Everything agent.
All related.

From the release notes

此版本重点提升会话可靠性、Windows 升级与 Agent 执行、文件路径处理和对话记录稳定性。 发布渠道：稳定版 · v1.38.10 English → · 网页版完整更新日志 → 使用攻略 应用会话所有权 — 了解会话在应用中的所有权和管理方式。 独立会话 — 了解独立会话的工作原理及其生命周期。 管理页面 — 探索会话和归档的管理页面。 对话记录架构 — 对话记录架构及其组件的详细信息。 Windows Agent Shell — 了解持久 PowerShell 如何在 Windows 上保留状态并报告命令完成情况。 Windows 安装程序验收 — Windows 安装程序构建的验收标准。 概览 Reasonix v1.38.10 — Reasonix 1.38.10 此版本重点提升会话可靠性、Windows 升级与 Agent 执行、文件路径处理和对话记录稳定性。 发布日期：2026-09-17 Desktop 36 · CLI 2 桌面端更新 重点内容 [Desktop] 更可靠的会话管理 — 会话现在在重启后保持身份和历史记录，归档和回收站处理得到改进。 (#10440, #10396, #10437) [Desktop] 更快的本地历史加载 — 切换标签页时，本地会话历史立即显示，无需等待运行时启动。 (#10414, #10423) [Desktop · CLI] Windows 路径和升级修复 — 修复了 Windows 文件 URI、CLI 诊断、旧会话升级和安装程序身份问题。 (#10459, #10451) [Desktop] 可展开会话与工作区上下文 — 侧栏现在每个项目或分组显示 5 个会话，可分批展开，新会话支持项目和分支上下文。 (#10426) [Desktop] 重启后保留压缩上下文 — 压缩后的模型上下文现在在切换会话、重启和导入旧会话后仍然保留。 (#10430) [Desktop] 修复用户消息历史分页 — 发送的消息在历史页面回收或确认超出可见窗口时不再重复或消失。 (#10438) 新增 [Desktop] 统一模型凭据恢复 — 桌面端与 CLI 现在共享凭据槽、回执和恢复服务，防止缺失凭据进入新对话轮次。 (#10460) [Desktop] 统一 Windows 桌面启动入口 — Reasonix.exe 现在是 Windows 的规范 GUI 入口，同时保留旧启动器的兼容性。 (#10439) [Desktop] 后台对话通知 — 后台对话现在无需切换标签页即可播放提示音并显示通知。 (#10382) 改进 [Desktop] 项目树展开与缓存逻辑 — 项目树现在使用五行初始窗口和五行增量，缓存失效与可见性分离。 (#10447) [Desktop] 默认按创建时间排序 — 此版本首次启动时，现有工作台排序偏好一次性重置为创建时间。 (#10444) [Desktop] 拦截 Windows 便携版共享目录启动 — 现在在 Electron 启动前拒绝从 UNC 和映射远程驱动器启动 Windows 便携版。 (#10442) [Desktop] 修复目录迁移的未读基线 — 迁移的历史记录不再因占位读取基线而显示蓝色未读点。 (#10428) [Desktop] 修复规范会话未读状态 — 未读状态现在只跟踪可见的终端助手结果，而不是元数据更改。 (#10424) [Desktop] 本地会话持久化操作与运行时解耦 — 会话操作现在通过显式 SessionRef 解析，改进了标题 CAS 和生命周期防护。 (#10409) [Desktop] 自动修复官方供应商协议与端点冲突 — 当 URL 唯一匹配目录路由时，现在会自动修复冲突的官方供应商协议/端点设置。 (#10411) [Desktop] 按正式构建身份限制桌面更新器 — 更新器现在在 canary、preview、test、dev 和 prerelease 构建中禁用，自动检查限制为每六小时一次。 (#10406) [Desktop] 统一会话归档、回收站与历史恢复 — 归档会话现在使用一个持久生命周期：正常会话 → 回收站 → 恢复或永久删除。 (#10396) [Desktop] 完善旧版对话迁移并排除重复恢复副本 — 桌面升级现在跨代发现并整合旧版历史，不重复恢复副本。 (#10395) [Desktop] 修复消息身份、工具状态与持久 PowerShell — 单次发送不再渲染两次，Windows Agent 执行使用带结构化完成的持久 PowerShell。 (#10392) [Desktop] 将命令面板搜索移至侧栏底部 — 命令面板搜索操作从主题栏移至侧栏底部的第一个插槽。 (#10390) [Desktop] 关闭最后一个标签页后自动收起工作区 — 关闭最后一个工作区标签页现在会收起整个右侧停靠栏，并将其空间返回给对话。 (#10388) [Desktop] 修复迁移后会话 AI 重命名失败 — AI 重命名现在从规范持久历史中读取用户消息，修复了迁移对话的重命名失败。 (#10389) [Desktop] 统一会话同步与完成机制 — 已完成、取消或重新连接的聊天不再丢失可见文本或显示过时的运行状态。 (#10385) [Desktop] 暂停与恢复改为追加式终结 — 停止一个轮次并打开另一个对话不再导致持久历史不可读。 (#10383) [Desktop] 恢复项目树并整合会话恢复修复 — 恢复了 ProjectTree 工作台侧栏，并整合了会话、历史、聊天显示和终端修复。 (#10380) [Desktop] 升级 Desktop SessionID 会话架构 — 桌面规范会话现在仅按 SessionID 持久化，防止切换或重启后消失。 (#10326) 修复 [Desktop] 保持迁移会话的身份并清除恢复后的错误提示 — 启动和继续读取历史时复用已迁移的会话身份，避免元数据刷新后重复导入。历史恢复成功后清除错误状态，不再残留错误的聊天警告。 (#10472) [Desktop] 稳定 Windows 发版验证 — 修复了 Windows 验证缺陷，包括在 Windows 上运行仅限 POSIX 的测试和 shell 探测缓存污染。 (#10468) [Desktop · CLI] 统一 PowerShell 冷启动握手预算 — PowerShell 现在从进程启动到连接接受和就绪帧接收有 30 秒预算。 (#10464) [Desktop] 修复重开会话的历史缓存复用 — 规范历史窗口现在在同一会话以新标签 ID 重新打开时复用热驻留。 (#10423) [Desktop] 串行化回收站恢复与永久删除 — 回收站删除现在在刷新、锁等待、重试和重启期间保留用户的原始确认快照。 (#10437) CLI 端更新 重点内容 [Desktop · CLI] Windows 路径和升级修复 — 修复了 Windows 文件 URI、CLI 诊断、旧会话升

From the release notes

此版本重点提升会话可靠性、Windows 升级与 Agent 执行、文件路径处理和对话记录稳定性。 发布渠道：稳定版 · v1.38.10 English → · 网页版完整更新日志 → 使用攻略 应用会话所有权 — 了解会话在应用中的所有权和管理方式。 独立会话 — 了解独立会话的工作原理及其生命周期。 管理页面 — 探索会话和归档的管理页面。 对话记录架构 — 对话记录架构及其组件的详细信息。 Windows Agent Shell — 了解持久 PowerShell 如何在 Windows 上保留状态并报告命令完成情况。 Windows 安装程序验收 — Windows 安装程序构建的验收标准。 概览 Reasonix v1.38.10 — Reasonix 1.38.10 此版本重点提升会话可靠性、Windows 升级与 Agent 执行、文件路径处理和对话记录稳定性。 发布日期：2026-09-17 Desktop 36 · CLI 2 桌面端更新 重点内容 [Desktop] 更可靠的会话管理 — 会话现在在重启后保持身份和历史记录，归档和回收站处理得到改进。 (#10440, #10396, #10437) [Desktop] 更快的本地历史加载 — 切换标签页时，本地会话历史立即显示，无需等待运行时启动。 (#10414, #10423) [Desktop · CLI] Windows 路径和升级修复 — 修复了 Windows 文件 URI、CLI 诊断、旧会话升级和安装程序身份问题。 (#10459, #10451) [Desktop] 可展开会话与工作区上下文 — 侧栏现在每个项目或分组显示 5 个会话，可分批展开，新会话支持项目和分支上下文。 (#10426) [Desktop] 重启后保留压缩上下文 — 压缩后的模型上下文现在在切换会话、重启和导入旧会话后仍然保留。 (#10430) [Desktop] 修复用户消息历史分页 — 发送的消息在历史页面回收或确认超出可见窗口时不再重复或消失。 (#10438) 新增 [Desktop] 统一模型凭据恢复 — 桌面端与 CLI 现在共享凭据槽、回执和恢复服务，防止缺失凭据进入新对话轮次。 (#10460) [Desktop] 统一 Windows 桌面启动入口 — Reasonix.exe 现在是 Windows 的规范 GUI 入口，同时保留旧启动器的兼容性。 (#10439) [Desktop] 后台对话通知 — 后台对话现在无需切换标签页即可播放提示音并显示通知。 (#10382) 改进 [Desktop] 项目树展开与缓存逻辑 — 项目树现在使用五行初始窗口和五行增量，缓存失效与可见性分离。 (#10447) [Desktop] 默认按创建时间排序 — 此版本首次启动时，现有工作台排序偏好一次性重置为创建时间。 (#10444) [Desktop] 拦截 Windows 便携版共享目录启动 — 现在在 Electron 启动前拒绝从 UNC 和映射远程驱动器启动 Windows 便携版。 (#10442) [Desktop] 修复目录迁移的未读基线 — 迁移的历史记录不再因占位读取基线而显示蓝色未读点。 (#10428) [Desktop] 修复规范会话未读状态 — 未读状态现在只跟踪可见的终端助手结果，而不是元数据更改。 (#10424) [Desktop] 本地会话持久化操作与运行时解耦 — 会话操作现在通过显式 SessionRef 解析，改进了标题 CAS 和生命周期防护。 (#10409) [Desktop] 自动修复官方供应商协议与端点冲突 — 当 URL 唯一匹配目录路由时，现在会自动修复冲突的官方供应商协议/端点设置。 (#10411) [Desktop] 按正式构建身份限制桌面更新器 — 更新器现在在 canary、preview、test、dev 和 prerelease 构建中禁用，自动检查限制为每六小时一次。 (#10406) [Desktop] 统一会话归档、回收站与历史恢复 — 归档会话现在使用一个持久生命周期：正常会话 → 回收站 → 恢复或永久删除。 (#10396) [Desktop] 完善旧版对话迁移并排除重复恢复副本 — 桌面升级现在跨代发现并整合旧版历史，不重复恢复副本。 (#10395) [Desktop] 修复消息身份、工具状态与持久 PowerShell — 单次发送不再渲染两次，Windows Agent 执行使用带结构化完成的持久 PowerShell。 (#10392) [Desktop] 将命令面板搜索移至侧栏底部 — 命令面板搜索操作从主题栏移至侧栏底部的第一个插槽。 (#10390) [Desktop] 关闭最后一个标签页后自动收起工作区 — 关闭最后一个工作区标签页现在会收起整个右侧停靠栏，并将其空间返回给对话。 (#10388) [Desktop] 修复迁移后会话 AI 重命名失败 — AI 重命名现在从规范持久历史中读取用户消息，修复了迁移对话的重命名失败。 (#10389) [Desktop] 统一会话同步与完成机制 — 已完成、取消或重新连接的聊天不再丢失可见文本或显示过时的运行状态。 (#10385) [Desktop] 暂停与恢复改为追加式终结 — 停止一个轮次并打开另一个对话不再导致持久历史不可读。 (#10383) [Desktop] 恢复项目树并整合会话恢复修复 — 恢复了 ProjectTree 工作台侧栏，并整合了会话、历史、聊天显示和终端修复。 (#10380) [Desktop] 升级 Desktop SessionID 会话架构 — 桌面规范会话现在仅按 SessionID 持久化，防止切换或重启后消失。 (#10326) 修复 [Desktop] 保持迁移会话的身份并清除恢复后的错误提示 — 启动和继续读取历史时复用已迁移的会话身份，避免元数据刷新后重复导入。历史恢复成功后清除错误状态，不再残留错误的聊天警告。 (#10472) [Desktop] 稳定 Windows 发版验证 — 修复了 Windows 验证缺陷，包括在 Windows 上运行仅限 POSIX 的测试和 shell 探测缓存污染。 (#10468) [Desktop · CLI] 统一 PowerShell 冷启动握手预算 — PowerShell 现在从进程启动到连接接受和就绪帧接收有 30 秒预算。 (#10464) [Desktop] 修复重开会话的历史缓存复用 — 规范历史窗口现在在同一会话以新标签 ID 重新打开时复用热驻留。 (#10423) [Desktop] 串行化回收站恢复与永久删除 — 回收站删除现在在刷新、锁等待、重试和重启期间保留用户的原始确认快照。 (#10437) CLI 端更新 重点内容 [Desktop · CLI] Windows 路径和升级修复 — 修复了 Windows 文件 URI、CLI 诊断、旧会话升

From the release notes

Enhance LLM_DEBUG with took and first token time, short embedding to …

From the release notes

Release v4.7.9

From the release notes

Mutable DEV CLI build for fc512f8d2cc0b4835bf2bedf9ef42b1d78b3b11c. Install: curl -fsSL https://kortix.com/install | KORTIX_CHANNEL=dev bash Defaults to https://dev-api.kortix.com.

From the release notes

New Choose which account a connector runs as, per call. A connection now carries the accounts you may run it as, and the account is chosen when the connector is called rather than pinned to the whole session. The connector catalog, the SDK and the CLI all report the available accounts and the pinned default, and connectors/describe shows accounts instead of only tool counts. An ambiguous connector call is refused, not guessed. When several accounts are reachable and none is named or pinned as default, the call is denied with account_required instead of silently picking one — no more mail sent from the wrong mailbox. Choosing a connector account is discoverable in the CLI rather than a dead end. Improved A connector you have your own accounts for reads as connected, not "needs setup", and the admin connector list now agrees with what each caller can actually reach. Connector accounts are per connection instead of a connector-level strategy, and connections the old strategy flag made unreachable are revoked. The sandbox daemon's internals were reorganised behind explicit service boundaries, with its HTTP controllers separated from its adapters. Fixed A failed identity check no longer signs you out. A failed getUser round trip used to end the browser session; it no longer does, and the new-connection screen offers Upgrade when you are at the project cap. Connector authorization: finalizing a connection that the whole project can use now requires the connections-manage capability, matching the gate the connect step already applied. Previously a role with connector-write but not connections-manage could complete a project-wide shared connection. Auto-created connections no longer claim to be the default, and the CLI's account column no longer truncates away the pinned-default marker. Stale connector banner copy and a dead connector detail shell removed; the account list renders for every direct provider. What's Changed fix(auth): a failed getUser round trip no longer signs the browser out; /new offers Upgrade at the project cap by @markokraemer in #7352 refactor(connectors): credentials are a call-time choice — one access rule, no session gate, no strategy flag by @markokraemer in #7326 refactor(kortixd): introduce internal harness service boundaries by @DimitrijeGlibic in #7220 test(api): wait for the second lifecycle drain instead of a 200 ms margin by @markokraemer in #7357 fix(connectors): account-aware modal header, dead ConnectorDetail shell removed, reconcile re-activates a revoked account by @markokraemer in #7360 fix(build): quick-queue snapshot uses a reverse scan, unbreaking API typecheck on main by @markokraemer in #7368 fix(security): connect/finalize gates a project-owned account on connections.manage by @markokraemer in #7366 chore(staging): bring staging to main (connector-creds, git-connection-logout, 48 commits) by @markokraemer in #7356 Release v0.13.23 — Per-call connector accounts, no more guessed account, an

From the release notes

docs(release): prepare v1.38.10 notes Summary: Generate a bilingual, product-focused draft from merged pull request metadata. Reuse the selected release-bound PR when one is available. Verification: Validate the catalog, citations, bilingual fields, and rendered GitHub release notes before committing. docs(release): retain reviewed v1.38.10 notes and bind final CI repair Preserve the reviewed bilingual product changes and citations. Only update six guide links to the latest main-v2 source after the test and CI lifecycle repair. Catalog validation and all 11 release-note tests pass. Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: SivanCola 32437197+SivanCola@users.noreply.github.com

From the release notes

Initial release release(typesafe): bump to 0.0.1a2 (#40576) feat(typesafe): experimental AutoModeMiddleware (#40545) feat(typesafe): experimental ModelRouterMiddleware (#40543) fix(typesafe): trace usage metadata (#40570) feat(typesafe): TypeSafeClassifier (#40542)

From the release notes

Release API image v2.11.359

From the release notes

What's Changed fix(core): align conditional approvals with validated tool arguments by @seratch in #5066 fix(core): deliver tool-not-found output on server-managed resume by @hyeonsang010716 in #4947 fix(sandbox): keep command paths POSIX on a Windows host by @hyeonsang010716 in #4958 fix(sessions): handle concurrent async SQLite startup by @00200200 in #4987 fix(sessions): do not forward session limit as Conversations page size by @tiagovilasboas in #4961 fix(tracing): do not cache a missing OPENAI_API_KEY by @hyeonsang010716 in #5017 fix(extensions): include shell and apply_patch calls in AdvancedSQLiteSession's tool usage by @rioyu123 in #4982 fix(extensions): reap subprocesses on early stream close by @Hughhhhcoder in #4804 fix(extensions): preserve exact IDs in AdvancedSQLiteSession.delete_branch by @JiangLLM in #4980 Documentation & Other Changes docs: align repository security and contributor guidance by @jbeckwith-oai in #4962 docs: clarify Unix-local execution boundaries by @seratch in #5064 docs: update translated pages by @seratch in #5065 docs(examples): drop the duplicated word in the Temporal trace name by @simpleqt in #5079 chore: enable Python Dependabot coverage with release-age cooldowns by @jbeckwith-oai in #4964 chore: add code ownership and required release review guidance by @jbeckwith-oai in #4965 fix(deps): update vulnerable dependencies without changing SDK APIs by @jbeckwith-oai in #4963 chore(deps): update fastapi requirement from >=0.120.0 to >=0.141.1 in /examples/realtime/twilio_sip by @dependabot[bot] in #4966 chore(deps): update uvicorn requirement from >=0.38.0 to >=0.52.4 in /examples/realtime/twilio_sip by @dependabot[bot] in #4968 chore(deps): update openai requirement from <4,>=3.0.0 to >=3.8.0,<4 in /examples/realtime/twilio_sip by @dependabot[bot] in #4969 chore(deps): bump typing-extensions from 4.14.1 to 4.16.0 by @dependabot[bot] in #4970 chore(deps-dev): bump grpcio from 1.76.0 to 1.83.1 by @dependabot[bot] in #4971 chore(deps): bump sqlalchemy from 2.0.43 to 2.0.52 by @dependabot[bot] in #4973 chore(deps-dev): bump cryptography from 45.0.7 to 50.0.0 by @dependabot[bot] in #4975 ci: add Python and Actions CodeQL security scanning by @jbeckwith-oai in #4974 fix(ci): deploy docs after dependency and workflow updates by @dependabot[bot] in #4976 fix(deps): update python-multipart requirement from >=0.0.31 to >=0.0.32 in /examples/realtime/twilio by @dependabot[bot] in #5030 fix(deps): update python-dotenv requirement from >=1.2.2 to >=1.2.3 in /examples/realtime/twilio by @dependabot[bot] in #5031 fix(deps): update starlette requirement from >=1.3.1 to >=1.6.0 in /examples/realtime/twilio_sip by @dependabot[bot] in #5033 fix(deps): update python-dotenv requirement from >=1.2.2 to >=1.2.3 in /examples/realtime/twilio_sip by @dependabot[bot] in #5034 fix(deps-dev): bump pymdown-extensions from 11.0.1 to 11.0.2 by @dependabot[bot] in #5036 fix(deps): bump runloop-api-client from

From the release notes

Release API image v2.11.358

From the release notes

Release API image v2.11.357

From the release notes

What changed Fixes Finding B of #3327: memory_search({ smart: true }) was reporting SmartRetrieval's internal RRF fusion score in the similarity field instead of the underlying retrieval relevance score — e.g. a raw retrieval relevance of 0.8 was showing up as 0.03252247488101534. similarity now carries the raw retrieval relevance (matches smart: false behavior) the fused SmartRetrieval ranking score is now exposed separately as rankingScore threshold behavior is unchanged (still applied pre-RRF/MMR/recency) 14 new regression tests across the SmartRetrieval package and MCP handler Fixed in #3340 (thanks @takagibit18). Finding A of #3327 (reasoningBank/vectorBackend reporting enabled:true but routing through bridge-fallback) remains open and unclaimed. Note on version history 3.42.1–3.42.3 were published to npm without matching version-bump commits on main. Verified their tags (v3.42.0, v3.42.1, v3.42.3) are all ancestors of this release, so 3.42.4 is a strict superset of everything previously shipped. Co-Authored-By: RuFlo ruv@ruv.net

From the release notes

MiniMax H3 on Monid @hypit/provider-monid now serves @hypit/minimax-h3@1#minimax-h3 through Monid's minimax /v1/video/minimax-h3, alongside the four ByteDance Seedance endpoints it already mapped. The model has a third bundled gateway Provider next to HypiHub, HiAPI, Pollo and TokenDance. (#286) H3 takes the same role-tagged ModelArk content array as the Seedance endpoints, names its model in the body, and carries neither the generate_audio nor the web_search field they take. Three differences are handled at the route: resolution is required by the endpoint while the model's port is optional, so a request stating none is sent at 2K, the resolution the HypiHub Provider also selects. The endpoint takes no adaptive ratio for text-to-video, while frame mode resolves the framing from the uploaded image and reference-to-video defaults to adaptive. A text-to-video request carrying no aspect-ratio is reported unsupported before any reference is resolved; the other two modes are sent as adaptive. H3 returns its video at content.url rather than the ModelArk video_url. Monid addresses an endpoint by provider and path, so each mapping now carries the provider that relays it instead of every run assuming bytedance. The rest of Monid's catalogue carries no other model this Distribution describes. Its ElevenLabs endpoints are voice-id based while @hypit/elevenlabs-speech@1#eleven_ttv_v3 is voice design and requires a voice description they have no field for, and the H3 Fast, Max and Max Turbo variants declare different models with their own value domains. Merged pull requests PR Author #286 feat(provider-monid): serve MiniMax H3 @rponeawa Contributors @rponeawa Update For a global CLI installation: npm install -g @hypit/hypit@0.2.5 hypit --version For a project-local installation: npm install --save-exact @hypit/hypit@0.2.5 npm exec --no -- hypit --version The installed Hypit Skill updates separately through the tool used to install it. Internal format and logical module identifiers remain @1. Script syntax is unchanged from 0.2.0. Updating does not migrate or rewrite saved video projects.

From the release notes

Release API image v2.11.356

From the release notes

Partner gateway Providers and media frame rate This release adds four bundled Providers for third-party gateway services and fixes a frame rate defect in local media execution that affects ffmpeg 7.x. Partner gateway Providers The Distribution now carries four Providers that serve the installed models each service offers. Each takes an API key credential and reads no pricing. Every one maps only the models the Distribution already describes, reports each service's input limits through supports() before any reference is resolved, and accepts Seedance person-reference declarations without transmitting them, since none of these APIs carries such a field. (#282) @hypit/provider-tokendance, credential tokendance.api-key. Seedance 2.0, 2.0 Fast, 2.0 Mini and 2.5, Seedream 5 Lite and MiniMax H3, through the Ark and MiniMax protocols TokenDance documents. MiniMax inputs are uploaded through TokenDance's file API. @hypit/provider-hiapi, credential hiapi.api-key. Seedance 2.0, 2.0 Fast, 2.0 Mini and 2.5, Seedream 5 Lite, MiniMax H3, GPT Image 2, Nano Banana 2 and Pro, and Grok Imagine video and 1.5 Preview, each submitted as one task to HiAPI's POST /v1/tasks with an Idempotency-Key, so a retried submission returns the original task. @hypit/provider-pollo, credential pollo.api-key. MiniMax H3, Grok Imagine 1.5 Preview, GPT Image 2, and Nano Banana 2 and Pro, through Pollo AI's per-model generation paths. Pollo offers no Seedance or Seedream model. @hypit/provider-monid, credential monid.api-key. Seedance 2.0, 2.0 Fast, 2.0 Mini and 2.5 through Monid's bytedance endpoints, with reference media uploaded through the Monid workspace file system. Models and deployment services describes each service and what its Provider covers. The Providers guide and the generation quickstart carry the same list, and the installed Skill points at the service page once a user names one of these services. Local media execution normalize and transform compiled an exact frame domain and asserted the encoded result against it, but did not tell ffmpeg what that frame rate was. Both filter chains end in setpts, which marks the filtergraph output link's frame rate as unknown, and -frames:v bounds the frame count without establishing a rate. On ffmpeg 7.x the encoder then fell back to 25 fps. Requesting a rate above 25 fps produced a frame count that no longer matched the plan and failed the operation's own assert, so a 30 fps project could not be normalized. Requesting a rate below 25 fps filled the count and wrote the container at 25 fps with the wrong duration, with nothing reporting it. Both operations now pass the compiled rate, as render-still-video already did. (#284, reported in #283) ffmpeg 6.1.6, 8.1.2 and 9.0.1 derive the rate correctly and were never affected. On those versions the encoded output is byte identical before and after this change, across both encoder branches and both input branches. The repository's own CI installs ffmpeg 6.1.1 on Linux and 9.0.1 o

From the release notes

Release API image v2.11.355

From the release notes

Release API image v2.11.354

From the release notes

Release API image v2.11.353

From the release notes

Release API image v2.11.352

From the release notes

New Private provider key pools. Bring several keys for one provider, keep them private or share them with named members, and let each session hold its own key. Sessions fail over to the next key and report the earliest retry time when a pool is exhausted. Gemini and ChatGPT keys pool too, and pooled keys are managed in Models. Pool access is bound to the session owner, and a pool is checked against the model you actually picked. Print a whole conversation. Cmd+P prints the full session as a clean document — every message, not one clipped screen of the app. Queued prompts you can see and edit. Queued messages show in a list above the composer. Press Up to edit the last one, and Resume picks the queue back up. Microsoft Teams install reports what happened. One-click install publishes in the background, lands on Channels with a real status, and offers a retry with the actual reason when publishing fails. Astra is available as a ChatGPT subscription model. Improved Faster project boot. New sessions fetch the project from storage instead of cloning it, and start sooner. Faster session lists. A project's session list is now a page, not the whole inventory. Menus open instantly — menus, popovers, tooltips and the command palette no longer wait to appear. Smaller sandbox images, which start faster. Git connections are organized per account, with one instance backend and an identity that cannot drift. Verifying with GitHub no longer signs you out, and the new-connection screen always has a way out. Fixed Microsoft Entra directory sync. Patches apply atomically, large directories paginate, groups survive while a user is inactive or a sign-in session is stale, deactivation and single-member removal behave correctly, and users resolve across the whole directory. The gateway retries a slow internal call instead of answering "Gateway unavailable". Composer drafts survive session startup, and opening session settings no longer closes itself while the composer takes focus. One session lifecycle drain at a time per API task, so queued work is not claimed twice. Session attachments are more reliable, and model prices show correctly in the ChatGPT picker. Security hardening across session cursors, webhook signing keys, and the remaining high-severity scanner findings. Database migrations apply cleanly on staging and production. What's Changed fix(teams): one-click install publishes in the background, records its outcome, and shows it by @Ino-Bagaric in #7341 fix(api): run one session lifecycle drain at a time per API task by @sutharjay1 in #7343 fix(api): run one session lifecycle drain at a time per API task (staging) by @sutharjay1 in #7344 fix(git): verify-with-GitHub no longer signs out; /new is a git account manager with a way out by @markokraemer in #7330 chore(release): staging VERSION → 0.13.22 [skip ci] by @github-actions[bot] in #7347 chore(release): VERSION → 0.13.22 [skip ci] by @github-actions[bot] in #7348 Default pooled pro

From the release notes

Initial release feat(typesafe): TypeSafeClassifier (#40542)

From the release notes

Rolling, fully SYNTHETIC preview-database seed fixture baked from dev@ef2d42c798f4 (migration head: 20260917160000_add_expert_job_title). Preview environments restore fixture.dump.gz (custom-format, gzipped, platform schema only, including _prisma_migrations) BEFORE running a PR's own migrations, so schema changes run against populated tables. RESTORE: apply restore-preamble.sql to the target database first (it creates the platform schema and installs the vector + pg_trgm extensions into it, which pg_dump --schema=platform omits), then run 'gunzip -c fixture.dump.gz | pg_restore -d --no-owner' WITHOUT --exit-on-error (the dump's own CREATE SCHEMA collides benignly with the preamble's). SECURITY INVARIANT: this bake holds no cloud credentials and has no read path to any real database — it runs entirely against a throwaway Postgres container and contains only Faker-generated synthetic rows plus the already-public marketplace agent exports checked into autogpt_platform/backend/agents/. This asset is regenerated on every eligible dev push and weekly; treat it as disposable.

From the release notes

New Print a whole conversation. Cmd+P prints the full session as a clean document — every message, not one clipped screen of the app. Queued prompts you can see and edit. Queued messages show in a list above the composer. Press Up to edit the last one, and Resume picks the queue back up reliably. Private provider connection pools. Bring several keys for one provider, keep them private or share them with named members, and let a session hold its own key. Sessions fail over to the next key and report the earliest retry time when a pool is exhausted. Gemini and ChatGPT keys pool too, and pooled keys are managed in Models. Astra is available as a ChatGPT subscription model. Improved Faster project boot. New sessions fetch the project from storage instead of cloning it, and start sooner. Faster session lists. A project's session list is now a page, not the whole inventory. Menus open instantly. Menus, popovers, tooltips and the command palette no longer wait to appear. Smaller sandbox images, which start faster. Git connections are now organized per account, with one instance backend and an identity that cannot drift. Fixed Microsoft Entra directory sync. Patches apply atomically, large directories paginate, groups survive while a user is inactive or a sign-in session is stale, deactivation and single-member removal behave correctly, and users resolve across the whole directory. The gateway retries a slow internal call instead of answering "Gateway unavailable". Composer drafts survive session startup, and opening session settings no longer closes itself while the composer takes focus. Session attachments are more reliable. Model prices show correctly in the ChatGPT picker. Security hardening across session cursors, webhook signing keys, and the remaining high-severity scanner findings. Database migrations now apply cleanly on staging and production. What's Changed feat(sandbox): bake rg, fd, bat, jq, fzf and a shell tool floor into every image by @markokraemer in #7264 refactor(sandbox): remove the fast cold boot image, its flag, and the experiments it gated by @markokraemer in #7268 test(release): dismiss the welcome card before Save in the repository-access journey by @markokraemer in #7276 fix(cli): connectors apps searches the Composio toolkit catalog by @markokraemer in #7277 feat(project-snapshot): S3 boot acquires the project 3.5× faster than Git (307 vs 1,078 ms) and boots 0.8 s sooner — presign at create, no HEAD on the boot path, kortixd gated on OpenCode's "server listening" by @DimitrijeGlibic in #7242 fix: restore terminal recovery and keep message retries in their session by @sutharjay1 in #7267 fix(usage): show zero cost for ChatGPT subscriptions by @markokraemer in #7278 chore(release): VERSION → 0.13.19 [skip ci] by @github-actions[bot] in #7285 fix: refresh terminal provider credentials after sandbox resume by @markokraemer in #7286 fix(session): isolate optional connector failures and preserve prompt qu

From the release notes

Release API image v2.11.351

From the release notes

cut the revison v2.1.0-rc0

From the release notes

Changelog Features bb76eb6 feat: add magic link sign-in Bug fixes fbcb496 fix: keep empty forceLanguage empty so browser language detection works 0db629e fix: keep resolved RADIUS/push provider id through MFA setup enable step 4684d70 fix: read SAMLRequest case-insensitively in auth callback 91ae9d6 fix: redirect to the single OAuth provider only once instead of on every render 7bf924a fix: reset signin method rule when its name changes so WebAuthn stays visible

From the release notes

What's Changed docs: update toolkits, API spec, and meta tools data by @sdkrelease[bot] in #4419 docs(auth): note the auto-populated connection display_name by Anshu Garg (@anshugarg15) in #4503 docs: update toolkits, API spec, and meta tools data by @sdkrelease[bot] in #4510 fix(cli): block SSRF in proxy binary downloads by Alberto Schiabel (@jkomyno) in #4511 Full Changelog: https://github.com/ComposioHQ/composio/compare/@composio/cli@0.4.2-beta.397...@composio/cli@0.4.2-beta.398

From the release notes

Release nightly created from 1ff3d96 at 2026-09-17 21:18:32+08:00

From the release notes

Local execution and request correctness File-backed credentials now share the Result store's native Windows replacement implementation. CLI and Worker processes can read and replace complete credential values without a process-local queue. Failed replacement preserves the previous value; unsupported filesystem operations report their errors. Windows OS credential replacement no longer deletes the previous value before writing. Child process timeouts and output failures terminate the helper. Explicit file/platform credential directories outside the Host state root are accepted again. Profiles retain control of their selected storage location. Generation mappings reject explicitly supplied media fields that cannot be transmitted, before uploading or submitting a request. Optional booleans preserve both true and false. Media preparation probes the selected FFmpeg/FFprobe executables in their execution environment. A probe-only Program reports installation/preparation accurately without claiming to be a running service. The launcher resolves its physical Distribution directory before loading packages. This fixes Windows short-path aliases incorrectly making installed fonts appear outside their own Distribution. Installed-package CI now verifies both Windows and Linux, including component build, fonts, explicit browser preparation, rendering, export and full media decode. The project video Provider example carries first/last-frame person-reference classifications through its illustrative upload transport. These changes do not add silent downloads, retries, fallback browsers, lock-file protocols, central queues or persistent coordination metadata. Logical module and format identifiers remain @1; Script syntax is unchanged. Updating For a global CLI installation: npm install -g @hypit/hypit@0.2.3 hypit --version For a project-local installation: npm install --save-exact @hypit/hypit@0.2.3 npm exec --no -- hypit --version The installed Hypit Skill updates separately through the tool used to install it. Update that installation from this repository to receive the revised Runtime profile guidance, which separates preparation, process startup and credential connection. Updating either installation does not migrate or rewrite saved video projects. Repository-only change: the optional AI analysis workflow is manually triggered, reads one issue or PR and returns advice in the Actions summary. It has no issue/PR write permissions and is independent of the npm CLI.

From the release notes

Release API image v2.11.350

From the release notes

Windows media environment and credential store correctness This release includes Windows execution fixes and a change to credential directory selection. Existing Profiles with a file or platform Store path outside the Host state root are rejected by 0.2.2; this is a behavior change from 0.2.1. The new starter Store selection applies only to newly initialized Profiles. Windows media execution ffmpeg and ffprobe are started with a whitelist of environment variables so the Host environment stays out of media children. That whitelist was written for POSIX, so Windows children received neither %TEMP% nor %TMP% and failed to create their temporary files. Both the HyperFrames render path and the shared media execution path now carry a Windows list alongside the POSIX one. Host secrets are still withheld. (#259, #260) Credential storage runtime init writes a starter Profile that selects @hypit/credential-store-platform, so the Profile it produces opens and writes on Linux as well as macOS and Windows. macOS and Windows keep the platform locker under the same service and key, so an existing credential is still found. os, file and env remain selectable by name. (#247) A Profile's config.path for the file and platform stores is resolved against the Host state root and checked for containment. A path that escapes that root is rejected when the Store opens. An absolute path outside the root was previously accepted and now is not. (#265) Reads and replacements of one credential document take their turn within a process. resolve holds the document open for the whole read while put replaces it with a rename, and Node’s Windows rename implementation uses MoveFileExW, which can refuse replacement while the destination is open. The process-local ordering reduces this overlap within one process; it does not coordinate separate CLI and Worker processes. Cross-process replacement failures remain in 0.2.2. Keys stay independent. (#276) The Windows Credential Manager helper is bounded to the same 10 second timeout as the macOS security(1) call, so a stuck locker fails instead of hanging the CLI. (#261) Runtime and CLI programs down treats a Program with no valid Hypit PID and no start command as having nothing to stop, and reports No external programs to stop instead of exiting 1. The starter Profile's media.local and hyperframes.local declare only probe(), so a ready probe means the tool is installed, not that a process is running. A startable Program that is ready without a Hypit PID still reports not-ours. (#272) The overview help lists programs prepare|up|status|down, and programs and runtime without a selected Profile report the same recovery sentence that build, pricing, auth and the execution commands already use. (#266) Three packages declare the workspace packages their sources import. (#253) Documentation The English quickstart's Build workflow section carries the dependency preparation, .gitignore, shared asset library and archive command materi

From the release notes

Changelog 726a263 feat: refresh Doubao models to the current Ark lineup

From the release notes

What's new 🌐 Introducing Rowboat Spaces. Every teammate. Their own agent. One Space. Rowboat becomes a multiplayer personal assistant for work. A Space brings your team’s conversations, files, and whiteboards together, and everyone brings their own Rowboat: their own memory, tools, and model keys. Your personal assistant can now work alongside your teammates and their assistants in the same room. 🤖 Ask in the Space. Your Rowboat takes it from there. Type @rowboat what is missing? and your Rowboat picks it up on your machine, draws on your Brain, email, meetings, and notes, and brings the answer back as “You (via Rowboat)”. Your personal knowledge stays on your machine; the Space holds what you and your agent choose to share. The room gets a 👀 while work is underway and a ✅ when it settles. If your Rowboat needs a decision, permission, or a private conversation, it leaves a ❗ and asks in your own chat. You can also open the agent conversation behind a Rowboat reply to follow the work. 🎨 From a shared whiteboard to something you can ship. Sketch together with live cursors, then ask @rowboat turn this board into a proposal. Agents can draw on the same board and work with the same files as the team. Markdown documents have version history, diffs, and restore; concurrent text edits merge line by line, with conflicts surfaced for resolution. Open documents and presentations in place, or link a discussion to a file and read them side by side. When the proposal is ready, @rowboat implement the proposal can hand it to Claude Code or Codex on your machine, using the Space’s files as context, and bring the result back for review. 💬 The everyday parts of teamwork are here too. Spaces includes threaded conversations, DMs, notes to yourself, reactions, polls, scheduled messages, and mentions. The composer has formatting controls, link previews, attachments, and voice dictation. An Activity view gathers mentions, DMs, replies in followed threads, and reactions on your messages. Unread state follows your account across devices, and notifications take you back to the conversation. On macOS, the Dock icon shows a count for mentions and DMs, or a dot for other unread activity. 🚪 Start a Space and bring your team. Spaces is now available by default. Open Spaces → Create a server, sign in with Google or Microsoft, and use Copy invite link to bring teammates in. A server is your team’s home for its Spaces; you can create or join more than one. For teams that want to run their own infrastructure, Harbor, the server behind Spaces, is open source and self-hostable. It also exposes an MCP server, so other agents can participate using the same tools as Rowboat. See the Spaces guide or explore Spaces on our website. 🪟 Keep your assistant wherever the work is. Move a conversation between the full Assistant page, a right sidebar, and a floating window. Keep several floating chats open, resize and reorder them, or minimize them into tabs along the bottom. The app now ope

From the release notes

本版修复远端主机使用本机模型时报 unknown model，并移除 IM 机器人网关。 修复 远端主机使用本机模型时，新会话、模型列表和切换模型不再落到远端自己的配置，不再报 unknown model #10448 所选模型在本机不存在时，报错会说明缺模型的是本机，并提示如何改用远端自己的模型 (8b2a05c) 移除 reasonix bot 与 QQ、飞书、微信机器人网关；配置文件里已有的 [bot] 段原样保留，不影响 1.x (f352dfb) 升级须知 模型来源为本机的远端主机，若内核低于 2.18.1，下次连接会自动下载并替换远端内核 (437b982) macOS 已用 Developer ID 签名并公证，打开时不需要清除隔离属性。Windows 包未签名，SmartScreen 会告警：选「更多信息」，再选「仍要运行」。Linux 从 .deb 安装。 三个平台都能自更新：Linux 走自己的包，Windows 运行下一个安装器，macOS 替换自身 bundle。

From the release notes

本版修复 Windows 窗口打不开、读图模型误报看不到图、问题框不弹出，并精简内置文档。 修复 Windows 安装目录带 AppContainer 包授权时窗口打不开；启动时自动移除这类授权 #10435 读图模型不再声称看不到已附加的图片 (44150b0) Goal 续跑时不再提示读图模型"无法读图" (44150b0) 切换模型、重载扩展或切换工作区后，问题框和审批框不再被吞掉 (921a04b) 变更 内置文档除 README、使用指南、CLI 参考外只保留英文版，中文提问会检索到英文文档 (68b031a) 移除 内置文档检索中的版本历史，其内容停留在旧产品线的 1.24.1 (99c5360) macOS 已用 Developer ID 签名并公证，打开时不需要清除隔离属性。Windows 包未签名，SmartScreen 会告警：选「更多信息」，再选「仍要运行」。Linux 从 .deb 安装。 三个平台都能自更新：Linux 走自己的包，Windows 运行下一个安装器，macOS 替换自身 bundle。

From the release notes

Signed-off-by: omnigent-ci[bot] <294685417+omnigent-ci[bot]@users.noreply.github.com>

From the release notes

🛡️ Security This release fixes four security issues, all of them reached through web_fetch_tool or OpenTelemetry instrumentation. See each advisory for full details and affected versions. GHSA-vmxc-h2x2-jmf3 (moderate): the cloud-metadata and private-IP blocklists could be bypassed with an IPv6 zone identifier on a URL opted into local network access, via FileUrl(force_download='allow-local') or web_fetch_tool(allow_local_urls=True). Both are off by default. Reported by @euriconicacio. (#8401) GHSA-fpf4-vwcp-v4hp (moderate): web_fetch processed responses in superlinear time on the event loop, in both the HTML conversion and the charset decode, so a single attacker-chosen page could stall every agent in the process. Reported by @BrianWillows. (#8397, #8418, #8433) GHSA-22h6-qm39-v87j (low): web_fetch_tool's domain lists were compared as written rather than in the form the resolver uses, so a blocked domain could be reached under another spelling. (#8407, #8421) GHSA-4x9p-g9wm-8q7f (low): with InstrumentationSettings(include_content=False), spans still carried exceptions, error statuses, instructions and the output template. Reported by @BrianWillows. (#8403, #8408, #8419, #8428) Patched in 2.44.0 (v2) and 1.107.6 (v1). What's Changed ⚠️ Compatibility Notes Make RunContext.enqueue() safe from worker threads by @adtyavrdhn in #7758 Require a JSON Content-Type on UI adapter requests by @DouweM in #8396 Dispatch a capability @durable_operation called from a per-request hook, instead of silently running it inline by @DouweM in #8395 🚀 Features Add a Vercel AI SDK and Eve migration skill by @adtyavrdhn in #8357 Give AgentRunResult a stable serialized shape, and settle a finished stream into one by @DouweM in #8269 Add a Storage page so persistence is findable by @DouweM in #8336 Emit EnqueuedMessagesEvent when a realtime session delivers enqueued content by @DouweM in #8134 🐛 Bug Fixes Forward blocked_domains to OpenAI web search by @adtyavrdhn in #8323 Fix OpenAI hosted tool_search pairing in AG-UI by @adtyavrdhn in #8313 Let Agent.from_spec() construct without a model, deferring the requirement to run time by @DouweM in #8339 Let RealtimeSession.close() finish its teardown when the closing task is cancelled by @DouweM in #8142 Run a capability's prepare_tools over every tool the availability gate admits by @DouweM in #8071 Bound the realtime session event queue while nothing is iterating it by @DouweM in #8133 Report a run's own usage on its span, not the conversation total by @DouweM in #8417 New Contributors @mrchatam made their first contribution in #8286 @vedjaw made their first contribution in #7028 Full Changelog: v2.43.0...v2.44.0

From the release notes

🛡️ Security A maintenance release for the v1 line, carrying the v1 backports of the four security fixes released in 2.44.0. See each advisory for full details and affected versions. GHSA-vmxc-h2x2-jmf3 (moderate): cloud-metadata and private-IP blocklist bypass via an IPv6 zone identifier, on a URL opted into local network access. Reported by @euriconicacio. (#8402) GHSA-fpf4-vwcp-v4hp (moderate): superlinear response processing in web_fetch, in both the HTML conversion and the charset decode, blocking the event loop. Reported by @BrianWillows. (#8399, #8434) GHSA-22h6-qm39-v87j (low): web_fetch_tool domain lists compared as written rather than in the form the resolver uses. (#8409) GHSA-4x9p-g9wm-8q7f (low): OpenTelemetry spans carrying exceptions, error statuses, instructions and the output template under include_content=False. Reported by @BrianWillows. (#8404, #8422, #8429) Also carried over: credentials are now dropped on a safe_download redirect whenever the full origin changes, not only the hostname (#8410). This was fixed on the v2 line some time ago and had never been backported. Patched in 1.107.6; all four are also patched on the v2 line in 2.44.0. What's Changed Every code change in this release is one of the security fixes above. Maintenance alongside them: Repair v1 CI and its mcp, anthropic, and duckduckgo extras (v1 backport) by @DouweM in #8406 Pin FastA2A below 1 in the a2a extra (v1) by @DouweM in #8412 Full Changelog: v1.107.5...v1.107.6

From the release notes

fix(api): align Go proxy contract parity (#19723)

From the release notes

Cross-platform credential storage Example Runtime Profiles now explicitly select @hypit/credential-store-platform, so the same Profile supports authentication on Linux, macOS and Windows. The package delegates to existing stores with a documented policy: macOS: Keychain through credential-store-os. Windows: Credential Locker through credential-store-os. Linux: unencrypted, owner-private local files through credential-store-file. Other platforms report an unsupported-platform error. A missing credential or backend failure never switches storage. No credentials are migrated. You can still select os, file or env directly; the CLI starter continues to select file. The policy lives in the new package, with no Core or protocol changes. This release also restores the previous English and Chinese README wording. Update npm install -g @hypit/hypit@0.2.1 hypit --version For a project-local installation: npm install --save-exact @hypit/hypit@0.2.1 npm exec --no -- hypit --version Internal format and logical module identifiers remain @1. Script syntax is unchanged from 0.2.0. See #246 for the implementation and validation.

From the release notes

What's Changed feat: backport app sources, artifact scroll and Windows MCP fixes by @bolichen97 in #11339 fix: refresh cached highlight worker CSP on insider by @bolichen97 in #11374 Full Changelog: v0.7.0-insider.3...v0.7.0-insider.4

From the release notes

What's changed Hypit 0.2.0 preserves authored text across semantic annotations and separates local resource preparation from execution. Script keeps authored display spacing. Chinese, Korean and mixed-script text retain their written separators. Speech tokens, display words and explicit Dual Text alignment groups remain distinct; spaces do not become timing tokens. Comments and annotations do not split complete words or detach their punctuation. Semantic markers have explicit delimiters. Selection and Moment markers use @{...}. Left/right affinity remains available, and Studio editing preserves the surrounding prose. Package documentation, examples and the Hypit Skill use the new syntax. Local resources are prepared explicitly. WhisperX prepares the selected ASR model, sentence data and requested alignment models before offline inference. Browser capture and video download also report missing preparation instead of installing during use. Resource choices remain with their owning packages and selected profiles. WhisperX accepts authored language codes. SVML language="ko" and CLI transcription can select Korean and other languages supported by the chosen endpoint. Model availability belongs to that endpoint, without a central cross-provider language whitelist. Joined caption backgrounds follow actual wrapping. Fine measures the complete Cue through the existing browser-program extension, including inactive words. Internally wrapped words receive full background coverage, translucent overlaps paint once, and borders follow the joined exterior. Existing semantic scheduling still determines activation. Portable diagnostics and tests. Credential-store guidance names explicit supported alternatives. Concurrency and process-exit tests cover independent writers and process reaping, and CLI tests use platform-resolved paths. Updating to 0.2.0 This release changes Script marker syntax. Update the executable and Skill through their respective installation channels, then update the Sources you intend to use with 0.2.0. For a global npm installation: npm install -g @hypit/hypit@0.2.0 hypit --version hypit version --check For a project-local installation, update that project's dependency and lockfile deliberately, for example: npm install --save-exact @hypit/hypit@0.2.0 npm exec --no -- hypit --version An existing ^0.1.x dependency range does not select 0.2.0 through an ordinary npm update. The executable and the installed Skill are separate. For a global Hypit Skill installed with the skills CLI, update only that Skill with npx skills update hypit -g; other installers use their own update method. Have the Agent read the updated Skill before continuing. Updating either installation does not silently rewrite a video project. Script markers and affinity Purpose Previous word's end Next word's start Selection start @{~name} @{name} Selection end @{/name} @{/name~} Moment @{~name!} @{name!} For example, @part text @/part becomes @{part} text @{

From the release notes

What's Changed Features Support aliases as connection identifiers Record reasons for deployment creation failures Collect human feedback and pause events in tracing Add llm_overlay context variable to route agent roles to models Carry task_prompt and output in agent execution payloads Validate platform integrations during crew setup Add platform tools to JSON crew wizard Expose CrewAI Platform application catalog Add OpenRouter as a supported embedding provider Bug Fixes Accept CRLF in inline skill definitions Load text file URLs through the safe fetcher Close SQLite connections in kickoff task outputs storage Carry from_cache on the native tool path's ToolUsageFinishedEvent Raise ValueError instead of bare raise in uploader Support non-primitive types in SQLiteFlowPersistence Improve coding agents instructions Read JSON checkpoints as UTF-8 Overwrite stale poetry.lock backup on Windows Key streamed tool calls by wire index in Azure Preserve file data content parts in Gemini Honor read_only on update() and recall() access times Send reasoning_effort to every OpenAI reasoning model Prevent crashes in the run TUI when streamed output contains a literal [...] Reject replay when stored tasks differ Use sys.platform guards for mypy compatibility on Windows Request the forced final answer as a user turn Keep null in the task output schema embedded in the prompt Align DOCXSearchTool with standard RAG fixed schema pattern Documentation Add xpu to embedding device options List all workspace packages Contributors @ASTion24, @BlueX888, @HUAN2022A, @RaycarlLei, @Rohitkanithi, @SWAPI03, @SharoonSharif, @ShivangiRay, @Theater-ahyeon, @Vidit-Ostwal, @gamal1osama, @github-actions[bot], @joaomdmoura, @lorenzejay, @mhbuehler, @modusensus, @monkscode, @Rohitkanithi, @roli-lpci, @vinibrsl, @wangtaotaotao95

From the release notes

What's Changed fix(cli): curated help for every command family and a help command by Alberto Schiabel (@jkomyno) in #4421 Full Changelog: https://github.com/ComposioHQ/composio/compare/@composio/cli@0.4.2-beta.396...@composio/cli@0.4.2-beta.397

From the release notes

What's Changed feat(docs): redesign social preview cards by Malay Vasa (@Malayvasa) in #4502 feat(cli): plugin-adoption telemetry for setup and the plugin hint by Kshitij Jhunjhunwala (@KJ-11) in #4496 Full Changelog: https://github.com/ComposioHQ/composio/compare/@composio/cli@0.4.2-beta.395...@composio/cli@0.4.2-beta.396

From the release notes

What's changed Local rendering prepares, inspects and launches the same explicitly selected browser. Profiles can select an exact browser version, an existing executable or one archive mirror. Failed downloads do not switch sources, and npm dependencies no longer download an unselected Puppeteer browser. Workers and capture processes resolve the installation actually being executed. Rendering preserves element animation, waits for visual resources, improves process cleanup and samples Studio previews on source frames. Windows media execution retains the required process environment, OAuth preserves the complete authorization URL, and yt-dlp locates its Python service through its declared package. Credential replacement and deletion no longer require reading the previous value. The new explicitly selected file CredentialStore supports writable local credentials on Linux, without changing existing OS-store selections. Package resolution follows npm optional-dependency precedence; font resource errors retain their actual cause and installation guidance uses the owning package's declared version. Windows workspace tests, Telegram links and bilingual testing guidance are corrected. Updating Install the release with npm install -g @hypit/hypit@0.1.14. Prepare a selected local renderer explicitly with hypit programs up --runtime <profile> --endpoint <renderer> or hypit runtime up. doctor inspects without installing. Browser selection belongs to the Profile; use its chromePath or browserVersion setting instead of old browser-path environment hints. Mirror and cache settings are documented in the local renderer README. For Linux writable credentials, explicitly select the file CredentialStore in the Profile before login. File storage is unencrypted and requires a private directory; existing OS credentials are not automatically migrated. Merged changes: #241 and #242. The publication workflow runs Linux/Windows checks and installs, renders, exports and decodes the packaged CLI before publishing that same tarball to npm.

From the release notes

Fixed Projects with thousands of sessions no longer slow the app down. The session list now loads one page at a time — 50 sessions, with a Load more control — instead of returning every session on every refresh. On a project with 12,617 sessions that is a 46 KB response instead of about 11 MB, and the sidebar was refreshing it every five seconds. Opening, sharing, stopping, restarting or renaming a session now works however old the session is. These controls used to look the session up in the full list, so once a project grew past the first page the controls could quietly disappear. Improved Date grouping (Today, This week, Older), filters and search work as before, applied to the sessions you have loaded. A new database index serves the paged list, so a request no longer sorts every session in the project. Security The pagination cursor is encrypted and tied to both the project and the person viewing it. It cannot reveal a session they are not allowed to open, and it cannot be reused by anyone else. What's Changed Release v0.13.20 — Faster session lists in large projects by @github-actions[bot] in #7316 Full Changelog: v0.13.19...v0.13.20

From the release notes

Changelog New Features: Azure OpenAI Responses: Added AzureOpenAIResponses to use the Responses API with Azure OpenAI deployments. See cookbook. Elasticsearch: Added Elasticsearch vector database with vector, keyword and hybrid search. See cookbook. DocumentationMarkdown: Added a transform for Knowledge.sync_pages that converts Mintlify and Fumadocs components into plain Markdown. See cookbook. Improvements: MCPConfig: Added root_host, path and path_aliases to serve MCP on a dedicated hostname or a custom endpoint path. MCP Server Card: /mcp/server-card now returns pretty-printed JSON. Bug Fixes: MCP Tools: Preserve AudioContent from MCP tool results as audio artifacts. Slack: Deduplicate event retries by event_id instead of dropping them. Remote Agents: RemoteAgent.role and RemoteTeam.role are now properties that return the role instead of a bound method; use .role, not .role(). Readers: TextReader, MarkdownReader and FieldLabeledCSVReader now accept text streams. PPTXReader: Read text inside grouped shapes. JSONReader: async_read() now uses the chunking strategy's async achunk(). Chunking: Preserve newlines and tabs when cleaning text before chunking. CSVReader: async_read() keeps every data row across pages with RowChunking(skip_header=True). SitemapReader: Keep filenames like myindex.html intact and continue discovery past an invalid .xml.gz sitemap. Tools & Workflows: Keep falsy values like 0, False and [] in sync tool results and Parallel workflow step outputs. agnoctl: agno connect now reads and writes client configs as UTF-8 (agnoctl 0.2.1). DynamoDB: Serialize booleans as native BOOL values. Breaking Changes: CodingTools: run_shell is now opt-in (enable_run_shell=True), and restricted mode runs commands without a shell. PublicSurface: With authorization=True and PublicSurface(mcp=True), MCP now accepts only localhost by default; add your domain to MCPConfig(allowed_hosts=[...]). What's Changed fix: read text from grouped shapes in PPTXReader by @devYRPauli in #10037 feat: add support for Elasticsearch vector database by @sannya-singal in #10018 [feat] Add Azure OpenAI Responses model by @rastagan-git in #9841 [fix] Accept text streams in FieldLabeledCSVReader by @n1uz1 in #10020 [fix] Preserve MCP AudioContent as audio artifacts by @devansh173 in #10036 fix: serialize booleans as BOOL instead of {"N": "True"} in DynamoDB storage by @Anai-Guo in #9978 [fix] Use UTF-8 for infra file helpers by @Ghraven in #10113 docs: fix broken external links in cookbook by @Shxiao101 in #9980 fix: pretty-print MCP server-card JSON by @ashpreetbedi in #10084 fix: detect public MCP access alongside JWT REST auth by @ashpreetbedi in #10083 feat: add optional documentation Markdown source transform by @ashpreetbedi in #10094 feat: configure MCP hostname and endpoint routing by @ashpreetbedi in #10090 fix: add missing @Property to RemoteAgent.role and RemoteTeam.role by @kausmeows in #10172 cookbook: export DocumentationM

From the release notes

本版的主线是让"检查通过"意味着它所说的那件事：模型可用性探测此前从不发工具，于是一个只会聊天的中转站通过了验证、在第一条消息上失败，而设置面板仍标着"已验证"；一个 2024 年才有的可选字段被无条件发给每个 OpenAI 兼容端点，让不认它的网关拒掉整场会话。界面一侧，一个中文词在两处字典里有两种英文，英文窗口上的"保存"按钮读作 Transcript；补全菜单的内置动词由内核按进程语言渲染，于是中文机器上的英文窗口配一份中文菜单。另一条主线是老外壳的退役：Wails 那套连同它的 CGO 与 GTK/WebKitGTK 构建要求一次性移除，Studio 从此只有一个外壳；内核与控制器随之做了一轮按主题的归位。自 2.16.0 起 39 个提交，净减两万六千行。 升级路径：无需人工步骤。三平台照旧自更新——Linux 走 .deb，Windows 运行下一个安装器，macOS 替换自身 bundle。已安装的 2.10 及更早的 Wails 版本仍由现有的接管路径升到本版，那条路径不是 Wails 代码，未被此次退役触及。 修复 检查通过，然后第一条消息就失败 模型可用性探测现在带一个工具。Reasonix 是代理，一个递不进工具的模型跑不了一回合。探测此前只发 messages 和 max_tokens，于是一个应答聊天、拒绝 tools 数组的中转站通过了验证，用户的第一条消息才失败——而那一行仍显示"已验证"。被拒时它再问一次、这次不带数组，只有当去掉数组正是让请求答上来的那件事时才报"tools"：结论来自第二次尝试成功，不是拒绝里的某个词。凭据错误与限流不给重试——第二次会因第一次的同一个理由被拒——所以那条"只试一次"的规则对它们原样保留。连接按钮从未声称超出它所证明的（它只列模型，它自己的注释就这么写），但"已连接，OpenAI 兼容"读起来像"可以用了"。现在那一行说得出"端点应答聊天、拒绝工具调用"，这一句话把人送去换网关，而不是送进一场注定失败的会话。 不再为一个 token 计数器赔上整场会话。stream_options 被无条件发给每个 OpenAI 兼容端点。它是 2024 年的新增项，比它更早的、或转发时较真的网关会拒掉带着它的请求，而它要的只是用量记录——一个 token 数，不是一回合需要的任何东西。于是一个别的代理都能驱动的中转站，在这里每一条消息都被拒。客户端现在对每个端点问一次：请求体被拒就去掉这个字段重试，若这样答上来了，此后不再发它。按端点问一次而不是每次被拒都问，所以一个因别的理由拒绝请求体的网关不会被永远反复试探；值得一问的状态码是点名的三个——400、413、422——而不是整个 4xx，因为 402 的意思是账户付不出钱，再发一次什么也不会变。实测于一个为此搭的网关：此前每一回合都失败，现在完成一回合，字段发了一次、再没发过。 一个中文词，一种英文读法 五个中文词条在两份字典里各有一份定义，而目录是十个模块展开进一个对象——同一个键写两遍，留下的是后展开的那一份，另一种读法从此不可达。其中三处渲染了错的词：记录 在指标那边是 "Save"、在运行图那边是 "Transcript"，运行图赢了，于是那个记录模型上下文窗口的按钮读作 Transcript；还原 在窗口控件是 "Restore"、在文件是 "Revert"，文件赢了，于是最大化窗口的还原按钮自称 Revert；改动文件 在权限规则里指 edit_file 这个工具，在顶栏指改动文件数，顶栏赢了。一个中文词承担两种意思，修法是两个中文词，不是更聪明的查表：记录值的按钮改称"保存"，窗口控件改称"向下还原"（Windows 中文里它就叫这个）。另外两处（模型、取消）两种写法渲染相同，失去那份从未胜出的定义；十二个写了两遍但值相同的键失去它们的副本。 补全菜单说窗口的语言，不说进程的。内置动词由内核措辞、在窗口里被读，两者不必一致：一台 CLI 说中文的机器上可以开着英文窗口。端点此前读配置里的桌面语言，而在没有设置时——那是默认值，意思是"跟随机器"——落到本进程按自己的 $LANG 装的那份目录。于是中文机器上的英文窗口，英文输入框下面挂着二十一条中文菜单。窗口才是"它正在画什么"的权威：配置里有值它早已采纳，没有值它跟随自己的区域设置，而那是内核读不到的。所以语言由请求携带，配置只回答那些什么都不说的客户端。 变更 Wails 外壳退役 一次事务，而不是逐件删除：desktop/next 与它的事件泵、remote_pump.go、/rx-replay、前端的 WAILS_* 常量与总线分支、apiPaths/apiPrefixes 白名单、vendored 的 go-webview2 分支，以及 Wails 依赖连同它的 CGO 与 GTK/WebKitGTK 构建要求。先删任何一件留下的都是一个坏掉的外壳，而不是一棵更小的树。 允许它发生的是那个条件本身：不再有任何只有那个外壳才有的语义，加上此前记为"未验证"的那一条——发布线。studio-v2.11.0（2026-09-01）到 2.16.0 已经七次发出 Electron 包，而从已安装的 Wails 2.10 跨过来这件事由维护者确认过，不是在这里假定的。 两个后果值得说出来。desktop 模块现在不含 CGO，所以 CI 与贡献者都不再为构建它安装 GTK 或 WebKitGTK，webkit2_41 标签从每条命令里消失。而老外壳装下的东西原样不动：winuninstall、Electron 外壳的接管路径、studio_line 的二进制名——已安装的 2.10 靠它们变成 2.11，它们都不是 Wails 代码。 迁移契约变成 docs/STUDIO_SHELL_BOUNDARIES.md：它立下的规则活得比迁移本身长，它记的那些账不必。契约里跟踪的展示层缺口原样转为已知缺口，并补上主机侧缺失的崩溃捕获。 内核随之失去只有那个外壳够得到的东西：update.LocalApplication（一个自身即应用的进程，Electron 的不是）、update.Here、appidentity 的 AppUserModelID 应用与快捷方式修复（Electron 自报 appId）、crashreport.CaptureStudioPanic。AppUserModelID 本身留下——Windows 的通知仍带着它。 内核按主题归位 四个各自占着一个包三分之一的文件，按它们服务的那个界面拆开，代码与规则一行未改： controller.go 3232 → 1430。七组方法回到它们的主题已经拥有的文件，回合入口（Send、RunTurn 与它们驱动的循环）拿到自己的一个。留下的是构造函数与访问器——每个主题都要读的那份会话状态——这就是行为搬到状态旁边之后，控制器剩下的东西。审批桥拿到的不止一次搬家：决定一次工具调用跑不跑的 gateApprover 此前住在 controller.go，在敏感路径之外；approval_bridge.go 现在被声明为敏感，评审与覆盖

From the release notes

Changes since langchain==1.4.0 release(langchain): 1.4.1 (#40498) fix(langchain): preserve open MCP object arguments (#40414) fix(langchain): correct InterruptOnConfig documentation (#40140)

From the release notes

Fixes hooks: quote-aware split for RUFLO_HOOK_CLI_OVERRIDE (#3344) — the test-only CLI-override parser in plugins/ruflo-core/scripts/ruflo-hook.cjs did a naive .split(' '), which broke process.execPath into C:\Program plus stray tokens on a standard Windows Node install (C:\Program Files\nodejs\node.exe). Found while verifying #3332's Windows argv fix on a real Windows machine rather than windows-latest CI (whose hosted Node path has no space). No production impact — RUFLO_HOOK_CLI_OVERRIDE is never set outside the test harness. release: spawn npm.cmd with shell:true on Windows (#3346, #3348) — scripts/stage-internal-runtime-bundles.mjs and scripts/prepare-root-publish.mjs both spawned npm.cmd directly without shell:true, which throws EINVAL on any current Node/Windows combo (CreateProcess can't launch a .cmd, and Node has refused to shell out to one implicitly since CVE-2024-27980). This blocked every Windows publish of @claude-flow/cli and the claude-flow umbrella package. Found live while publishing this release from a Windows machine. Verification plugins/ruflo-core/scripts/test-hooks.mjs: 32/32 (was 29/32 before #3344) v3/@claude-flow/cli/__tests__/ruflo-hook-windows-argv.test.ts, escape-cmd-arg.test.cjs, mcp-launch.test.cjs, scripts/smoke-ruflo-hook-cjs.mjs, cross-platform/env-var audits: all clean Full Windows publish pipeline (@claude-flow/cli → claude-flow → ruflo) exercised end-to-end on a real Windows machine after the fixes landed latest / alpha / v3alpha dist-tags verified aligned at 3.42.3 across all three packages 🤖 Generated with RuFlo

From the release notes

Feature: Elixir alias/import/require/use targets now resolve onto the module's defmodule node across files, so the internal module dependency graph is no longer dropped as dangling. Only top-level modules are indexed (a nested defmodule, labeled with its bare inner name, cannot capture an unrelated use <Name> from another file), and a same-file reference is left unresolved so it cannot clobber the structural contains edge (#3603, thanks @ayushcodes10). Feature: a Rust self.method() call now resolves to a method defined on the same type in another file (the common split-impl-block layout), pooling methods across every impl of one type and refusing to link when two unrelated types share a bare name (#3602, thanks @ayushcodes10). Feature: a Ruby member call obj.foo on a known-type receiver now resolves to a method foo inherited from a superclass, including across files, using the same conservative promotion as the implicit-self resolver — a single owning class, matching method kind, and one unambiguous ancestry chain, or it stays dangling (#3585, thanks @oleksii-tumanov). Fix: every edge endpoint written to graph.json is now a declared node. An imports/imports_from/re_exports edge to an external module (stdlib, a third-party dependency) mints a typed external stub node instead of leaving a dangling endpoint that loaders materialise as an attribute-less phantom, and a cross-repo merge unifies the same external module into one global node instead of fragmenting it per repo; sourceless external call targets stay suppressed (#2873, #2878, thanks @AromalBiju1). Fix: a Markdown code-span mention edge now survives an incremental rebuild instead of being pruned as an unauthored link, and a dotted span (Foo.bar) resolves using its qualifiers as evidence, rejecting misleading matches such as time.sleep or pyproject.toml (#3587, thanks @AstroMined).

From the release notes

Changelog Features fb38eae feat: add group add/delete actions to group tree page 39a1bba feat: support popup mode for Captcha providers Bug fixes 0f629eb fix: URL-encode OAuth state on interactive login, consent, prompt, signup and post-logout redirects 1d65427 fix: allow only global admins to change applications' custom HTML f82480a fix: allow org admins to import groups, roles and permissions via xlsx 15e3c21 fix: authorize get-permissions-by-submitter against the signed-in user d484bf5 fix: bind token exchange to the audience and organization of the target application f88e92c fix: block intranet addresses in non-built-in organizations' webhooks d8c06ad fix: check token revocation in token exchange, enforce MFA and verified email when binding social logins 1f4e299 fix: enforce SAML audience, validity window, InResponseTo and replay protection in SP 4c27d13 fix: index token by organization and user to speed up forbidding a user 7fc8c7f fix: mask global cert private keys from org admins 322f84d fix: move password history out of user table to avoid MySQL row size limit cd938de fix: restore Edit LDAP, organization transactions and MCP Store buttons missing from the new frontend 6a99ac7 fix: restrict email, SMS and notification providers to the caller's organization 0af4692 fix: restrict resource storage providers and file paths to the caller's organization and user f81531b fix: scope application credentials to their own organization 47558d3 fix: scope get-resources to the caller and restrict direct listing to admins 0aafb57 fix: scope rename triggers to their own organization 58aa959 fix: scope user rename to the user's own organization

From the release notes

Fixes Model-owned request rules now run when inputs are bound, complete requests are sealed, and generation is planned. Seedance's reference-audio restriction applies consistently to admitted files and upstream component outputs, with an actionable WAV/MP3 conversion error. Successful local HyperFrames workers flush their completion message, close their resources and exit naturally. Failure and cancellation retain process-tree cleanup; cleanup limitations remain visible in diagnostics. Project directory aliases resolve at the Node filesystem entry point. Sources inside an aliased project are accepted, and history queries can still find a Source after its directory is removed. In-project names beginning with .. are accepted while actual parent traversal and symlink escapes remain rejected. Contributor documentation now matches Node 22.15+ and the current release process. The Chinese Studio Companion guide uses the matching filename. Seedance documentation clarifies the existing explicit person-reference attributes for reference images, reference videos and first/last frames. Authors set these on each input; they are not inferred from media or inherited from earlier Results. Distribution checks A separate npm workflow installs the prepared tarball outside the checkout, compiles a project component, loads its font, renders and exports an eight-second video, and decodes the complete output. Publication uses the checked tarball. Linux and Windows repository tests also run before publication. Includes the original contributions from #216, #217 and #218, integrated in #220. Update npm install -g @hypit/hypit@0.1.13 For a project-local installation, update that project's ordinary npm dependency instead. This release updates the npm installation; no separate Skill update is required. Logical interfaces and file formats remain at @1, with no new request, Blob, Need or Result metadata.

From the release notes

What's Changed docs: update Python SDK reference from source by @sdkrelease[bot] in #4479 fix(core): stop dropping auth field metadata from toolkit auth details by Mo (@decknamec) in #4411 docs: broaden harness example category by Brendan O'Leary (@olearycrew) in #4492 docs: project API key permissions by mukund-composio in #4246 docs: add product architecture guides and KB entry points by Soham Basu (@sohambasu963) in #4294 docs(openai): replace assistants examples with responses by Srija Vuppala (@srijavuppala) in #4165 fix(core): contain async Pusher subscription errors by CoralGarden52 in #4448 docs(py): render raises sections and normalize reST in SDK reference by Alberto Schiabel (@jkomyno) in #4491 docs: update documentation for new changelog entries by @sdkrelease[bot] in #4356 docs: add Atomic Agent to Composio Connect clients by Dudka (@sosidudku1) in #4490 perf(cli): store large execute output by byte size, not tokens by Alberto Schiabel (@jkomyno) in #4483 New Contributors Mo (@decknamec) made their first contribution in #4411 mukund-composio made their first contribution in #4246 Srija Vuppala (@srijavuppala) made their first contribution in #4165 Dudka (@sosidudku1) made their first contribution in #4490 Full Changelog: https://github.com/ComposioHQ/composio/compare/@composio/cli@0.4.2-beta.394...@composio/cli@0.4.2-beta.395

From the release notes

The PageIndex SDK, local or cloud — vectorless, reasoning-based RAG, end to end. Much faster indexing — the PageIndex Flash engine gets the tree from layout stats: no LLM involved for the structure generation itself, LLMs only write the node summaries, and tree expansion proposes a wave of nodes concurrently. client = PageIndexClient() client.submit_document("report.pdf") client.chat("What does the report conclude?") Index, to chat, to agent integration, one client. Local mode needs no server, no vector DB, no PageIndex API key. Highlights Flash engine: the local default (mode="standard" keeps the classic LLM pipeline). Embedded bookmarks are consumed when trustworthy, and tree optimization is on by default — optimize="merge" for the deterministic LLM-free pass, "full" (default) adds LLM expand, which runs a wave of nodes concurrently instead of one round-trip at a time. One complete surface, local and cloud: PageIndexLocalClient(storage_path=...) is the same client as cloud — submit, tree, page content, chat, and agent tools all present in both modes, so code moves between them unchanged. Cloud documents, your own model: api_key decides where your documents live; a configured chat model decides who answers — and the two combine. PageIndexClient(api_key="pi-...", chat_model="openai/gpt-5.2") runs the same in-process document-QA engine over the live cloud tool set. Page content flows through your process to your provider on your credentials; doc_id targets at the prompt level; enable_citations stays with the managed chat. Agent integration: the cloud MCP tool contract, in-process — client.agent_tools() (plain functions), as_openai_tools(), as_anthropic_tools(), as_claude_mcp(), plus one-call openai_agent_config() / anthropic_runner_config() / claude_agent_config() bundles and agent_instructions() for the system prompt. Cloud clients get the live server tool set over the MCP bridge (read-only endpoint by default); local clients get the in-process subset with the same schemas and envelopes — agent prompts port unchanged. Chat surfaces: chat() — question in, answer out, on any backend; chat(stream=True) shows the run as it happens, thinking and tool calls woven into the text, or as typed events via .events; chat(protocol="responses" | "messages") drives the OpenAI Responses or Anthropic Messages API natively with that protocol's own shapes, and chat_completions() keeps the OpenAI-compatible envelope — all with doc_id targeting, streaming, honest usage accounting, and prompt-cache continuity across turns. Transcripts append verbatim: a protocol lane's output goes back into the next request unchanged. Model & connection knobs: index_model / chat_model, index_backend / chat_backend (and per-call backend) passed verbatim to each lane — LiteLLM-routed providers, keyless OpenAI-compatible servers, Azure/Bedrock/Vertex included. index= / chat= slots: the grouped spelling of the flat arguments — a string shorthand or a mapping (index={"model": ..

From the release notes

🚀 Nexent：开源智能体平台 / Nexent: Open Source Intelligent Agent Platform 我们很高兴地宣布Nexent v2.6.0 正式发布！🎉 Nexent 是一个开源智能体平台，能够将流程的自然语言转化为完整的多模态智能体 —— 无需编排，无需复杂拖拉拽。基于 MCP 工具生态，Nexent 提供强大的模型集成、数据处理、知识库管理、零代码智能体开发能力。我们的目标很简单：将数据、模型和工具整合到一个智能中心中，使日常工作流程更智能、更互联。 We are excited to announce that Nexent v2.6.0 is released! 🎉 Nexent is an open-source agent platform that turns process-level natural language into complete multimodal agents — no diagrams, no wiring. Built on the MCP tool ecosystem, Nexent provides model integration, data processing, knowledge-base management, and zero-code agent development. Our goal is simple: to bring data, models, and tools together in one smart hub, making daily workflows smarter and more connected. 新功能 / New Features 重构模型管理页面：统一模型编辑对话框，新增模型容量自动推荐（内置 LiteLLM 模型目录、离线可用）、自定义参数校验与推理参数配置。 / Revamped the model management page: unified model edit dialog, automatic model capacity suggestions (bundled LiteLLM catalog, works offline), custom parameter validation, and inference parameter configuration. 新增完整会话线程（Thread）管理能力，覆盖创建、切换与并发控制，并配套遥测指标。 / Added full conversation thread management, covering creation, switching, and concurrency control, with telemetry support. 上线外部记忆体系：智能体可在最终回答后自动提取记忆，支持外部记忆提供方检索（top_k、超时可配置），并提供 AIDP Memory 插件与插件热加载。 / Introduced the external memory system: agents automatically extract memories after final answers, with configurable external provider search (top_k, timeout), an AIDP Memory plugin, and hot-reloadable plugins. 优化 Agent 创建体验：NL2Agent 生成流程与落地卡片打磨，新增 NL2Agent 优化建议，提示词同步升级。 / Improved the agent creation experience: polished NL2Agent generation flow and landing cards, with new NL2Agent optimization suggestions and upgraded prompts. 检索引用交互升级：回答中的引用支持统一悬浮卡片与点击定位高亮。 / Upgraded retrieval citations: references in answers now support unified hover cards with click-to-focus highlighting. 重构统一标签（Label）系统，支持标签库权限、资源打标与筛选，并为多版本 Agent 提供协作版本标签。 / Refactored the unified label system with tag library permissions, resource tagging, and filtering, plus collaborative version labels for multi-version agents. 支持模型优先级排序，可按优先级为智能体选择模型。 / Added model priority sorting so agents can select models by priority. 沙箱适配 Anthropic Skills，可直接运行 Anthropic 技能格式。 / Adapted the sandbox to run Anthropic skills natively. S3 上传/下载工具支持用户自定义配置。 / The upload_to_s3 and download_from_s3 tools now support user configuration. 日志支持落盘为文件，便于排查与归档；服务启动时自动恢复中断的后台任务。 / Logs are now written to disk as files for easier troubleshooting, and interrupted background tasks are automatically recovered on service startup. Bug 修复 / Bug Fixes 完成上下文预算契约迁移，增加有界压缩与真实溢出恢复，长对话不再丢失上下文。 / Completed the context budget contract migration with bounded compaction and real overflow recovery, preventing context loss in long conversations. 修复会话链路多处问题：返回后恢复会话、线程切换保留智能体、会话标题使用会话模型生成、正确渲染执行代码与对比模式下的历史保留。 / Fixed several conversation-flow issues: conversation restore after returning, agent retention across thread switches, session-model-based title generation, correc

From the release notes

Two fixes on 0.1.11. Seedance rejects m4a reference audio at both layers. No Seedance provider accepts the container, so the Surface decode rejects an admitted .m4a file at authoring time, and the referenceAudio binding Producers apply the same check at runtime to Blobs produced by upstream components. Both errors direct converting the file to WAV or MP3 — <media:ExtractAudio> already produces canonical WAV. Runtime gap reported by @Flandern1211 in #219. Local HyperFrames render cleanup no longer requires ps. On hosts whose sandbox denies process enumeration, the post-render cleanup probe failed and discarded every completed render. Descendants are now enumerated with parent-scoped pgrep, the whole tree receives SIGKILL before the wait loop so reparenting reaps each process, and a completed render survives a failed best-effort cleanup. Reported by @goatyyc in #199, with an alternative fix proposed by @521Peter in #204. Update npm install -g @hypit/hypit@0.1.12 Logical interfaces and file formats remain at @1.

From the release notes

1.10.0 (2026-09-16) Features add @experimental annotation for unstable ADK APIs (69914f4) add EventActions.agentState for session-resumability checkpoints (8505a01) add one-way ADK Java to Kotlin engine interop module (baff584) add opt-in ResumabilityConfig flag for plain-text continuation auto-resume (f0fb722) Add unique event_id to BigQuery agent analytics events (15c70de) bridge ADK Java plugin onRunErrorCallback to the Kotlin engine (4ee155b) fire ADK Java plugin tool callbacks for native Kotlin tools (fda5a10) run an ADK Kotlin-engine runner behind the ADK Java Runner API (6ff8241) Bug Fixes dev: serve the dev UI and its assets under a /dev-ui path (6866919) memory: format search timestamps as epoch millis (5269b87) models: set additionalProperties on schemas for OpenAI strict structured outputs (4b058cd) sessions: forward the caller-supplied session id in VertexAiSessionService (77eae25) sessions: validate HTTP status before parsing Vertex AI session responses (4192aca)

From the release notes

Patch Changes 5b015ad: Removed the hidden e2b template build command (alias bd). It printed a V1 deprecation notice and exited 1; the V1 build system it fronted is gone from the API. A script still calling it now gets commander's unknown-command error with the same exit code. Build templates with e2b template create or the Template SDK; e2b template migrate still converts a V1 project (see https://e2b.dev/docs/template/migration-v2). Updated dependencies [956e3ab] Updated dependencies [e1fbe86] Updated dependencies [80496c0] Updated dependencies [5b015ad] Updated dependencies [80496c0] Updated dependencies [e1fbe86] e2b@2.50.0

From the release notes

Minor Changes 5b015ad: Removed the V1 template build operations and schemas from the generated API clients. The API no longer serves them (runtime 87968fc1e1fa); control planes carrying that change answer 410 Gone. Template builds go through the Template SDK. Visible removal, classified minor: the JS paths namespace loses POST /templates, POST /templates/{templateID}, POST /templates/{templateID}/builds/{buildID} and POST /v2/templates, and components['schemas'] loses TemplateLegacy, TemplateBuildRequest and TemplateBuildRequestV2; the Python e2b.api.client.models package loses TemplateLegacy, TemplateBuildRequest and TemplateBuildRequestV2, and e2b.api.client.api.templates loses the post_templates, post_templates_template_id, post_templates_template_id_builds_build_id and post_v2_templates modules. No SDK method accepted or returned them; code that imported these names directly must drop the import. The regenerated clients also pick up a documented 429 on most operations, a 409 on template create (v3), the upload-request headers on the build file-upload link, minLength: 1 on the v2 build source fields, and a deprecation marker on the always-empty logs field of the build status. Patch Changes 956e3ab: Apply the request headers the API returns with a template layer-file upload link. Azure Blob Storage requires x-ms-blob-type on the upload request, which its signed URL cannot carry, so COPY instructions failed on Azure-backed clusters. GCS- and S3-backed clusters return no headers and are unaffected. e1fbe86: Bump @connectrpc/connect and @connectrpc/connect-web to 2.2.0 (fixes Http2SessionManager.verify() hanging when the connection closes mid-verification and stops delivering stream messages after cancellation). 80496c0: Fix uploads of non-native ReadableStreams in the browser silently sending the text [object ReadableStream] instead of the data. Buffering a stream drained it with new Response(stream), which accepts any async iterable on Node (an undici extension) but only its own stream class in a browser, stringifying anything else. Streams are now drained through the reader, which every implementation supports. 80496c0: Report a sandbox killed mid-request as an actionable TimeoutError in the browser. When the connection to a sandbox drops mid-request the SDK probes the sandbox's health to tell a killed sandbox apart from a transient network blip, but the probe only ran for connection-dropped wordings it recognized, and the browser's (network error) was missing — so killing a sandbox while a command was running surfaced a generic SandboxError: [unknown] network error instead of TimeoutError: ... The sandbox was killed or reached its end of life while the request was in flight. e1fbe86: Bump undici to 7.29.1 and the optional undici8 fallback to 8.10.2 to pick up the September 2026 security fixes.

From the release notes

Minor Changes 5b015ad: Removed the V1 template build operations and schemas from the generated API clients. The API no longer serves them (runtime 87968fc1e1fa); control planes carrying that change answer 410 Gone. Template builds go through the Template SDK. Visible removal, classified minor: the JS paths namespace loses POST /templates, POST /templates/{templateID}, POST /templates/{templateID}/builds/{buildID} and POST /v2/templates, and components['schemas'] loses TemplateLegacy, TemplateBuildRequest and TemplateBuildRequestV2; the Python e2b.api.client.models package loses TemplateLegacy, TemplateBuildRequest and TemplateBuildRequestV2, and e2b.api.client.api.templates loses the post_templates, post_templates_template_id, post_templates_template_id_builds_build_id and post_v2_templates modules. No SDK method accepted or returned them; code that imported these names directly must drop the import. The regenerated clients also pick up a documented 429 on most operations, a 409 on template create (v3), the upload-request headers on the build file-upload link, minLength: 1 on the v2 build source fields, and a deprecation marker on the always-empty logs field of the build status. Patch Changes 956e3ab: Apply the request headers the API returns with a template layer-file upload link. Azure Blob Storage requires x-ms-blob-type on the upload request, which its signed URL cannot carry, so COPY instructions failed on Azure-backed clusters. GCS- and S3-backed clusters return no headers and are unaffected.

From the release notes

Two fixes for installations from npm, both reported against 0.1.10. A contributor checkout links these packages through node_modules; a published distribution resolves them through the machine package root and the Distribution resolver, which is where both failures lived. Authored fonts resolve from an npm installation. hypit packages install places an upstream asset under the machine package root, and the version selection now reads the same optional declaration that install was addressed to, so hypit check passes after the install it asks for. Reported by @daodaobing in #206 and by @wangyufeng336 in #211, fixed by @daodaobing in #208. Local HyperFrames rendering runs from an npm installation. The capture process preloads the package resolution its launcher installed, so the render entry point resolves @hypit/hyperframes. Reported by @daodaobing in #207 and by @wangyufeng336 in #211, fixed by @daodaobing in #209. Update npm install -g @hypit/hypit@0.1.11 Logical interfaces and file formats remain at @1.

From the release notes

What's Changed feat: label and review tier for pinned app registries (backport #11155) by @bolichen97 in #11184 Full Changelog: v0.7.0-insider.2...v0.7.0-insider.3

From the release notes

Fix terminal connections that fail after a sandbox resumes because its provider credentials have changed. Refresh rejected provider credentials once for reads and discard stale credentials after a failed WebSocket handshake. Preserve application redirects and sandbox authentication errors. Do not replay writes when provider authentication fails. What's Changed chore(release): staging VERSION → 0.13.19 [skip ci] by @github-actions[bot] in #7284 fix: release provider credential recovery after sandbox resume by @markokraemer in #7287 Release v0.13.19 — Reliable terminal connections after sandbox resume by @github-actions[bot] in #7288 Full Changelog: v0.13.18...v0.13.19

From the release notes

Signed-off-by: omnigent-ci[bot] <294685417+omnigent-ci[bot]@users.noreply.github.com>

From the release notes

Fixed Terminal opens on a stopped workspace. Opening Terminal requests a wake even before the first shell exists. Connection retries have a deadline and preserve shell output. Session switches keep message retries in the right workspace. Missing conversations stop repeated 404 or 410 requests without clearing cached history. Sessions keep their own agent. A session could be switched to an agent from a different project after a prompt that named no agent, which removed its access to connectors and the Kortix CLI. Every session now always runs as its own agent, and a session that was affected recovers on its next message. The first message on the home screen no longer gets stuck in the composer after a slow start. Project pages load cleanly. The project home shows the Kortix mark while it loads instead of a flashing placeholder. Session history is saved before a session stops, and responses from the in-sandbox model proxy are no longer returned in a corrupted encoding. Admin project list shows each project's own session counts. Customize settings show what you saved. After saving an agent or connector setting, a reload could show the previous value for up to a minute. Improved Connector activity uses consistent names. Connector calls appear as connector calls. Repository access is explicit. Agents now declare whether a session gets the project's files, replacing the older workspace modes. Older API clients keep reading restricted manifests correctly. Previews route correctly and set up shared sandboxes in isolation. Release checks are more reliable: preview environments use their own database, confirm the running runtime, and wait for completed agent artifacts before asserting on them. They also verify cold terminal wake and session-switch routing, allocate sufficient preview frontend memory, and check Docker readiness. What's Changed chore(release): VERSION → 0.13.16 [skip ci] by @github-actions[bot] in #7249 chore(release): VERSION → 0.13.17 [skip ci] by @github-actions[bot] in #7253 Replace workspace modes with explicit repository access by @markokraemer in #7240 fix(preview): a legacy HS256 session opens preview origins again by @markokraemer in #7258 fix(web): a timed-out first prompt no longer stays in the home composer by @sutharjay1 in #7254 fix(web): project home loading paints the Kortix mark, not a skeleton by @sutharjay1 in #7263 fix(api): session tokens no longer adopt another project's agent by @markokraemer in #7262 Release candidate: main 3e7d039 → staging by @markokraemer in #7266 chore(release): VERSION → 0.13.18 [skip ci] by @github-actions[bot] in #7269 chore(release): staging VERSION → 0.13.18 [skip ci] by @github-actions[bot] in #7270 hotfix(staging): editor manifest reads are fresh on every replica; release setup polls retry transport errors by @markokraemer in #7273 fix(web): connector tool calls say "connector", not "app" by @markokraemer in #7272 fix(api): Customize editor reads show the committe

From the release notes

v3.42.1 — SONA env-default, review batch, and reliability fixes A patch release: one small feature (opt-in, fully backward-compatible) plus a batch of independently-reviewed bug fixes, a performance fix, and a dependency bump — all reviewed, tested, and merged in this pass. What's new for you SONA learning mode can now default from the environment RUFLO_INTELLIGENCE_MODE sets a fleet-wide default learning mode (real-time, balanced, research, edge, batch) without threading mode/sonaMode through every call site — useful for a managed deployment that wants every session to learn in a specific profile. An explicit per-call mode always wins; an unset or unrecognised value falls through to balanced, so a typo can never silently pick a profile. (Originally landed with a bug where one of the two places reading this variable only checked it once per process instead of per session — fixed before release, with new regression tests in both places.) Diagnostics and memory fixes ruflo doctor no longer guesses your memory driver from a raw table count (a heuristic that never actually correlated with which driver you're using) — it now reports only what it verifies directly. ruflo doctor now resolves the installed MetaHarness version correctly instead of reporting stale/incorrect data. memory_stats now answers from the live store instead of a snapshot that couldn't see in-flight writes — a working memory store no longer reports itself as uninitialized. Also fixes a namespace named __proto__ silently vanishing from the breakdown, and a hardcoded row cap that was truncating large stores without saying so. A database handle used during memory initialization now closes on every code path, including the failure path that used to leak it. Opening the graph database now verifies the handle is actually persistent before using it — a bad argument shape used to silently produce an in-memory-only graph that vanished on exit, with no error. Fixed a memory-key bug on Windows where a drive letter's colon wasn't sanitized consistently with how Claude Code derives the same key. Federation and security The federation gateway's untrusted-relay data is now read through the same validated envelope shape the gateway actually sends, closing a gap where some tools could miscount or silently see zero messages. WITNESS_ED25519_ROOT now actually resolves @noble/ed25519 for witness-manifest signing (this only affects the maintainer signing path — the separate node:crypto-based signature verification path was never affected). Performance AgentBBS now dedupes redundant peer-hello and watch-replay traffic, and computing the next sequence number no longer requires scanning an entire (potentially large) room log. Housekeeping swarm status/swarm stop now report a real swarm id and no longer get confused by a stale "stopped" state file shadowing a live swarm. Bumped agent-browser to 3.37.1 for recording/WebMCP improvements and a Windows headless-Chrome cleanup fix. Upgrading

From the release notes

fix(bedrock): report the expected key schema instead of a JSONDecodeE…

From the release notes

Major new features 🔀 Work across GitHub repositories: track multiple PRs in one session, switch between their summaries and diffs, and link or unlink PRs manually. Start managed sandboxes with multiple repositories and branches, cloned in parallel. (#6935, #6995) 🎛️ A smoother chat interface: configure workspace, agent, model, effort, and permissions in the redesigned composer. Find sessions with fuzzy search, open the model picker from the keyboard, and inspect running background commands from the task badge. (#6345, #7226, #7413, #7061) 🪟 Better Claude context management: Claude SDK sessions now show context-window usage and support /compact for native compaction, including with Claude Pro/Max logins. (#6119) 🗄️ Expanded sandbox and database support: provision managed sandboxes through Gensee, run Kubernetes sandboxes on dedicated node pools using tolerations, and use CockroachDB as Omnigent's application database. (#7229, #6881, #6360) Bug fixes 🔁 Pending approvals and requests for user input remain usable after server or runner restarts, and plans remain visible. (#6696, #6964, #6629) ↩️ Codex sessions resume more reliably, preserving conversation history, permission settings, and selected reasoning effort. (#7055, #7147, #7233, #6860) ⏳ Chat more accurately shows when an agent is working or finished, and completed answers remain visible when background tasks resume. (#6957, #6879, #7002, #7191) 🎨 Model selectors display readable names, and model rate-limit errors can be retried directly from chat. (#7260, #7235) ⎋ Escape reliably interrupts a working agent from the message input, and archived sessions no longer reappear in the sidebar. (#6871, #6481) 🛡️ Improved validation of uploaded agent settings and ensured tool approval retries retain the arguments that were reviewed. (#7457) 💜 Thanks to our community This release was shaped by the people who filed issues, opened PRs, and shared feature requests on Discord. Thank you for building Omnigent with us—keep the bug reports, ideas, and contributions coming! Full Changelog: https://github.com/omnigent-ai/omnigent/blob/main/CHANGELOG.md

From the release notes

Fixed Previews open again when you are signed in. A sandbox preview opened from a session (prod-p<port>-sbx-….p.kortix.com) showed "Sign in to open this preview" even though you were signed in. Sign-in tokens are still issued with the older HS256 signature while the published key set already holds the newer ES256 key. The API already verified those tokens with the auth server, but the preview gate refused them. It now makes the same check, so the preview panel, pasted preview links, and preview WebSockets open for the account that owns the sandbox. A forged or expired token is still refused, and a test now fails if any verifier caller handles these results in its own way. Internal The release gate quarantines TUN-6 on deployed targets only: Bun's WebSocket client sends no User-Agent, so the edge firewall refuses the tunnel handshake on staging and prod. The flow still runs locally. What's Changed chore(release): staging VERSION → 0.13.16 [skip ci] by @github-actions[bot] in #7248 chore(release): staging VERSION → 0.13.17 [skip ci] by @github-actions[bot] in #7252 test(release): quarantine TUN-6 — Bun WebSocket handshake has no User-Agent, WAF 403s it on staging/prod by @Ino-Bagaric in #7255 hotfix(preview): a legacy HS256 session opens preview origins again by @markokraemer in #7259 Release v0.13.17 — Signed-in previews open again by @github-actions[bot] in #7261 Full Changelog: v0.13.16...v0.13.17

From the release notes

typescript/v1.18.0 Auto-drafted from commits in typescript/v1.17.0..typescript/v1.18.0, grouped by conventional-commit type. Edit on the release page after publish if you want a polished writeup; the canonical release notes live on the website. 🚀 Features feat(storage): add bm25 search strategy (#4079) (495d154) feat(sandbox): include partial output in shell timeout errors (#4325) (bd4020f) feat(vended-tools): port Python notebook improvements to TS (#4281) (6c79888) feat(context-py): port strategy presets and agent rewire (#4282) (08ed4cf) feat(context-ts): add context strategy presets + rewire defaults to use class (#4256) (a8b1b90) feat(bidi): render live transcripts with audio output (#4287) (381ab48) feat(vended-tools): add web_fetch tool for TypeScript (#4153) (f800228) feat(context-py): add session manager integration (#4254) (bea15fe) feat(bidi): add transcript completion events (#4230) (4ece5b0) feat(bidi): share lifecycle hooks and add response completion hooks (#4280) (90a5a02) feat: automatically use openAI prompt-cache keys from session id (#4083) (7462908) feat(bidi): align model configuration and parameter overrides (#4255) (0d2beba) feat(vended-tools): include exit_code in shell tool result (#4269) (1efa1ee) feat(vended-tools): port notebook tool to Python (#4132) (545d9ed) feat(bidi): simplify Google and OpenAI model configs (#4189) (6d4f61f) feat(python): add backgroundTasks to Agent (#4226) (42e5a61) 🐛 Fixes fix(agent): make AgentResult.to_dict JSON-serializable with bytes (#4313) (39da4bb) fix(python): force structured output retry by tool name (#4263) (f627f54) fix(mistral): preserve image blocks in requests (#4200) (8423dd3) fix(bidi): align reference semantics with Agent and Model (#4286) (0a26e12) fix(skills-py): avoid leaking YAML parse error and dropping valid ski… (#4192) (9663bcf) fix(bidi): clean up CRT streams on shutdown (#4253) (7bda6c7) fix(models): route Bedrock Mantle openai.gpt-6-* to /openai/v1 (#4267) (056f325) fix(session): filter malformed immutable snapshot IDs (#4199) (6d6c8fc) fix(telemetry): use semantic system instructions for agent spans (#4222) (e5ebbd6) fix(models): resolve nested Bedrock model prefixes (#4221) (21b27d4) fix(context): fix various context manager parity items (#4228) (bebf3e8) fix(openai): emit tool results before user text to preserve tool_use/tool_result adjacency (#4234) (5f636bb) ♻️ Refactoring refactor(bidi): refine model audio configuration (#4303) (a1e7e4c) refactor(bidi): rename model configuration validator (#4297) (a04273c) refactor(python): share repository session methods through LocalAgent (#4257) (08cc6e7) 📚 Documentation docs(context-manager): split into modes, strategies, and presets pages (#4336) (af902b6) docs(context-management): update to ContextManager strategy API (#4332) (d1e1d0a) docs(blog): add framework-agnostic evaluation post (#4114) (8709dd7) docs: ai usage reflection blog (#4148) (1ede04b) docs(changelog): sync st

From the release notes

python/v1.56.0 Auto-drafted from commits in python/v1.55.1..python/v1.56.0, grouped by conventional-commit type. Edit on the release page after publish if you want a polished writeup; the canonical release notes live on the website. 🚀 Features feat(anthropic): non-clobbering seam for Anthropic-direct server-side tools (web search) (#3568) (a0cbf29) feat(storage): add bm25 search strategy (#4079) (495d154) feat(sandbox): include partial output in shell timeout errors (#4325) (bd4020f) feat(vended-tools): port Python notebook improvements to TS (#4281) (6c79888) feat(context-py): port strategy presets and agent rewire (#4282) (08ed4cf) feat(context-ts): add context strategy presets + rewire defaults to use class (#4256) (a8b1b90) feat(bidi): render live transcripts with audio output (#4287) (381ab48) feat(vended-tools): add web_fetch tool for TypeScript (#4153) (f800228) feat(context-py): add session manager integration (#4254) (bea15fe) feat(bidi): add transcript completion events (#4230) (4ece5b0) feat(bidi): share lifecycle hooks and add response completion hooks (#4280) (90a5a02) feat: automatically use openAI prompt-cache keys from session id (#4083) (7462908) feat(bidi): align model configuration and parameter overrides (#4255) (0d2beba) feat(vended-tools): include exit_code in shell tool result (#4269) (1efa1ee) 🐛 Fixes fix: deliver background results through the registered management tool (#4347) (45c2ffd) fix(agent): make AgentResult.to_dict JSON-serializable with bytes (#4313) (39da4bb) fix(python): force structured output retry by tool name (#4263) (f627f54) fix(mistral): preserve image blocks in requests (#4200) (8423dd3) fix(bidi): align reference semantics with Agent and Model (#4286) (0a26e12) fix(skills-py): avoid leaking YAML parse error and dropping valid ski… (#4192) (9663bcf) fix(bidi): clean up CRT streams on shutdown (#4253) (7bda6c7) fix(models): route Bedrock Mantle openai.gpt-6-* to /openai/v1 (#4267) (056f325) ♻️ Refactoring refactor(bidi): refine model audio configuration (#4303) (a1e7e4c) refactor(bidi): rename model configuration validator (#4297) (a04273c) refactor(python): share repository session methods through LocalAgent (#4257) (08cc6e7) 📚 Documentation docs(context-manager): split into modes, strategies, and presets pages (#4336) (af902b6) docs(context-management): update to ContextManager strategy API (#4332) (d1e1d0a) docs(blog): add framework-agnostic evaluation post (#4114) (8709dd7) docs: ai usage reflection blog (#4148) (1ede04b) docs(changelog): sync strands-agents/harness-sdk python/v1.55.1 (#4259) (66101c7) ✅ Tests test(mcp-py): pin native OTel trace continuity on the mcp 2.x version (#4131) (7edb796) 👷 CI ci(docs): bump vite from 8.2.2 to 8.3.0 in /site (#4319) (b14fef6) ci(docs): bump node-html-parser from 9.0.2 to 9.0.4 in /site (#4264) (1b13f6c)

From the release notes

What's new 🌐 Introducing Rowboat Spaces. Every teammate. Their own agent. One Space. Rowboat becomes a multiplayer personal assistant for work. A Space brings your team’s conversations, files, and whiteboards together, and everyone brings their own Rowboat: their own memory, tools, and model keys. Your personal assistant can now work alongside your teammates and their assistants in the same room. 🤖 Ask in the Space. Your Rowboat takes it from there. Type @rowboat what is missing? and your Rowboat picks it up on your machine, draws on your Brain, email, meetings, and notes, and brings the answer back as “You (via Rowboat)”. Your personal knowledge stays on your machine; the Space holds what you and your agent choose to share. The room gets a 👀 while work is underway and a ✅ when it settles. If your Rowboat needs a decision, permission, or a private conversation, it leaves a ❗ and asks in your own chat. You can also open the agent conversation behind a Rowboat reply to follow the work. 🎨 From a shared whiteboard to something you can ship. Sketch together with live cursors, then ask @rowboat turn this board into a proposal. Agents can draw on the same board and work with the same files as the team. Markdown documents have version history, diffs, and restore; concurrent text edits merge line by line, with conflicts surfaced for resolution. Open documents and presentations in place, or link a discussion to a file and read them side by side. When the proposal is ready, @rowboat implement the proposal can hand it to Claude Code or Codex on your machine, using the Space’s files as context, and bring the result back for review. 💬 The everyday parts of teamwork are here too. Spaces includes threaded conversations, DMs, notes to yourself, reactions, polls, scheduled messages, and mentions. The composer has formatting controls, link previews, attachments, and voice dictation. An Activity view gathers mentions, DMs, replies in followed threads, and reactions on your messages. Unread state follows your account across devices, and notifications take you back to the conversation. On macOS, the Dock icon shows a count for mentions and DMs, or a dot for other unread activity. 🚪 Start a Space and bring your team. Spaces is now available by default. Open Spaces → Create a server, sign in with Google or Microsoft, and use Copy invite link to bring teammates in. A server is your team’s home for its Spaces; you can create or join more than one. For teams that want to run their own infrastructure, Harbor, the server behind Spaces, is open source and self-hostable. It also exposes an MCP server, so other agents can participate using the same tools as Rowboat. See the Spaces guide or explore Spaces on our website. 🪟 Keep your assistant wherever the work is. Move a conversation between the full Assistant page, a right sidebar, and a floating window. Keep several floating chats open, resize and reorder them, or minimize them into tabs along the bottom. The app now ope

From the release notes

🐈 nanobot v0.3.5 brings the workbench to the terminal — and makes conversations easier to continue across the browser, terminal, and chat apps. The headline is one agent, more places to work. Run nanobot for the native terminal client or nanobot webui for the browser. Both use the same gateway and saved WebSocket conversations. Terminal sessions can be resumed, forked, inspected, and detached without building a separate agent runtime. The browser workbench has grown around real multitasking: grouped conversation panes, temporary chats, persistent sidebar ordering, session mentions, a skills marketplace, portable Agent Plugins, and browser OAuth for remote MCP servers. Automations now have task and calendar views, clearer scheduling controls, and per-run replies. Mobile composition, settings, context charts, app logos, and branded mentions received another round of focused polish. The other story is continuity you can inspect. Context usage and compaction are visible, interrupted work has clearer recovery controls, and session history is more durable across pagination, reconnects, and upgrades. This release also tightens attachment, process, workspace, and email-authentication boundaries. Please read the upgrade notes before replacing an existing installation. Highlights A native terminal workbench — Start with nanobot, switch sessions, paste supported clipboard images, complete app and skill references, inspect context and file diffs, queue follow-ups, and use /detach to leave the gateway running. Browser conversations side by side — Arrange up to four topics in resizable groups, keep sidebar ordering, drag saved sessions into mentions, and use temporary chats when you do not want a persistent topic. Automations that are easier to manage — Switch between task and calendar views, inspect run replies, and adjust schedules with clearer controls. Editing a job preserves pending executions; deleting one no longer leaves navigation blocked. Visible context and steadier long sessions — Per-round token/cache charts, context-compaction progress, paged history search, cumulative summaries, and file-read deduplication scoped to the actual model context. Apps and skills with clearer identities — SkillHub discovery, portable Agent Plugins, remote MCP OAuth and token refresh, visible connection failures, and consistent app logos and brand names in chat. A better small-screen experience — Home-screen installation through PWA support, narrower composer controls, draggable context sheets, touch-friendly chart details, and settings navigation that matches its opening direction. Broader model and search choices — DeepSeek Responses and V4 vision support, Eden AI and OrcaRouter gateways, AnySearch, online OAuth model catalogs, and more resilient provider fallback and Responses-history handling. More reliable channels and runtime boundaries — Rich Telegram streaming, improved Feishu onboarding, safer QQ attachments, trusted email authentication results, o

From the release notes

2.9.1 (2026-09-15) Highlights This release focuses on improving visibility into model reasoning steps when utilizing Claude's adaptive thinking capabilities. Claude Integration: Access the step-by-step reasoning process of Claude models by ensuring visible thoughts are requested during adaptive thinking tasks. (ba0d542) Bug Fixes request visible thoughts for Claude adaptive thinking (ba0d542)

From the release notes

Feature: Terraform module calls with a literal local source (./… or ../…) now resolve to a directory-scoped module node, exposing the caller→implementation topology (e.g. environment → application → base); remote and registry sources and source expressions are left unresolved and never fabricate a target. After upgrading an existing graph, run graphify update . once to regenerate Terraform ids and topology (#3571, thanks @vstepko). Feature: a bare implicit-self call in a Ruby method now resolves up the inheritance chain to a method defined on a superclass, including across files. The promotion is conservative — it only upgrades an existing inferred edge when a single owning class, a matching method kind, and one unambiguous ancestry chain all agree, and it bails on mixins, reopened classes, and dynamic superclasses rather than guess (#3569, thanks @oleksii-tumanov). Feature: an inline code span in a Markdown file (e.g. `Widget.render`) now emits a references edge to the symbol it names, but only when exactly one corpus callable matches — generic words and ambiguous names link to nothing, so docs join the graph without a false-edge firehose (#3562, thanks @AstroMined). Fix: a prose file whose stem is the bare plural "tokens" (e.g. TOKENS.md) is no longer excluded as a credential dump; singular token.md and other keywords' bare plurals still exclude, and content-based secret detection is unchanged (#3527, thanks @HARSHAVARDHAN-RAJU5). Fix: a call whose only resolution target is a sourceless external stub no longer emits a spurious calls edge, so an unresolved external name stops accruing a phantom god-node (#3156, thanks @DevChiniwala). Fix: calls made inside a module-level anonymous closure (an IIFE or a top-level callback) are now attributed to the file node instead of being dropped (#3124, thanks @ayushcodes10). Fix: cross-repo merge now respects C# namespace/using scoping — a parked member call binds to a same-named type in another repo only when the caller's namespace or an in-scope using resolves it, so a third-party receiver no longer binds to an unrelated repo-local type (#3360, thanks @DevChiniwala). Fix: a bare Lua require("mod") statement (no assignment) now emits an imports edge, including the chained require("lazy").setup({}) LazyVim idiom, so Neovim configs graph correctly instead of coming out as disconnected files (#3320, thanks @A-Levin). Fix: the Claude/CodeBuddy PreToolUse guard hooks now carry a timeout, so a wedged filesystem stat can no longer stall a tool call indefinitely (#3314, thanks @ayushcodes10). Fix: id-prefix reconstruction is now Unicode-aware — a non-Latin path segment is slugified with the canonical normalizer instead of being dropped, so files under non-ASCII directories keep stable ids (#3559, thanks @ayushcodes10). Fix: cohesion_score now excludes self-loops (e.g. a recursive call) from the edge count, so a community's score stays within 0..1 instead of being inflated above 1 (#3558, thanks @jordanale

From the release notes

Changelog a325fac chore: bump version to v3.20.78 [skip-release] [skip ci] b2950b3 feat(core): SQL roles_query in union role mode 9c1192e feat(core): per-role read grants in sources mode 6357681 feat(core): union role mode and trusted $groups variable aebaff0 fix(dialect): in and nin with array variables on MySQL and MariaDB text columns 33cd5e6 fix(dialect): nin with an array variable on MongoDB 3b32e85 fix(qcode): reject role filters that compile to nothing 75795ad refactor(core): rename the groups variable to $user_groups

From the release notes

2.1.0 (2026-09-14) Miscellaneous Chores integrations: Synchronize adk versions Dependencies The following workspace dependencies were updated dependencies @google/adk bumped from ^2.0.0 to ^2.1.0

From the release notes

2.1.0 (2026-09-14) Features Add ContainerCodeExecutor for Docker-sandboxed code execution (#541) (a52b669) dev: add --min_instances/--max_instances flags to agent_engine deploy (#899) (0dc138c) dev: add Origin validation to the dev server (DNS-rebinding defense in depth) (#557) (f280eba) Bug Fixes build: stop compiling the dev CLI entrypoint into dist/cjs (#861) (da9a492) cli: quit adk run on a whitespace-padded exit (#896) (0568990) cli: stop staging an empty node_modules and a zero-byte lockfile into the deploy image (#894) (fd4ac9a) cli: surface model errors and make agent stack traces readable (#625) (9508dee) cli: use GOOGLE_GENAI_USE_ENTERPRISE instead of deprecated GOOGLE_GENAI_USE_VERTEXAI (#897) (6dc4a3d) core: make transfer and runtime tool confirmation work across runtimes (#808) (6e251bb) deploy: pass shell:true to spawn on win32 to fix ENOENT for .cmd binaries (#888) (b46196b) dev: let --bundle false actually load an agent (#791) (91357fd), closes #714 dev: print the command help after a CLI usage error (#892) (206dc97) dev: route AgentFile.load multiple-apps/agents warnings through AdkLogger (#879) (905e730) guard reads in the dev server against DNS rebinding (#744) (0fe9581) remove the no-op @ts-ignore on the esbuild-shim-plugin import (#805) (9566743) Dependencies The following workspace dependencies were updated dependencies @google/adk bumped from ^2.0.0 to ^2.1.0

From the release notes

2.1.0 (2026-09-14) Features Add ContainerCodeExecutor for Docker-sandboxed code execution (#541) (a52b669) Add VertexAiRagMemoryService (Vertex AI RAG memory backend) (#543) (8543e9c) auth: emit a PKCE S256 code challenge on the OAuth2 authorization URI (#820) (6db4727) core: export resetLogger from the @google/adk public API (#882) (d4bdbd3) models: add a Chrome Prompt API BaseLlm for the on-device model (#843) (77b0803) plugins: add ReflectAndRetryToolPlugin and ReflectAndRetryModelPlugin for self-healing error recovery (#631) (1b6ea6d) Port the typed-errors module (errors/) from adk-python (#583) (fab19d0) Rethinking ForwardingArtifactService Interface (Part 2/2) (#510) (cea2dca) Rethinking ForwardingArtifactService Interface and Removing Unnecessary Parameters in ADK JS (Part 1/2) (#508) (77809e3) tools: port the adk-python v0.1.0 retrieval tools (#854) (5ec1c94) Bug Fixes a2a: harden agent-card resolution against SSRF and a silent file read (#832) (1bb1120) a2a: propagate request metadata to agent execution (#752) (#768) (bb2dd8f) agent-registry: declare the synthesised card default modes as media types (#878) (2024ab2) agents: carry request-input resume inputs in their own type (#822) (8664b6e) always emit an array for the OpenAPI argument schema required key (#895) (dc30492) artifacts: list artifacts nested under another artifact (#858) (caa2b4d) auth: redact OAuth2 authorization code from malformed-callback-URI log (#640) (08f4f24) auth: reject an empty credentialKey before storing the credential (#818) (8f792bb) build: make the web build loadable outside Node (#837) (5b9c306) check every function response, not just the first, when resolving which agent to resume (#810) (aee975f) close two remaining author-blind confirmation-resolution paths (#755) (f9d6d91) context: respect isolation scopes during compaction (#830) (cda8433) core: configure VertexAiSessionService from environment for vertexai:// URIs (#898) (600b756) core: drop the unused @mikro-orm/reflection dependency and pin the DB-driver manifest contract (#814) (6ea9ce5) core: gracefully handle unregistered tool calls (#790) (6be4195) core: keep winston out of the published browser build by splitting the logger by platform (#813) (c2c2ecb) core: load artifacts and MCP resources when other tools are call… (#639) (b723201) core: make built-in tools resolvable by name (#811) (e7fab37) core: make transfer and runtime tool confirmation work across runtimes (#808) (6e251bb) core: redact inline data payloads from the resumption debug event dump (#817) (0c9d9ba) core: render an empty code block when an executableCode part has no code (#868) (649afd0) core: stop a surviving grandchild from hanging code execution (#793) (173462a), closes #622 core: type the live session mock in gemini_llm_connection_test (#849) (bfe904e) drop the redundant ExtendedInteractionSSEEvent self-cast in the streaming loop (#827) (e9f187f) make the web build bundle and run in a b

From the release notes

2.1.0 (2026-09-14) Features Add ContainerCodeExecutor for Docker-sandboxed code execution (#541) (a52b669) Add VertexAiRagMemoryService (Vertex AI RAG memory backend) (#543) (8543e9c) auth: emit a PKCE S256 code challenge on the OAuth2 authorization URI (#820) (6db4727) core: export resetLogger from the @google/adk public API (#882) (d4bdbd3) dev: add --min_instances/--max_instances flags to agent_engine deploy (#899) (0dc138c) dev: add Origin validation to the dev server (DNS-rebinding defense in depth) (#557) (f280eba) models: add a Chrome Prompt API BaseLlm for the on-device model (#843) (77b0803) plugins: add ReflectAndRetryToolPlugin and ReflectAndRetryModelPlugin for self-healing error recovery (#631) (1b6ea6d) Port the typed-errors module (errors/) from adk-python (#583) (fab19d0) Rethinking ForwardingArtifactService Interface (Part 2/2) (#510) (cea2dca) Rethinking ForwardingArtifactService Interface and Removing Unnecessary Parameters in ADK JS (Part 1/2) (#508) (77809e3) tools: port the adk-python v0.1.0 retrieval tools (#854) (5ec1c94) Bug Fixes a2a: harden agent-card resolution against SSRF and a silent file read (#832) (1bb1120) a2a: propagate request metadata to agent execution (#752) (#768) (bb2dd8f) agent-registry: declare the synthesised card default modes as media types (#878) (2024ab2) agents: carry request-input resume inputs in their own type (#822) (8664b6e) always emit an array for the OpenAPI argument schema required key (#895) (dc30492) artifacts: list artifacts nested under another artifact (#858) (caa2b4d) auth: redact OAuth2 authorization code from malformed-callback-URI log (#640) (08f4f24) auth: reject an empty credentialKey before storing the credential (#818) (8f792bb) bind one appName in the Vertex AI compaction e2e test (#848) (675c2f3) build: make the web build loadable outside Node (#837) (5b9c306) build: stop compiling the dev CLI entrypoint into dist/cjs (#861) (da9a492) check every function response, not just the first, when resolving which agent to resume (#810) (aee975f) cli: quit adk run on a whitespace-padded exit (#896) (0568990) cli: stop staging an empty node_modules and a zero-byte lockfile into the deploy image (#894) (fd4ac9a) cli: surface model errors and make agent stack traces readable (#625) (9508dee) cli: use GOOGLE_GENAI_USE_ENTERPRISE instead of deprecated GOOGLE_GENAI_USE_VERTEXAI (#897) (6dc4a3d) close two remaining author-blind confirmation-resolution paths (#755) (f9d6d91) context: respect isolation scopes during compaction (#830) (cda8433) core: configure VertexAiSessionService from environment for vertexai:// URIs (#898) (600b756) core: drop the unused @mikro-orm/reflection dependency and pin the DB-driver manifest contract (#814) (6ea9ce5) core: gracefully handle unregistered tool calls (#790) (6be4195) core: keep winston out of the published browser build by splitting the logger by platform (#813) (c2c2ecb) core: load artifacts and MCP resources wh

From the release notes

chore: bump version to v3.20.78 [skip-release] [skip ci]

From the release notes

chore: bump version to v3.20.78 [skip-release] [skip ci]

From the release notes

chore: bump version to v3.20.78 [skip-release] [skip ci]

From the release notes

chore: bump version to v3.20.78 [skip-release] [skip ci]

From the release notes

chore: bump version to v3.20.78 [skip-release] [skip ci]

From the release notes

chore: bump version to v3.20.78 [skip-release] [skip ci]

From the release notes

chore: bump version to v3.20.78 [skip-release] [skip ci]

From the release notes

What's new 🌐 Introducing Rowboat Spaces. Every teammate. Their own agent. One Space. Rowboat becomes a multiplayer personal assistant for work. A Space brings your team’s conversations, files, and whiteboards together, and everyone brings their own Rowboat: their own memory, tools, and model keys. Your personal assistant can now work alongside your teammates and their assistants in the same room. 🤖 Ask in the Space. Your Rowboat takes it from there. Type @rowboat what is missing? and your Rowboat picks it up on your machine, draws on your Brain, email, meetings, and notes, and brings the answer back as “You (via Rowboat)”. Your personal knowledge stays on your machine; the Space holds what you and your agent choose to share. The room gets a 👀 while work is underway and a ✅ when it settles. If your Rowboat needs a decision, permission, or a private conversation, it leaves a ❗ and asks in your own chat. You can also open the agent conversation behind a Rowboat reply to follow the work. 🎨 From a shared whiteboard to something you can ship. Sketch together with live cursors, then ask @rowboat turn this board into a proposal. Agents can draw on the same board and work with the same files as the team. Markdown documents have version history, diffs, and restore; concurrent text edits merge line by line, with conflicts surfaced for resolution. Open documents and presentations in place, or link a discussion to a file and read them side by side. When the proposal is ready, @rowboat implement the proposal can hand it to Claude Code or Codex on your machine, using the Space’s files as context, and bring the result back for review. 💬 The everyday parts of teamwork are here too. Spaces includes threaded conversations, DMs, notes to yourself, reactions, polls, scheduled messages, and mentions. The composer has formatting controls, link previews, attachments, and voice dictation. An Activity view gathers mentions, DMs, replies in followed threads, and reactions on your messages. Unread state follows your account across devices, and notifications take you back to the conversation. On macOS, the Dock icon shows a count for mentions and DMs, or a dot for other unread activity. 🚪 Start a Space and bring your team. Spaces is now available by default. Open Spaces → Create a server, sign in with Google or Microsoft, and use Copy invite link to bring teammates in. A server is your team’s home for its Spaces; you can create or join more than one. For teams that want to run their own infrastructure, Harbor, the server behind Spaces, is open source and self-hostable. It also exposes an MCP server, so other agents can participate using the same tools as Rowboat. See the Spaces guide or explore Spaces on our website. 🪟 Keep your assistant wherever the work is. Move a conversation between the full Assistant page, a right sidebar, and a floating window. Keep several floating chats open, resize and reorder them, or minimize them into tabs along the bottom. The app now ope

From the release notes

New Project-level provider and model access controls. Owners and admins choose which model providers and which models a project may use, from one model management view that also holds provider links. Kortix managed models sit beside your own providers in the same list, a ChatGPT subscription shows its access beside its credentials, and the picker distinguishes "not shown" from "blocked" so a member sees why a model is unavailable. GPT-6 Astra is in the managed catalog. Select kortix/gpt-6-astra with image input, tools, and the supported reasoning efforts. Routing, prices, capability limits, and sandbox fallbacks are all in step. The desktop app always has a way out. The shell has no browser toolbar, so a page without an in-app exit was a dead end. Every such page now has Close, the Go menu has Back (Cmd/Ctrl+[) and Home (Cmd/Ctrl+Shift+H), a renderer crash offers Reload or Go Home instead of an empty window, and the Electron route allowlist now matches the web middleware's exactly. Improved One noun: project. The interface said "workspace" while the SDK, CLI, API, routes, manifest, and docs all said "project". Screen copy now says project in every one of the nine languages, each with its own word rather than a find-and-replace. The sandbox /workspace directory, the manifest's per-agent workspace: boundary, and Slack workspaces keep their names. Create a project in a specific account. The Switch project menu, already grouped by account, gains a "Create a project in {account}" row for each account where you are an owner or admin, including accounts that have no projects yet. The old global link never said which account it would land in. The SSO and SCIM setup wizards work again. Picking a provider did nothing, and Back or the step rail snapped you to the first step with a render loop in the console. Both are fixed, and a browser journey now walks every step of both wizards on every deploy. Tunnel file transfers are verified end to end. Binary files sent through the Computer Tunnel are checked by digest at the destination, an approval for one exact file no longer grants its parent directory, permission decisions are serialized so an approve and a deny cannot both win, and orphaned upload tasks are stopped. Fixed The sandbox model proxy asked upstream for compressed responses and could not decode them. It now requests identity encoding, so a session on a model that answered with zstd no longer fails its first turn. Internal Project snapshot config provider v2 (S3). Sessions can boot from a prebuilt, blob-less snapshot of the project's committed tree with the history hydrated off the boot path, falling back to Git when no snapshot is ready. Every environment gets a private snapshot bucket and the task-role grant; only staging names its bucket, so production sessions keep the Git path until the in-region measurement lands. The project-snapshot task role gets s3:ListBucket, so a missing object is a 404 and not a denied build, and the grant

From the release notes

New Review Center is on for every project. It is no longer behind a feature flag — the routes, registry, contract, web surfaces, CLI, and docs all ship it by default. A public App can recognise you. public used to mean both "anyone with the link may open this" and "nobody is ever recognised". Now an App you share outside your company still greets your own team, shows them the controls a visitor should not see, and records who acted. Improved Creating an account is reachable again, and it lands you in the account you created. The control had no live entry point, and the one path that reached it dropped you back where you started. There is now a "Create an account…" row in the workspace switcher, and creating an account opens its first workspace with that account selected. Connector categories show their true size. The catalogue counted one page of results and headed every category with "· 1". Categories are now grouped over the complete catalogue on the server, and "View all" filters to the set its heading counted. Secret intake forms read better. Field hints written by an agent become real links, every non-form step shares one status notice, and the header no longer runs under the close button. Modals opened over modals stack correctly. A modal opened while another was open could render underneath it; it now takes the layer above, and the switch control meets contrast in both themes. Session hover cards line up. Sessions carrying a Slack, schedule, or shared marker opened their card inset from the sidebar edge; every row now anchors at the same place. Fixed Sending a message no longer makes the transcript jump twice. An idle send moved the view down and then glided it back. The send is now recognised as the working turn, so the view moves once. Internal 4xx denials are logged as warnings, not errors. Roughly 43% of production error-level lines were expected denials — expired tokens, a project-scoped token refused a cross-project read, an agent missing a grant — which buried real faults. Severity now follows the status class; the lines stay queryable and Sentry behaviour is unchanged. A dropped audit batch now says why. The log recorded the whole failing statement plus its bound parameters — IP addresses, user agents, account and project ids — while hiding the SQLSTATE that distinguishes a transient timeout from a permanent constraint violation. It now reports the code and cause, with no statement text and no parameters. A forced test exit raises a workflow annotation instead of a line buried in a 40,000-line CI log. The staging deploy asserts the host-only access cookie from main, matching what staging already checked. What's Changed feat(apps): a public App can recognise you without shutting anyone else out by @markokraemer in #7197 chore(release): staging VERSION → 0.13.15 [skip ci] by @github-actions[bot] in #7211 chore(release): VERSION → 0.13.15 [skip ci] by @github-actions[bot] in #7212 fix(web): stack modals opened over

From the release notes

This release repairs several v1 runtime surfaces that were silently broken since v1.50.0 and removes the deprecated useRenderTool shim from React Native.

From the release notes

Merge pull request #1083 from rowboatlabs/copy-download feat(spaces): add file copy and download menu actions

From the release notes

This release trims the dependency footprint of the Teams and Slack channel packages and fixes Telegram code-block formatting. The headline change is a breaking one for self-hosted Teams users: the Microsoft Agents SDK and Express are now optional peer dependencies.

From the release notes

docs(release): prepare v1.38.9 notes Summary: Generate a bilingual, product-focused draft from merged pull request metadata. Reuse the selected release-bound PR when one is available. Verification: Validate the catalog, citations, bilingual fields, and rendered GitHub release notes before committing. docs(release): restore v1.38.9 upgrade notes after regeneration Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: SivanCola 32437197+SivanCola@users.noreply.github.com

From the release notes

docs(release): prepare v1.38.9 notes Summary: Generate a bilingual, product-focused draft from merged pull request metadata. Reuse the selected release-bound PR when one is available. Verification: Validate the catalog, citations, bilingual fields, and rendered GitHub release notes before committing. docs(release): restore v1.38.9 upgrade notes after regeneration Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: SivanCola 32437197+SivanCola@users.noreply.github.com

From the release notes

docs(release): prepare v1.38.9 notes Summary: Generate a bilingual, product-focused draft from merged pull request metadata. Reuse the selected release-bound PR when one is available. Verification: Validate the catalog, citations, bilingual fields, and rendered GitHub release notes before committing. docs(release): restore v1.38.9 upgrade notes after regeneration Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: SivanCola 32437197+SivanCola@users.noreply.github.com

From the release notes

What's Changed docs(core): fix tools.getInput and tools.proxyExecute examples by Alberto Schiabel (@jkomyno) in #4481 fix(cli): bare --stream, dead Slack slug in help, tools info exit code, unknown listen slug by Nishant Gupta (@2nishantg) in #4344 New Contributors Nishant Gupta (@2nishantg) made their first contribution in #4344 Full Changelog: https://github.com/ComposioHQ/composio/compare/@composio/cli@0.4.2-beta.393...@composio/cli@0.4.2-beta.394

From the release notes

Signed-off-by: omnigent-ci[bot] <294685417+omnigent-ci[bot]@users.noreply.github.com>

From the release notes

Merge pull request #1075 from rowboatlabs/p-sidebar-enhancements feat(renderer): refine sidebar activity and chat access

From the release notes

本版的主线是让屏幕上写着的东西与内核里的事实是同一件：任务面板读内核发布的那份清单而不再从转录里把它猜回来，一次被拒的工具调用带着主机给它的身份而不是一句话，卡片名出内核真正跑的那个能力，能力清单在调用被花掉之前就说出它要的参数。工作台补上四件它做不到的事——会话内查找、代码块高亮、消息改写重发、能力名到卡片。回合多了一个出口：把清单交回用户，而不是被主机推着往下走。界面一侧是一轮以实测为准的整理（阅读尺寸与栏宽、等宽面、汉字下的小型大写、浅色的灰、卡片的高度成本），以及三个早已失灵的门。自 2.15.0 起 43 个提交。 升级路径：无需人工步骤。三平台照旧自更新——Linux 走 .deb，Windows 运行下一个安装器，macOS 替换自身 bundle。 新增 工作台：四件它做不到的事 会话内查找。转录只挂载几十张卡片，所以浏览器自带的查找只看得见屏幕上那些：200 条消息时 DOM 里有 59 张卡，第一条根本找不到。查找改读行本身——saidBy() 对 Item 是穷尽的，新卡片必须回答它携带什么文本。命中用高亮注册表画，而不是包一层标签：那等于重写 React 拥有的 markdown，而那些卡片在每一个增量上重渲染。 代码块高亮。rehype-highlight 像 katex 一样惰性加载，输出的是 class 而非内联颜色——自己写十六进制的高亮器会离开令牌系统、从此不跟随主题。调色板是四个新令牌，注释与标点走 --faint 和 --muted，跟随对比度档位。没有标注语言的围栏不上色：猜一门语法等于把散文画成代码。 一条消息可以改写后重发。由内核已有的两件事组合而成：把对话回卷到那一回合，然后发送。RewindResult 的 TS 镜像少了 conversationOk，于是"动了文件但对话仍在"与"对话被截断"无从分辨——那会把消息发两次。类型补全，这一对进了 wireparity。 调用解析到的能力会到达卡片，仅当名字尚未说出它时：tool:grep 写在 Search 旁边是一件事说两遍，skill:review 不是。 任务清单，与回合的出口 面板画的是清单主人所持的状态。GET /todos 一直在发布那份权威清单——最新一次 todo_write 与其后每一次 complete_step 推进的合并——而窗口从来没调用过它。面板自己从转录里解析 todo_write 的参数重建一份，那是 rebuildTodoState 的第二份实现，并在两处漂开：它不重放 complete_step，所以每一次签收都丢了；它不看回执，所以一份被内核拒绝的清单成了计划。每一次重建——挂载、面板重绑、流中断——都把面板重置回"模型最后写下的样子"。已签收的条目渲染成当前项、此后任何调用都推不动它，这就是"待办面板清不掉"从外面看的样子。 任务清单根本不可能从转录导出：推进不是 todo_write 调用，而被拒绝的写入是。所以 fromHistory 不再尝试，窗口去问；回合结束再读一次，因为有些移动没有事件可说（一次取消会让内核重建自己的状态）；读取失败时面板保持原样而不是清空。 投影在状态移动时就欠下，而不只是 id 变了的时候。withTodoIdentityTail 只要每个步骤 id 还能在视图里读到就认为不欠——而 id 永远读得到，它们就在模型自己的 todo_write 参数里。底下的状态不会留在原地：complete_step 在主机侧推进那份权威清单而模型什么也没写。于是一份主机已经翻过页的清单还按模型最后写下的样子读着，下一次签收点的是主机已不认为是当前的那一项——complete_step 拒绝它，代价是满上下文下的一整轮，且学不到任何关于任务的东西。23 个录制会话里 complete_step 在 43 次调用中失败 16 次，这一类是其中最大的一组。 一个回合可以把清单交回用户，而不是被驱动。带着未完项停下来此前一律读作"活没干完"，continueUntilReady 会拿主机写的一句话把它推上至多八轮。对模型能独自推进的步骤这是对的；对下一步输入在用户手上的步骤是错的——呈现一项并等待回复，与提前退出无从分辨，于是一份每回合走一项的清单根本没法走：回合停下、主机重启它、整张清单在一轮里全吐出来。 await_user 成为除 ask 与 conclude_blocked 之外第三种由模型决定的收尾，它点名等待中的那一项和用户需要提供什么，清单原样保留，就绪判定豁免未完项，续跑循环不运行。守卫是结构性的：必须有一份未完清单、step_id 必须点中其上未完成的一项、必须有可交互的用户——无头运行拒绝它并被指向 conclude_blocked。 工具表面与能力 一个工具在 schema 里，就是此刻可调用的。此前存在三种状态而模型只分得清两种：在 schema 里且可调用、注册了但藏在 use_capability 后面、以及在 schema 里却会被拒——最后一种不是任何人的本意。update_goal 就住在第三种里：534 个真实会话中它被调用 121 次，拒绝 116 次。 把工具钉住不动的那条理由被量掉了（benchmarks/tool-surface-cache）。"改 schema 会把缓存前缀搅浑"对任何位置的改动为真、对末尾的追加为假，而没有任何东西测过哪一条适用。实测：同一个工具在头部要 896 个缓存 token，在中部 384，追加在尾部不花任何东西；一个尾部工具在"有/无"之间来回三次，缓存读数稳定在 1920 与 2048 之间、没有一次未命中尖峰——因为缺席的那份表面是在场那份的字节前缀，一条缓存同时服务两者。于是 ProviderSchemas 分两段排序：稳定工具在前，本回合准入的上下文工具在后。 能力清单说出一次调用要被held到的参数。藏在 schema 之外的能力靠列目录再按 id 调用来够到，而清单只有 id 没有参数，第一次调用只能是猜。534 个会话里 12 次调用对 3 次 inspect，5 次带回"它需要……"。契约错误在调用被花掉之后才打印的两张表，现在在调用之前就在清单里：必填字段名，和每一个声明过的属性。从目标自己的 schema 经注册表读出——不是注册表工具、或 schema 没声明属性的条目什么也不说，而不是报一张空表让读者当成"它不要参数"。 卡片名出内核真正跑的那个东西。内核注册的 54 个工具里有 22 个在两张表里都没有条目，于是一次经 use_capability 的调用渲染成它自己 title-case 过的 id，顶着那个"折角的纸"标记。一次委派、一次记忆写入、一次安装全都读作文件读取，而类别颜色——那一行唯一的警示——是中性的那个。install_source 是其中最糟的一个：它不是只读的，它安装一个 MCP 服务器，而它根本没有身份。兜底标记改为中性圆点："·" 是一份文档，一个没有条目的名字此前在声称自己读过文件。 重新打开的转录说得出代理够到的是哪个能力。实时转录读作 "Remember · 2026-08-

From the release notes

@mastra/turso@0.1.6

From the release notes

@mastra/upstash@1.4.6

From the release notes

@mastra/valkey@0.2.3

From the release notes

@mastra/valkey-streams@0.5.2

From the release notes

@mastra/vercel@1.6.0

From the release notes

@mastra/voice-google-gemini-live@0.14.10

From the release notes

@mastra/voice-openai-realtime@0.14.0

From the release notes

@mastra/voice-xai-realtime@0.2.10

From the release notes

mastra@1.30.0

From the release notes

mastracode@0.40.0

From the release notes

What's Changed refactor: extract shared hardened gh runner for all gh spawn paths (#342) by @bolichen97 in #2407 feat(security): operator keystone extension for OAuth consent-endpoint allowlist (#1341) by @bolichen97 in #2402 fix: redact model-authored tool titles on all deny surfaces (#2274) by @bolichen97 in #2389 feat(chat): message-level pinning with pinned messages panel by @RohanK6 in #1676 ci(review): expand Opus to correctness blocking + sync fork prompt by @iamwhatever in #2379 feat(mcp-pool): Toggle all switch for poolable MCP servers by @chenmingwei23 in #2433 fix(knowledge): make the three auto-ingest paths opt-in by @bolichen97 in #2448 fix(dev-fleet): report slow actions as indeterminate, not as a fake percentage by @chenmingwei23 in #2439 test(dashboard): make chat-pins transient-I/O test portable on Windows by @iamwhatever in #2454 feat(voice): add a push-to-talk key for voice input by @CrysisDeu in #1608 feat(watchdog): detect a stalled turn from outside the turn's read loop by @chenmingwei23 in #2466 feat: receive images, voice and files on the Weixin channel by @chenmingwei23 in #2444 fix(dev-fleet): refuse the sync when pip cannot reinstall into the live venv by @chenmingwei23 in #2445 fix(prepare-pr): re-mirror the local review gate onto the current CI contract by @iamwhatever in #2456 feat(voice): show info banner + mic-click modal on remote instances by @cixuuz in #2471 refactor: centralize title hydration by @JoernZheng in #2467 fix(test_wait_tool_early_end): scope the wait-tool fake clock to the wait loop thread by @tlobinger in #2460 fix(terminal): complete subcommands for tools the user installed by @dwu96 in #2429 feat: first-party fixed-argv carve-out in sandbox fail-close (#1563) by @bolichen97 in #2428 feat: verify internal-API session claims via unix-socket peer creds (#302) by @bolichen97 in #2424 fix: re-vet cron jobs at fire time; move SEL key to trust/ (#1881) by @bolichen97 in #2421 build: target Node 24 LTS with a single 22+ floor everywhere (#1070) by @bolichen97 in #2415 feat: pin tunnel sessions to daemon-verified tailnet peer (#1762) by @bolichen97 in #2411 fix(appstore): compute app trust fields server-side (#580) by @bolichen97 in #2408 fix: detect and audit tier downgrades in nested-sandbox passthrough (#691) by @bolichen97 in #2406 fix: per-root hardlink-scan budgets with loud truncation (#646) by @bolichen97 in #2387 fix: make revocation generation cover refresh tokens (#2028) by @bolichen97 in #2388 ci: gate vendored _vendor tree behind a sha256 manifest (#492) by @bolichen97 in #2392 fix(dashboard): restore isort-clean import order in the dashboard auth module by @iamwhatever in #2479 feat(instances): carry the kiro-cli context so a sent session resumes by @iamwhatever in #2260 revert: withdraw multi-account Telegram until a bot is governable (#2203) by @chenmingwei23 in #2476 docs(autosde): forbid new work on the gateway boot path by @CrysisDeu in #2472 docs: revis

From the release notes

0.7.0 insider 1

From the release notes

Signed-off-by: omnigent-ci[bot] <294685417+omnigent-ci[bot]@users.noreply.github.com>

From the release notes

v3.42.0 — Swarm, Memory & Security Reliability Batch This release lands 14 fixes produced by Ruflo's own nightly "Dream Cycle" research pipeline — each one found a real bug or gap in the codebase, built a fix, proved it with before/after tests, and (in several cases) survived an independent adversarial review before merging. Nothing here is speculative: every change below has a passing regression test that failed before the fix and passes after it. What's new for you 🐝 Swarm consensus actually uses trust weights now When multiple agents vote on a decision, Ruflo has long computed a "trust weight" for each voter — but that number was silently discarded before it ever reached the vote count. Every strategy (Raft, Byzantine, gossip) was voting as if every agent were equally trustworthy. Now the weight is actually used, so a vote from an agent with a strong track record counts for more than a vote from an untested one. 🔍 Vector search is faster and more accurate HNSW product-quantization (a memory-compression technique for embeddings) was built and tested but never actually wired into the search path — searches were running on meaningless data. It's now dispatched correctly, more than doubling search accuracy (recall) in our benchmark. MMR re-ranking (used to diversify search results so you don't get 10 near-duplicate hits) was re-tokenizing every candidate on every pass, even when it didn't need to. It's now ~6x faster with byte-identical results. Memory deduplication used to only catch exact byte-for-byte duplicates. It now also catches near-duplicates (paraphrases, reformatted text) using the embeddings Ruflo already computes — no extra cost, same-quality matching. Fixed a bug where a memory pattern's retrieval score (how well it matched your query) was overwriting its learned reliability (how trustworthy that pattern has proven to be over time) — which meant the "only trust high-confidence patterns" safety check was silently checking the wrong number. 🔐 Security hardening MCP tool governance is now enforceable. Ruflo has shipped a governance policy file (audit logging, per-session call limits) for a while, but nothing in the running server ever actually read it. It's now wired in as an opt-in flag (RUFLO_MCP_ENFORCE_POLICY=1) — off by default, so nothing changes unless you turn it on. The call-limit for that policy now resets properly. It used to be a lifetime cap that, once hit, locked a session out permanently until restart. It's now a rolling time window, matching how every other rate limiter (Cloudflare's, etc.) does it. Hive-mind voting can no longer be Sybil-attacked. Previously, any caller could cast a vote under a made-up agent name and it would count toward quorum — meaning one caller could out-vote everyone else with fabricated identities. Votes are now checked against the actual roster of agents that joined the hive. Joining or leaving a hive-mind now requires proof you're allowed to. This closes the gap the vote-fix abov

From the release notes

Align the package version with CopilotKit 1.71.2. The API and implementation are unchanged from 0.1.0.

From the release notes

Align the package version with CopilotKit 1.71.2. The API and implementation are unchanged from 0.1.0.

From the release notes

Initial release of @copilotkit/intelligence-mastra: native processor, skill tools, and agent wrapper for verified learned skill delivery, including resumed tool execution. #7129

From the release notes

Initial release of @copilotkit/intelligence-langgraph: verified learned skill delivery for native LangGraph agents, with skill tools and invocation handling. #7072 #7129

From the release notes

This release adds native Intelligence runtime support to @copilotkit/runtime and improves how the Inspector handles ephemeral threads. Features Native Intelligence runtimes and shared conformance (#6967): The TypeScript runtime now consumes the released @ag-ui/mcp-apps-middleware@^0.1.0 and runs against the same cross-language conformance suite as CopilotKit's native runtimes. This release also includes several runtime hardening fixes: Enforces MCP tool visibility filtering and unambiguous account/proxy selection, rejecting proxy requests outside the selected agent scope. Preserves identity and request-boundary contracts — SDK update precedence and mounted routing are retained, browser identity aliases can no longer override the authenticated user, and stop requests are cloned before application authentication consumes the body. Validates AG-UI tool arguments against their JSON schemas, keeping relaxed structured output limited to the A2UI tool. Preserves safe failure reporting and completion analytics — failures are reported without private diagnostic payloads, and completion counts are retained for streams containing RUN_ERROR. Advertises and consumes the January MCP Apps MIME type correction (text/html;profile=mcp-app). Fixes Preserve the ephemeral Threads upgrade path in the Inspector (#7098): The Inspector no longer treats any Threads list endpoint as durable Threads support. OSS apps using an in-memory agent runner can now inspect ephemeral conversations while still seeing a clear path to durable Threads: The full Rich Threads setup CTA is shown when Threads are unavailable, or when Intelligence is off and no local threads exist. When the first ephemeral thread appears, the Inspector switches to the thread list with a "Keep your threads" banner explaining that history can disappear on restart. "Make them permanent" opens the full setup view inline, with a sticky "Back to your threads" link to return to local history. The banner and setup override are removed once Intelligence becomes available. Remove optional feedback prompt copy from Inspector onboarding (#7099): Dropped an unnecessary optional diagnostic-feedback instruction from the shared feature setup prompt used by Threads and Learning, while retaining the instruction not to reveal credentials.

From the release notes

What's Changed perf(cli): list connected accounts once per execute by Daksh (@sudodaksh) in #4475 Full Changelog: https://github.com/ComposioHQ/composio/compare/@composio/cli@0.4.2-beta.392...@composio/cli@0.4.2-beta.393

From the release notes

What's Changed docs: explain token custody architecture and deployment options by Brendan O'Leary (@olearycrew) in #4446 fix(cli): restore automatic plugin setup on install by Kshitij Jhunjhunwala (@KJ-11) in #4485 Full Changelog: https://github.com/ComposioHQ/composio/compare/@composio/cli@0.4.2-beta.391...@composio/cli@0.4.2-beta.392

From the release notes

What's Changed perf(cli): move the compiler and tokenizer out of the executable by Daksh (@sudodaksh) in #4469 Full Changelog: https://github.com/ComposioHQ/composio/compare/@composio/cli@0.4.2-beta.390...@composio/cli@0.4.2-beta.391

From the release notes

Fix agentdb_pattern-store/agentdb_pattern-search now surface degraded: true whenever the response comes from anything other than a healthy reasoningBank controller (the bridge-fallback and memory-store-fallback paths alike), instead of only in a note string a caller has no structural reason to check. Fixes #3288. Includes a follow-up correction (b9ca64f80) after review caught that the initial fix only handled the case where the controller registry is entirely absent, missing the more common case where the registry exists but ReasoningBank itself is unusable (controller: 'bridge-fallback') — the degraded flag is now applied at the result.controller === 'reasoningBank' boundary in both handlers rather than per fallback-label, so it can't be reopened by a future fallback label. See #3323 for the full discussion. This does not address why vectorBackend/reasoningBank fail to activate in a given deployment — that's tracked separately in #1228 (ADR-053) and #2296.

From the release notes

Release v4.7.8

From the release notes

What's new Three harnesses from the YC Paper Club talk are in. Prime Agent (Prime Intellect, 95.5% on ARC-AGI-3 with Opus 5), QM (Y Combinator's multiplayer harness for work), and OpenJarvis (Stanford's local-first stack). 167 harnesses, stars captured 2026-09-13. Three new decision guides (16 total). Why the harness matters more than the model: the measurements (ARC-AGI-3 30% to 95.5% on the same weights, SWE-bench Pro 23% to 52%, Cursor 46% to 80%, Harness-Bench rank correlation -0.05), a table of twenty labs, papers, and practitioners with checked links, and what the claim does not mean, from the YC Paper Club session of September 7. Managed vs self-hosted always-on agents: Grok Bot, Claude Managed Agents, QM, OpenClaw, Hermes, OpenJarvis. Who owns the computer, who pays for idle time. The best AI agent harnesses in 2026, ranked by category: the top three per category from the live data, regenerated on every weekly refresh. README and machine-readable surfaces. The intro now carries the long-horizon numbers, the two eras of harnesses, and the pairing rule. Two new FAQ entries (the thesis, Grok Bot) flow to the README, llms.txt, harnesses.json, and the site's FAQPage. llms.txt states the thesis in its blurb; the site adds llms-full.txt (llms.txt plus every guide). Site. Every guide page carries TechArticle JSON-LD with real published and modified dates, breadcrumbs on all project, category, and guide pages, and a per-URL sitemap date. Guide links now resolve on the site (they previously pointed at .md paths that 404 under /compare/). MCP. compare prefers a guide that names the compared projects in its title. The 0.5.2 package on PyPI already serves the new projects and guides from main; the tie-break ships in the next package release.

From the release notes

Security fix Federation event identity: fetchRecent/fetchManyOn/fetchChannel now apply the signature-verified id/pubkey/created_at after the publisher's own message content, instead of before — a publisher could otherwise put those keys in their own content and overwrite their verified identity downstream, which the swarm claims ledger (reduceClaims) trusts for release/handoff authorization. Same fix applied to x_federation_channel_read in @claude-flow/cli, which additionally now verifies event signatures at all (reqEvents() previously trusted unverified events). The live ruflo-x-gateway production service was redeployed with this fix ahead of this release. Windows hook argument escaping: ruflo-hook.cjs's Windows shell:true npm-shim invocation path now escapes every argument (cmd.exe metacharacter neutralization), so a hook-derived value can no longer be reinterpreted as a separate command or redirection. See #3322 for the full technical writeup and test coverage.

From the release notes

Hermes Agent v0.21.3 (v2026.9.14) Release Date: September 14, 2026 Patch release. This tag rolls up the ~338 PRs merged since v0.21.2 into a stable tagged release for downstream consumers (Docker images, Hermes Cloud, hosted deployments). It exists so the remote-gateway sign-in fixes below reach Cloud agents, which auto-update to the newest release tag. What this patch ships for remote Desktop / Cloud users Remote dashboard sessions no longer expire on refresh bursts (#110061, fixes #55712; salvage #71548 @Doud-FR, #55717 @liuhao1024). Both refresh paths on the gateway (cookie gate and the desktop's native bearer route) now coalesce concurrent requests carrying the same rotating refresh token, so a Desktop wake burst can no longer replay an already-rotated token into the Portal's reuse detection and revoke the whole session. Refresh also runs off the event loop, so a slow identity provider no longer freezes /api/status. Pairs with Portal-side NousResearch/hermes-portal#1209 (sliding 30-day idle horizon, 5-minute rotated-token grace). Also requested for this tag Long-lived processes stop leaking duplicate state.db writer handles (#110934, fixes #100896 #103339; salvage #107974 @kshitijk4poor, mapping @Rroven): gateway, dashboard/Desktop backend, ACP and CLI readers attach read-only and in-process writers share the registry handle, so the N live SessionDB handles precursor stops firing on a healthy topology. About this release Measured at commit 9b419a2d3c2657c192008e732149d61170b32c01, the window since v0.21.2 contains 1,036 non-merge commits across 2,642 changed files (+131,690 / −37,096) and 338 merged PRs. Also in the window, undocumented here on purpose: server→client JSON-RPC requests and a Pydantic wire-contract registry with generated TS/OpenRPC for the TUI/Desktop gateway (#110521, #110522); reasoning-effort selection on every model picker, a composer pill and per-auxiliary control in Desktop; OpenRouter OAuth PKCE login; HEIF/HEIC/AVIF image decoding; the Honcho peer-model setup rework; MCP OAuth refresh tokens bound to their issuer; a daily MCP re-auth nudge in Desktop; Wan 3.0, Kling 3.0 / Kling Image v3, MiniMax H3 Max Turbo, Gemini Omni Flash 1.1 and Meta Muse in the FAL catalogs; Slack pasted tables and the Agent Sessions API; multiplexed-profile isolation and gateway-liveness fixes; and the state.db WAL refusal on cross-VM filesystems. Full curated release notes for this window ship with v0.22.0, which will document everything from v0.21.0 onward — highlights, feature areas, and complete contributor credits. Nothing in this window is skipped. Updating hermes update (git installs), or re-run the installer one-liner. Docker / Hermes Cloud: images build from this tag (nousresearch/hermes-agent:v2026.9.14). Full changelog: v2026.9.11...v2026.9.14

From the release notes

What's Changed perf(cli): defer the TypeScript compiler and generation pipeline by Daksh (@sudodaksh) in #4468 Full Changelog: https://github.com/ComposioHQ/composio/compare/@composio/cli@0.4.2-beta.389...@composio/cli@0.4.2-beta.390

From the release notes

Two new homes for Ponytail. The headline: native Cursor support through hooks.json, with a scripts/cursor-hooks.js install that merges into your existing Cursor hooks file and leaves everything else in it alone. Also new: a Grok Build skills adapter, VS Code Copilot is finally detected correctly (no more Claude Code statusline nudge there), and the Claude.ai marketplace validator accepts the plugin again. What's Changed fix: drop commandWindows from hooks.json for Claude.ai marketplace validation (#593) by @prayag0one4 in #601 fix: detect VS Code Copilot via CLAUDE_PLUGIN_ROOT fallback (#528) by @krishrathi1 in #579 feat: add Grok Build native skills adapter (revive #561) by @p-clements in #661 docs: add Trendshift badge to READMEs by @DietrichGebert in #801 docs: add Trendshift monthly ranking badge by @DietrichGebert in #802 docs: put daily, weekly and monthly Trendshift badges in one row by @DietrichGebert in #803 docs: Built with Ponytail, Retriever by @DietrichGebert in #830 feat: add native Cursor hooks via hooks.json (#817) by @DietrichGebert in #869 New Contributors @prayag0one4 made their first contribution in #601 @p-clements made their first contribution in #661 Full Changelog: v4.9.0...v4.10.0

From the release notes

What's Changed fix(cli): stop tiktoken special-token literals from failing execute by Daksh (@sudodaksh) in #4464 Full Changelog: https://github.com/ComposioHQ/composio/compare/@composio/cli@0.4.2-beta.388...@composio/cli@0.4.2-beta.389

From the release notes

What's Changed perf(cli): cut 221ms and 44MB RSS off every CLI invocation by Daksh (@sudodaksh) in #4463 Full Changelog: https://github.com/ComposioHQ/composio/compare/@composio/cli@0.4.2-beta.387...@composio/cli@0.4.2-beta.388

From the release notes

🌐 English | 中文 This is a refinement release for the 2.1.8 multi-Agent version, building on it with a series of improvements and fixes across multi-Agent collaboration, model capabilities, and channel integration. 🤝 Multi-Agent Experience Further polish around using multi-Agent teams: Multi-Agent session selection: the in-session multi-Agent selector adds a lead-Agent marker that clearly separates the lead Agent from members, while simplifying the view in single-Agent setups. Collaboration fix: fixed failures when delegating tasks from non-lead Agents in multi-Agent group chats, now correctly identifying member Agents. Member Agent image rendering: fixed images generated by non-default Agents failing to render on the Web console and desktop client. Group-chat entry: the desktop Team page adds a group-chat entry, so you can start a group session directly from an Agent team. 📡 Channel Integration Fixes Fixed key issues along the console's channel start/stop path for more stable channel integration: Desktop channel integration fix: fixed channels failing to start in the desktop client. Channel stop fix: fixed a lost app_module lookup during the channel stop flow that caused disconnect errors (a regression introduced by #3141). QQ channel reconnection: fixed the QQ channel not reconnecting after a silent WebSocket disconnect; it now reconnects automatically. 🤖 Model Capabilities Model list configuration: model providers can now configure a model list, setting the model name, model type, context window, max output, and more. Thanks @alanyz106 (#3146) Multiple fallback models: when the primary model fails, CowAgent now tries an ordered fallback chain in turn, replacing the previous single fallback model. Thanks @alanyz106 (#3125) Fallback credential fix: fixed credential resolution during fallback, which now resolves credentials by the provider actually routed to rather than the global configuration. Thanks @alanyz106 (#3142) New image models: added gpt-image-2.5-flare and gpt-image-2.5-sunburst. Thanks @cowagent (#3140) 🛠 Improvements & Fixes Config file tolerance: tolerate a UTF-8 BOM in config.json, and stop the pre-login 401 polling to avoid invalid requests before sign-in. Image generation fixes: fixed Baidu Ernie IMAGE_CREATE crashing when unsupported, now returning an error message instead (Thanks @c020627 #3144). fixed Midjourney reply retries not recursing into the send logic correctly (Thanks @c020627 #3151). Voice fix: fixed silk voice transcoding by first decoding silk into a standalone wav and then re-encoding to mp3, avoiding overwriting the source file (Thanks @c020627 #3149). Media URL classification: determine media type from the resolved URL path, fixing misclassified media links (Thanks @c020627 #3150). Baidu Translate: fixed Baidu Translate not reporting the API error after all retries failed (Thanks @c020627 #3147). Background command detection: restrict run_in_background to long-running processes, preventi

From the release notes

v2026.09.14 — 三語正式版 / Trilingual release 繁體中文 reader-ux 三語 README 移除動畫 banner 下方重複的靜態圖與文件站捷徑，讓首頁入口更乾淨。動畫停止時仍保留完整構圖；文件站的減少動態、無 JavaScript 靜態 PNG 與三語 PDF fallback 都繼續保留。 來源 1 · 來源 2 maintenance 修正 Stage 6 五組範例的三語收合區塊，以及 CLI 工具表與 Stage 3 摘要中的行內程式碼：粗體、清單、行內程式碼與連結現在會由 MkDocs 正確解析，不再把 **Context** 或反引號這類 Markdown 記號直接顯示給讀者。新增 post-build gate，會檢查實際 HTML，而不是只看 .md 原始碼；程式碼範例中的記號仍被允許。 來源 1 learning-content Stage 7 改為「Agent 上線工程：可測、可看、可停、可恢復」。同一個查資料案例依序解釋 Harness、Loop、Graph、Eval、Observability、Approval、Checkpoint、Resume 與 Recovery；完整 Eval Case 明確包含 input、初始狀態、成功條件、禁止行為、選用參考答案、評分方法與案例資訊，避免把 Golden Set 誤解成只有輸入。 來源 1 learning-content Stage 7.5 改成「有證據才增加複雜度」的進階選讀，不再重教 Stage 7 基礎。可見核心詞表先用白話說明 Evaluator–Optimizer／Agent-as-Judge、Failure Injection／Chaos Eval、Autonomy Gradients／Trust Layers 與 Model–Harness Fit，再用兩組三語亮色圖協助讀者依失敗證據選擇最小做法。 來源 1 简体中文 reader-ux 三语 README 移除动画 banner 下方重复的静态图与文档站快捷入口，让首页入口更干净。动画停止时仍保留完整构图；文档站的减少动态、无 JavaScript 静态 PNG 与三语 PDF fallback 都继续保留。 來源 1 · 來源 2 maintenance 修正 Stage 6 五组示例的三语折叠区块，以及 CLI 工具表和 Stage 3 摘要中的行内代码：粗体、列表、行内代码与链接现在会由 MkDocs 正确解析，不再把 **Context** 或反引号这类 Markdown 标记直接显示给读者。新增 post-build gate，会检查实际 HTML，而不是只看 .md 源码；代码示例中的标记仍被允许。 來源 1 learning-content Stage 7 改为“Agent 上线工程：可测、可看、可停、可恢复”。同一个资料查询案例依次解释 Harness、Loop、Graph、Eval、Observability、Approval、Checkpoint、Resume 与 Recovery；完整 Eval Case 明确包含 input、初始状态、成功条件、禁止行为、可选参考答案、评分方法与案例信息，避免把 Golden Set 误解为只有输入。 來源 1 learning-content Stage 7.5 改成“有证据才增加复杂度”的进阶选读，不再重复讲解 Stage 7 基础。可见核心词表先用白话说明 Evaluator–Optimizer／Agent-as-Judge、Failure Injection／Chaos Eval、Autonomy Gradients／Trust Layers 与 Model–Harness Fit，再用两组三语亮色图帮助读者根据失败证据选择最小做法。 來源 1 English reader-ux The three READMEs remove the duplicate static-image and documentation-site shortcuts below the animated banner for a cleaner entry point. The complete composition remains visible when animation stops, while reduced-motion, no-JavaScript PNG, and trilingual PDF fallbacks remain available. source 1 · source 2 maintenance Fixed the trilingual collapsible sections in five Stage 6 examples, plus inline code in the CLI tool table and Stage 3 summaries. MkDocs now renders emphasis, lists, inline code, and links instead of showing Markdown markers such as **Context** or backticks to readers. A new post-build gate checks the actual HTML rather than source .md alone, while still allowing markers inside code examples. source 1 learning-content Stage 7 is now “Production Engineering for Agents: Testable, Observable, Stoppable, and Recoverable.” One research-assistant example progressively explains Harness, Loop, Graph, Eval, Observability, Approval, Checkpoint, Resume, and Recovery. A complete Eval Case explicitly includes input, initial state, success criteria, forbidden actions, an optional reference answer, grading method, and case metadata, so a Golden Set is not mistaken for inputs alone. source 1 learning-content Stage 7.5 is now an advanced reading map that adds complexity only when evidence justifies it, without reteaching Sta

From the release notes

Signed-off-by: omnigent-ci[bot] <294685417+omnigent-ci[bot]@users.noreply.github.com>

From the release notes

Changelog Features d60b5ae feat: add configurable password history check e1aa116 feat: add organization-level defaultTokenFields for new applications d6f2cb6 feat: show device, IP and last active time for each session e7f2799 feat: support app-level dark logo fbacaea feat: support group-level subscriptions ed417c1 feat: support invite-only pricing Bug fixes 3eb9231 fix(web): keep draft rows in key-value tables 0920d5d fix: carry the OAuth provider's email_verified through to the user (#5828) 251c76b fix: don't mark every group as having children in GetGroups() 6d6587d fix: keep CORS headers for concrete origins in proxy 991a5f1 fix: keep MFA providers in masked application so users can set up RADIUS MFA 0123751 fix: let RFC 7592 DCR endpoints bypass AutoSigninFilter 145f4c8 fix: only expire the logging-out user's tokens on session logout 5dd3de0 fix: record syncer errors instead of panicking at startup 4054ad5 fix: skip CORS headers when request has no Origin

From the release notes

Summary Removed unnecessary beartype hooks from three package imports. This prevents unwanted import-time instrumentation and dependency coupling. based on #15258

From the release notes

v1.70.3 Install npm install @copilotkit/react-core@1.70.3 @copilotkit/react-ui@1.70.3 @copilotkit/runtime@1.70.3 Changes Remove the inspectorLearning option from Fetch, Express, and Hono handlers. Inspector Learning now requires an Intelligence runtime with a configured Learning container, not debug mode or a separate opt-in. Remove this option from existing handler configurations. (#6957) React and A2UI Provider-level humanInTheLoop tools now wait for respond, keep parallel calls independent, and report an error when aborted instead of an empty success. (#6851) Custom useDefaultRenderTool renderers can return null to hide selected tool calls without a type error. (#6942) Add development-only warnings for unmatched tool-call renderers and A2UI surfaces that never render or cannot resolve their root component. React and web-component renderers now consistently prefer nested surface IDs. (#6802) Runtime and Inspector Preserve unread request payloads in the Express bridge when body-parser sets an empty body but skips parsing, including multipart and plain-text requests. (#6490) Home and launcher Learning status now comes from Learning's own endpoint rather than Memory availability. Enabled launcher toggles now appear green. (#6957) Documentation Correct the v1 useCopilotChat source documentation: appendMessage is the public programmatic-send method; sendMessage is internal. Include the v2 migration path. (#6940) Document frontend-tool setup requirements for nine integrations, including Built-in Agent, Agno, Deep Agents, and Strands. (#6874) Add Microsoft Agent Framework guidance for backend approval interrupts, alongside the tool-based approval workflow. (#6922) Add AWS AgentCore troubleshooting for HTTP 401: Missing Authentication Token, covering runtime credentials, expired tokens, invocation URLs, and request routing. (#6920) Full changelog

From the release notes

v2026.09.13 — 三語正式版 / Trilingual release 繁體中文 maintenance 發布前完整連結掃描確認 YIELD INTELLIGENCE 的 hosted MCP、替代路徑與 pricing 入口都回傳 404。三語精選目錄已移除這筆目前無法使用的服務與 repository snapshot，不以 README 仍寫著「live」取代實際端點證據。 來源 1 reader-ux 首頁下方的角色分流圖加入圖示動畫試版：只讓五個代表性角色圖示依序輕動，並支援停止／播放、減少動態與靜態 PNG。這組圖沿用三張舊原圖，因此圖中文字尚未完全對齊；英文圖的 Analysts 也比正文的 Knowledge workers 範圍窄。正式校正前，請以圖片下方保持展開的角色表為準。 來源 1 governance 貢獻規則改成繁中先定稿、英文與簡中在同一個 PR 同步。三語的概念、URL、數字、評分與安全限制必須一致；新增資源不再以會變動的 GitHub stars 當門檻，改用來源、維護狀態與教學價值判斷。 來源 1 learning-content Stage 7 新增可見的 Eval 基礎：Golden Set、Reference Set、Trial、Grader、Baseline、Regression 與 Holdout Set 都先用白話解釋，再保留正式名稱。三語同版圖把「定義成功、建立案例、重複測試、比較基準、人工複查」串成一個證據迴圈。 來源 1 examples Stage 7 的 Eval 範例改用 8 題版本化 JSON，分成 5 題開發案例與 3 題凍結 holdout。Ollama 與 Anthropic 共用同一個 runner，可選多次 trial、保存報告並和 baseline 比較；資料格式不相容或空輸出會直接失敗，離線測試不讀 API key。 來源 1 简体中文 maintenance 发布前完整链接扫描确认 YIELD INTELLIGENCE 的 hosted MCP、替代路径与 pricing 入口都返回 404。三语精选目录已移除这项目前无法使用的服务与 repository snapshot，不以 README 仍写着“live”取代实际端点证据。 來源 1 reader-ux 首页下方的角色分流图加入图标动画试版：只让五个代表性角色图标依次轻动，并支持停止／播放、减少动态与静态 PNG。这组图沿用三张旧原图，因此图中文字尚未完全对齐；英文图的 Analysts 也比正文的 Knowledge workers 范围窄。正式校正前，请以图片下方保持展开的角色表为准。 來源 1 governance 贡献规则改为繁中先定稿、英文与简中在同一个 PR 同步。三语的概念、URL、数字、评分与安全限制必须一致；新增资源不再以会变动的 GitHub stars 作为门槛，而是根据来源、维护状态与教学价值判断。 來源 1 learning-content Stage 7 新增可见的 Eval 基础：Golden Set、Reference Set、Trial、Grader、Baseline、Regression 与 Holdout Set 都先用白话解释，再保留正式名称。三语同版图把“定义成功、建立案例、重复测试、比较基准、人工复查”串成一个证据循环。 來源 1 examples Stage 7 的 Eval 范例改用 8 题版本化 JSON，分为 5 题开发案例与 3 题冻结 holdout。Ollama 与 Anthropic 共用同一个 runner，可选择多次 trial、保存报告并与 baseline 比较；数据格式不兼容或空输出会直接失败，离线测试不读取 API key。 來源 1 English maintenance The full pre-release link scan confirmed that YIELD INTELLIGENCE's hosted MCP endpoint, alternate route, and pricing endpoint all return 404. The unavailable service and its repository snapshot were removed from all three curated catalogs; a README that still says “live” does not outweigh direct endpoint evidence. source 1 reader-ux The role map below the homepage now has a pilot icon animation. Only the five representative role icons move in sequence, with stop/play controls, reduced-motion support, and static PNGs. Because it reuses three older source images, wording inside the images is not yet fully aligned; the English image's “Analysts” label is also narrower than “Knowledge workers” in the text. Until the artwork is corrected, use the always-visible role table below it as the canonical guide. source 1 governance Contribution rules now finalize Traditional Chinese first and synchronize English and Simplified Chinese in the same PR. Concepts, URLs, numbers, ratings, and safety limits must match across locales; new resources are judged by sources, maintenance, and teaching value instead of changing GitHub star counts. source 1 learning-content Stage 7 now teaches visible Eval foundations. Golden Set, Reference Set, Trial, Grader, Baseline, Regression, and Holdout Set are explained in plain language before retaining their formal names

From the release notes

The PageIndex SDK, local or cloud — vectorless, reasoning-based RAG, end to end. Much faster indexing — the PageIndex Flash engine gets the tree from layout stats: no LLM involved for the structure generation itself, LLMs only write the node summaries, and tree expansion proposes a wave of nodes concurrently. client = PageIndexClient() client.submit_document("report.pdf") client.chat("What does the report conclude?") Index, to chat, to agent integration, one client. Local mode needs no server, no vector DB, no PageIndex API key. Highlights Flash engine: the local default (mode="standard" keeps the classic LLM pipeline). Embedded bookmarks are consumed when trustworthy, and tree optimization is on by default — optimize="merge" for the deterministic LLM-free pass, "full" (default) adds LLM expand, which runs a wave of nodes concurrently instead of one round-trip at a time. One complete surface, local and cloud: PageIndexLocalClient(storage_path=...) is the same client as cloud — submit, tree, page content, chat, and agent tools all present in both modes, so code moves between them unchanged. Cloud documents, your own model: api_key decides where your documents live; a configured chat model decides who answers — and the two combine. PageIndexClient(api_key="pi-...", chat_model="openai/gpt-5.2") runs the same in-process document-QA engine over the live cloud tool set. Page content flows through your process to your provider on your credentials; doc_id targets at the prompt level; enable_citations stays with the managed chat. Agent integration: the cloud MCP tool contract, in-process — client.agent_tools() (plain functions), as_openai_tools(), as_anthropic_tools(), as_claude_mcp(), plus one-call openai_agent_config() / anthropic_runner_config() / claude_agent_config() bundles and agent_instructions() for the system prompt. Cloud clients get the live server tool set over the MCP bridge (read-only endpoint by default); local clients get the in-process subset with the same schemas and envelopes — agent prompts port unchanged. Chat surfaces: chat() — question in, answer out, on any backend; chat(stream=True) shows the run as it happens, thinking and tool calls woven into the text, or as typed events via .events; chat(protocol="responses" | "messages") drives the OpenAI Responses or Anthropic Messages API natively with that protocol's own shapes, and chat_completions() keeps the OpenAI-compatible envelope — all with doc_id targeting, streaming, honest usage accounting, and prompt-cache continuity across turns. Transcripts append verbatim: a protocol lane's output goes back into the next request unchanged. Model & connection knobs: index_model / chat_model, index_backend / chat_backend (and per-call backend) passed verbatim to each lane — LiteLLM-routed providers, keyless OpenAI-compatible servers, Azure/Bedrock/Vertex included. index= / chat= slots: the grouped spelling of the flat arguments — a string shorthand or a mapping (index={"model": ..

From the release notes

What's Changed Custom target connector sink API changes Container sinks now fulfill child target providers through per-action child slots (TargetActionSink.from_fn_with_children / from_async_fn_with_children); returning a ChildTargetDef list is deprecated, and in the Rust SDK that closure shape moved to from_async_fn_with_child_defs by @georgeh0 in #2407 Sinks whose callbacks compare equal but differ in type no longer share one sink identity, and callbacks that can't be weak-referenced (e.g. NamedTuple instances) now raise TypeError by @georgeh0 in #2404 fix(engine): confine a merged target-action batch failure to the failing component, instead of failing every component merged into the same sink call by @georgeh0 in #2406 Fixes for spurious re-execution of memoized functions Garbage-collecting an identical redefinition (re-run notebook cell, importlib.reload, function factory) no longer invalidates the live function's memo by @georgeh0 in #2405 copy.copy / copy.deepcopy of a coco function return the function itself, so dropping a copy no longer invalidates the original's memo by @georgeh0 in #2409 fix: component exception handlers receive the original typed Python exception instead of a wrapping RuntimeError by @hardness1020 in #2383 fix: preserve tuple and bytes component-path segments through state-store decoding, so target-state ownership for those components is no longer leaked or misattributed by @hardness1020 in #2403 Full Changelog: v1.0.22...v1.0.23

From the release notes

Signed-off-by: omnigent-ci[bot] <294685417+omnigent-ci[bot]@users.noreply.github.com>

From the release notes

Full Changelog: v1.2.45...v1.2.46

From the release notes

Full Changelog: v1.2.44...v1.2.45

From the release notes

Fix: graphify.serve now imports cleanly on Python 3.12 and 3.13. The chinese extra pins jieba-py from 3.12 onward (0.9.60 mistakenly kept the old jieba until 3.14, and its invalid regex escapes are a hard error on 3.12+), and the jieba import now suppresses the tokenizer's SyntaxWarning regardless of message or line so it never escalates under -W error. Fix: the git hook's rebuild-root guard now rejects a symlink-loop or dangling .graphify_root on Python 3.13, whose Path.resolve() no longer raises on a loop — the saved root must resolve to a real directory inside the repo before it is adopted.

From the release notes

Fix: atomic writes now fall back correctly on Windows WinError 17 (cannot move to a different drive), not just PermissionError — a shared os_replace_with_fallback copies through a temp in the target directory and restores the original if the swap fails, keeping install/export/cache writes crash-safe (#3508, thanks @ayushcodes10). Fix: files that could not be classified into any language or type are now surfaced (a count and top extensions in the console and GRAPH_REPORT) instead of vanishing from a "successful" run; noise and ignored paths are unaffected (#3511, thanks @ayushcodes10). Fix: a C or C++ header with no trailing newline no longer trips a tree-sitter syntax error — a newline is appended before parsing when the source lacks one (#3513, thanks @imanolpg). Fix: Office and Workspace document sidecars converted into graphify-out/converted/ are no longer dropped when graphify-out/ is gitignored — the tool stops ignoring its own output (#3504, thanks @ayushcodes10). Fix: a call inside an exported function (export function f(){ g() }, export const f = () => g()) now resolves, including calls to aliased imports (import { x as y }; y()) (#3346, thanks @abhay-codes07). Fix: a package.json exports map is now resolved by importer platform with the runtime condition preferred over types, so cross-package edges no longer drop to a non-existent .d.ts (#3487, thanks @dajiaohuang). Feature: Python 3.14 is now supported — optional dependencies without 3.14 wheels are gated to drop-in replacements (jieba to jieba-py, graspologic to graspologic-native) by interpreter version, and vulnerable dependency floors (setuptools, pypdf, yt-dlp) were raised (#3490, thanks @taazbro).

From the release notes

Fix: an incremental rebuild no longer drops cross-file concept nodes from files it didn't touch — global dedup during a merge now protects existing nodes from untouched files instead of collapsing same-labeled ones across them (#3477, thanks @hopstreax). Fix: extraction now falls back to sequential in-process work when the process pool cannot start (e.g. POSIX semaphore exhaustion), instead of aborting the whole run (#3497, thanks @curtismu7). Performance: Python symbol resolution is roughly 47% faster — path resolution is memoized, each file parses once across both resolution passes, and the tree walk is iterative rather than recursive; extraction output is unchanged (#3500 / #3501 / #3502, thanks @abhay-codes07). Fix: graphify explain now accepts a path::Symbol form to disambiguate a symbol that shares its name with its file, and the ambiguity hint now shows a form the resolver actually accepts (#3485, thanks @ayushcodes10). Fix: the git hook now keeps its rebuild root inside the repository — a committed .graphify_root pointing outside the worktree is ignored and falls back to the repo top, so a checked-in marker can't steer the hook to scan or write outside the tree (#3265, thanks @ayushcodes10).

From the release notes

0.10.18 (2026-09-12) fix(providers): don't throw on reasoning-only /v1/responses output (e116f4ed5cbcecb35f0ddbb350ff2fa0ea42eec3)

From the release notes

Signed-off-by: omnigent-ci[bot] <294685417+omnigent-ci[bot]@users.noreply.github.com>

From the release notes

What's Changed 🚀 Features Add a first-run banner describing the run, and open clai sessions with it by @DouweM in #7447 🐛 Bug Fixes Preserve text part boundaries after tool calls in OpenAIChatModel by @KaranJayakumar in #8235 fix(temporal): key tool opt-out checks on operation kind, not an MCP import by @adtyavrdhn in #8257 New Contributors @KaranJayakumar made their first contribution in #8235 Full Changelog: v2.42.0...v2.43.0

From the release notes

0.10.17 (2026-09-11) fix(providers): route /v1/responses when system content is cache-marked blocks (82f028e55e23c6b5f03413be2bdff819fef76ca6)

From the release notes

This release focuses on runtime MCP fixes, React hook performance, and improvements to the Inspector's onboarding experience.

From the release notes

MCP Serve MCP Apps: AG2 can now serve MCP Apps, with structured tool output and extension support. Input requests, both directions: An AG2 MCP server can ask the calling client for input, and an AG2 MCP client can answer a server that asks for it. Documentation Extension: TealTiger governance - output scanning output_scan: A new policy that inspects what a tool returns, rather than what it was asked to do. Every policy so far ran before the call; this one runs after, so a tool that returns something it should not can be caught on the way back. Contributed and maintained by @nagasatish007. Documentation Fixes: Middleware: Chat spans now record the full history and tool-call output, rather than a partial view. Documentation and project: Dead table-of-contents anchors and moved contribution-policy links repaired. What's Changed chore(deps): bump github/codeql-action/upload-sarif from 4.37.8 to 4.37.9 in the github-actions group by @dependabot[bot] in #3236 chore(deps): bump the uv group with 17 updates by @dependabot[bot] in #3237 fix(middleware): record full history and tool-call output on chat spans by @marklysze in #3238 feat(mcp): serve MCP Apps, with structured tool output and extensions by @vvlrff in #3235 feat(mcp): ask the calling client for input, and answer a server that asks by @vvlrff in #3230 feat(tealtiger): add output_scan governance policy by @nagasatish007 in #3240 docs: fix dead TOC anchors and moved contribution-policy links by @simpleqt in #3241 Fix flaky live LLM test assertions by @marklysze in #3247 Version bump to 1.0.5 by @marklysze in #3245 New Contributors @simpleqt made their first contribution in #3241 Full Changelog: v1.0.4...v1.0.5

From the release notes

Changelog 039ee52 chore: bump version to v3.20.77 [skip-release] [skip ci] 1e376f5 fix(openapi): isolate request-scoped upstream credentials

From the release notes

Hermes Agent v0.21.2 (v2026.9.11) — The state.db Patch Release Release Date: September 11, 2026 Patch release. v0.21.0 shipped a large rewrite of the session store's connection handling, and for some installs it made state.db fragile: second writers cancelling each other's locks, healthy databases reported as corrupt, one bad row killing sessions list. This release closes that class and rolls up everything else that landed on main in the four days since v0.21.1. About this release Measured at commit 04dd80a977f40b05e5b2054111747af07a61886a, the window since v0.21.1 contains 947 non-merge commits across 1,869 changed files (+182,504 / −15,564) and 312 merged PRs. 140 contributors appear in commits, co-author trailers, or salvage credits. ✨ Highlights state.db reliability campaign (six PRs, 44 issues closed) If your state.db broke after 0.21.0, this is the release for you. Six PRs fix the root causes rather than the symptoms: No more second writers. Profile gateways wrote hosted-room state into the root state.db every 5 seconds; the dashboard opened a writable handle on startup; cron's lifecycle guard did a raw open() on a live database (which cancels the gateway's POSIX locks — the classic "how to corrupt SQLite" recipe); doctor --fix would checkpoint under a live holder. All four are gone: hosted rooms live in shared-state.db, the dashboard opens read-only first, the guard goes through the tracked connection registry, and doctor --fix refuses a checkpoint it can't prove is safe. (#108076 — salvage #103489 @RikETS, #102682 @JoaoMarcos44, #108012 @Halldrix, #105428 @TaoMasterCoder) Healthy WAL databases stop wedging. OpenZFS (deleted) dentries and a close() racing an append_message both produced a sticky DeletedWalGenerationError on a perfectly good store; the read pool was handed out under an unconfirmed journal mode; a transient disk I/O error on WSL2 killed get_session on the first attempt; and a "state.db locked" banner was broadcast after the lock had already cleared. (#108082 — salvage #107411 @chelsealong, #105578 @ca-shrimp, #105711 @gaoanze888, #106958 @nikkoxgonzales; co-authored @QDung210, @fangliquanflq, @Sahilvishnaliya) FTS damage no longer kills your turn. An error scoped to the full-text-search index was classified as whole-file corruption and fail-closed the conversation. It's now fts_index: search degrades, the index rebuilds later, the transcript store is untouched. Same PR: doctor names structural damage honestly instead of "FTS write corruption", the FTS write probe catches the stale-index shape that passed every check while every write failed, .recover output no longer fails startup on orphan FTS5 shadow tables, header-zeroed databases recover instead of being refused, and the dashboard analytics poller returns a 503 instead of 520K tracebacks a day. (#108130 — salvage #97843 @SulthanZahran1 + #97841 @Finn763, #88604 #56824 #103657 @liuhao1024, #106890 @nftpoetrist, #103321 @jangomango76, #91413 @leegunwoo98, #102808

From the release notes

Added Added an opt-in authoritative run store through OrchestratorConfig.runStore and RunTasksOptions.runStore; it gives a logical run a durable lifecycle record, an execution lease, and a monotonically increasing fencing token so one worker at a time advances the run and a worker that was taken over cannot write after the takeover. Added the RunStore interface, the MemoryStoreRunStore adapter over any MemoryStore with compare-and-set, and the RunLedger and RunLeaseHandle surface for reading, cancelling, and resuming a run from outside its worker. Changed TeamConfig.maxConcurrency now bounds the agent pool for that team's runs instead of being accepted and ignored. It intersects with OrchestratorConfig.maxConcurrency, and the smaller value wins, so a team can narrow its own pool but never widen it past the orchestrator ceiling. A team that set a cap below the orchestrator's was previously running at the orchestrator value and now runs at its own lower value; raise or remove the team cap to keep the previous throughput. A cap that is not an integer of at least 1 is reported as an INVALID_TEAM_MAX_CONCURRENCY warning on onProgress and the orchestrator value applies. With a run store configured, execution-ownership and run-lifecycle writes fail closed instead of following the best-effort semantics of checkpoint and telemetry writes: a fenced-out checkpoint write never reaches the store, a lost lease stops the run at the dispatch gate, and a worker that lost its lease reports the fence failure rather than success. Runs with no run store configured are unchanged. The oma CLI default model for provider: 'deepseek' is now deepseek-flash (DeepSeek-V4.1-Flash), replacing deepseek-v4-flash. The retired name still resolves, and library users that pass an explicit model are unaffected. Provider docs, adapter JSDoc, the scaffolder env template, and the DeepSeek examples moved to the same name. The DeepSeek adapter JSDoc no longer states that thinking runs at high effort. The adapter forwards whatever effort the caller passes and sets none itself; runtime behavior is unchanged. Corrected published JSDoc that misstated runtime behavior, including onAgentStream coverage in runTasks(), done stream-event payloads per layer, retryable-error classification, and the RedactingStore example constructor call. The delegate_to_agent error returned when the pool has no free slot now points at both ceilings: "Increase maxConcurrency on the orchestrator or the team" rather than the orchestrator alone. Callers that match on that string should update it. The @open-multi-agent/core package description on npm was rewritten to lead with running in your own environment, approval gates, and a verifiable record of every run. LICENSE now names Shenzhen YuanASI Technology Co., Ltd. alongside the open-multi-agent contributors, package READMEs close with a maintainer line, and the author field in the published package.json files points at YuanASI. The MIT terms are unchanged.

From the release notes

Changes since langchain-core==1.6.2 release(core): 1.6.3 (#40407) feat(core): Allow model name and provider tracing metadata override based on gateway response (#40406) test(core): cover the deprecated .text() access path (#40243) docs(core): remove stale Args/Raises entries from FileCallbackHandler._write and ChatGeneration.set_text (#40211)

From the release notes

What's Changed .NET: Replace deprecated AWSSDK.Extensions.Bedrock.MEAI with AWS.Bedrock.MEAI by Ben Thompson (@BenGearset) in #7983 .NET: Remove Azure.AI.OpenAI dependency by Roger Barreto (@rogerbarreto) in #7986 .NET: Track and update A2A task state by SergeyMenshykh in #7998 .NET: Update Azure AI Projects to 3.0.0 beta 1 by Roger Barreto (@rogerbarreto) in #7995 .NET: Bump Aspire.Hosting from 13.5.2 to 13.5.3 by Dependabot (@dependabot)[bot] in #8029 .NET: [BREAKING] Add file_access_read_lines and move the line-numbering contract onto AgentFileStore by Anton Sokolovskyi (@antsok) in #7671 Bump Anthropic from 12.42.0 to 12.43.0 by Dependabot (@dependabot)[bot] in #8028 .NET: Bump AgentMemory.AgentFramework from 1.4.1 to 1.5.0 by Dependabot (@dependabot)[bot] in #8027 .NET: [BREAKING] Clarify A2A agent run modes by SergeyMenshykh in #8032 .NET: Clarify declarative workflow input serialization by 宇宙的山峰 (@UniversePeak) in #8046 .NET: Preserve streamed annotations in Foundry hosted responses by Roger Barreto (@rogerbarreto) in #7984 .NET: Improve Hosted Agent LRA Resilient Recovery Sample by Roger Barreto (@rogerbarreto) in #8082 .NET: Include workflow outputs in hosted agent responses by 宇宙的山峰 (@UniversePeak) in #8020 .NET: Improve inline skill argument error guidance by SergeyMenshykh in #8118 .NET: Bump Azure.Monitor.OpenTelemetry.Exporter from 1.5.0 to 1.8.3 by Dependabot (@dependabot)[bot] in #7887 [BREAKING] .NET: Isolate LocalCodeAct subprocess environment by Eduard van Valkenburg (@eavanvalkenburg) in #8159 .NET: Scope OpenAI hosting storage by isolation key by SergeyMenshykh in #8146 .NET: chore: updates source link dependency due to transitive vulnerability by Vincent Biret (@baywet) in #8179 .NET: chore: updates Microsoft owned testing dependencies by Vincent Biret (@baywet) in #8167 .NET: ci: update Microsoft.CodeAnalysis.NetAnalyzers and Microsoft.VisualStudio.Threading.Analyzers versions by Vincent Biret (@baywet) in #8166 .NET: chore: updates microsoft extensions and related packages to their latest version, as well as global.json dotnet sdk version to align by Vincent Biret (@baywet) in #8190 .NET: Bump Azure.AI.Projects to 3.0.0-beta.2 by Roger Barreto (@rogerbarreto) in #8165 .NET: fix: revalidate file skill paths before use by SergeyMenshykh in #8151 .NET: Opt in to client function forwarding for Responses hosting by Roger Barreto (@rogerbarreto) with @Copilot in #7844 .NET: chore: updates additional analyzers packages by Vincent Biret (@baywet) in #8198 .NET: Harden LocalCodeAct OS validation by Eduard van Valkenburg (@eavanvalkenburg) in #8239 .NET: upgrades xunit and other dependencies by Vincent Biret (@baywet) in #8202 .NET: ci/promote removed apis by Vincent Biret (@baywet) in #8229 .NET: fix: do not forward headers on redirect by Vincent Biret (@baywet) in #8164 .NET: updates open telemetry deps by Vincent Biret (@baywet) in #8253 .NET: Dotnet: reset $LASTEXITCODE per command in persistent Po

From the release notes

Changelog eadb4c1 chore: bump version to v3.20.76 [skip-release] [skip ci] 6c7711d fix: scope physical identifiers and detect schema changes

From the release notes

Kiro Crew stops being one agent on one machine: choose the harness that runs your sessions, start a chat on another crew you are connected to, and let crew members dispatch workers of their own. Chat gets quieter, with folded diffs, resumed sessions that stop re-reading their own context, and refused commands that tell the agent what it may do instead. Python 3.12 is the new floor. Before you upgrade Python 3.12 is now the floor: a host on 3.10 or 3.11 must move up before installing or updating, and every installer provisions 3.12 itself when the system package manager has none. The installer now brings its own Python: curl -fsSL https://download.crew.kiro.dev/cli.sh | sh provisions a pinned CPython rather than using the system one, so pass --system-python if you deliberately build against yours. The command gate no longer fences credential paths by reading the command text: the sandbox's bind masks are the fence now, so check that agent.sandbox in config.json is not off if you were relying on that text gate. An unattended auto-run plan now stops after two hours: raise it with orchestrator.max_plan_duration_seconds in config.json, or set 0 to remove the ceiling; a stage-gated plan is never cut. Pick the harness that runs your sessions (Preview) Claude Code, Codex and KAS are selectable harnesses: turn on Developer Mode in Settings → Developer (off by default), then pick one under Developer → Agent Backend, where each option says whether it is installed, missing on this machine, or waiting on a gateway restart. A tool pre-approved in Claude's own settings never reaches Crew's approval path, so its deny rules and audit log do not see that call, and Codex refuses to start while the sandbox is off. Monitor loops, project changes, follow-up cards and conversation reset work on every backend, where they previously failed closed outside Kiro CLI. Remote crews become one dashboard (Preview) Run a chat on another crew you are connected to: set instances.enabled in config.json and turn on Settings → Developer → Feature Previews → Chat on a crew (both off by default), then pick the peer under New chat on crew in the sidebar's new-chat menu; the transcript stays local while every turn runs on that crew. A session that runs elsewhere carries a server badge and the crew's name in your sidebar, so remote work is openable from the list you already read. Crews connect themselves on app load and on tab focus, on by default and switchable at Settings → Remote Instances → Auto-connect crews. Chat that stays out of the way Diffs start folded as a chip naming the file and its +N/-M counts, and Settings → Chat → Plain diffs (off by default) renders patches as plain monospace text instead. A sketch pad and a share action: the composer's + menu gains a Sketch row that draws an image and attaches it, and an assistant reply's Share as image action exports a branded PNG or prefills an X or LinkedIn post. The composer says where you are, and gets out of the

From the release notes

What's Changed fix(deps): backport the electron js-yaml 4.3.2 pin to release/0.6.0 by @bolichen97 in #10076 Full Changelog: v0.6.0-insider.4...v0.6.0-insider.7

From the release notes

Signed-off-by: omnigent-ci[bot] <294685417+omnigent-ci[bot]@users.noreply.github.com>

From the release notes

v2026.9.11 Download installers macOS Apple Silicon: Orkas-mac-arm64.dmg macOS Intel: Orkas-mac-x64.dmg Windows x64: Orkas-Setup.exe Highlights Organize project to-dos and queued work with the task board, updated task details, and clearer sidebar status. Connect more services with guided setup, task-scoped permissions, and API Key access for credit-based connectors. Browse inside Orkas with tabs, persistent sign-ins, and improved browser recovery after restarting. Improvements Respond to local coding-agent questions and cancel pending input without losing the surrounding task. Use updated built-in workflows for Office documents, visual content, and website analysis. Fixes Recover conversation history without duplicate replies, and preserve newer drafts while earlier messages are being sent. Keep search results working after a restart and restore missing download addresses for built-in marketplace updates. Keep keyboard focus where you place it and improve connector status labels and task-board keyboard controls. Full changelog Compare v2026.8.29...v2026.9.11

From the release notes

🌐 English | 中文 🤝 Multi-Agent Teams CowAgent is now a multi-Agent framework. You can create multiple Agents, each with its own role, to form a team and collaborate on complex tasks within a single session. Supported on the Web console, the desktop client, and channel integrations: Team management: create and maintain Agents, configure name, responsibilities, default model, skills, and knowledge, and copy the setup from an existing Agent. Resource isolation: each Agent has its own workspace, memory, and sessions; the knowledge base and skills can be shared or kept independent. Group collaboration: add multiple Agents to one session and use @ to direct a message to a specific member, with each Agent using its own configured model. Task delegation: a group chat has a lead Agent that can hand a task off to the member best suited for it based on their responsibilities, with support for multi-level delegation and configurable scope, depth, and timeout. Multiple channel instances: an IM channel of the same type can now run multiple instances, each bound to a single Agent or an Agent team. Supported on WeChat, WeCom smart bots, Feishu, DingTalk, QQ, Telegram, Slack, Discord, and more. Thanks @AaronZ345 (#2975, #2973), @zkjqd (#3096), @cowagent (#3118) Docs: Agent Teams ⏰ Scheduled Task Upgrades Scheduled tasks can now be created and managed from the console, with channel recipients and execution records added: Manual task creation: create, edit, and run tasks directly on the tasks page, and select the channel instance and recipient. Channel recipient management: maintain trusted recipients per channel instance, so a task can be delivered to a designated recipient rather than only the channel where it was created. Thanks @dajiaohuang (#3043) Execution records: a new execution-records page lets you review the history list and the details of each run. Thanks @mengluo04 (#3107) Concurrent-update fix: fixed tasks being lost on concurrent updates (Thanks @Whxuan0701 #3062). 📊 Context-Usage Visualization Context consumption is now visible and controllable, making token cost easier to manage: Context-usage view: a ring view replaces the previous clear-context button and shows a full breakdown of what makes up the context. Context actions: below the ring view you can trigger smart compaction, clear the context, or set the maximum context size. Context calculation: the usage breakdown is derived automatically from the model's reported usage, with agent_max_context_tokens available as a user-configurable cost cap. Thanks @chimyves (#3087), @sufan721 (#3079) 🤖 Model Capabilities New models: added deepseek-flash (DeepSeek V4.1 Flash), gpt-6-astra, claude-fable-5-1, qwen3.8-flash, glm-5.3-flash, gemini-3.8-flash, and the vision model deepseek-v4-flash-vision-exp (Thanks @a1094174619 #3094). Model fallback: configure a fallback model under Model Config > Primary Model, which takes over automatically when the primary model fails, preventing interr

From the release notes

Fix: a call to a Python function defined nested inside another function now resolves to that inner definition per lexical scope, instead of leaking to a same-named function elsewhere; direct recursion is preserved as a self-loop (#3410, thanks @hopstreax). Fix: submodule imports inside a PEP 420 namespace package (a directory with no __init__.py) now resolve to the target module instead of being dropped (#3429, thanks @flaukowski). Fix: a bare-name import of a module sitting next to the importing file (a flat script dir with no package) now resolves to that sibling — matching CPython's sys.path[0] behavior — without over-resolving a genuine third-party name (#3430, thanks @hopstreax). Fix: a PHP use Foo\Bar as Baz; import keys its local binding on the alias, so later Baz references resolve to the real class, and distinct aliased same-named classes stay separate (#3421, thanks @ayushcodes10). Fix: a directory literally named out is skipped as build output only when there is build-output evidence, so a real source out/ is no longer silently dropped (#3347, thanks @abhay-codes07). Fix: a shell script invoked in exec position through a variable path ("$SCRIPT_DIR/foo.sh") now resolves to the target when the variable holds a constant, matching the existing source handling (#3416, thanks @edwardselby). Fix: added the missing Iterable import in build.py so its type annotations resolve (they were an undefined name) (#3462, thanks @xiongjianxu). Fix: graphify global add no longer infers an empty repo tag for a path like /tmp/graph.json; it degrades to a sensible non-empty tag via the same helper merge-graphs uses (#3464, thanks @xiongjianxu). Feature: SQL extraction now emits index nodes for CREATE [UNIQUE] INDEX, linked to the table they index (#3467, thanks @L4XB). Fix: a TypeScript export * re-export no longer resolves a name to a same-named interface method — only module-level exports are candidates (#3436, thanks @L4XB). Feature: Rust module-level static and const declarations are now extracted as nodes (#3471, thanks @L4XB). Fix: an incremental rebuild no longer drops a node's rationale/summary when a semantic node's absolute source_file collides with its AST twin's relative path during dedup — source paths are normalized first (#3472, thanks @hopstreax). Fix: graphify install no longer aborts when the always-on registration target is unwritable (read-only or symlinked config); it skips that step with an actionable warning and still installs the skill (#3474, thanks @dajiaohuang). Fix: community labelling falls back to an installed claude CLI resolved at run time instead of pinning a path that expires (e.g. under snap/nvm), so labelling keeps working across updates (#3475, thanks @ktsang622). Fix: the all extra now includes psycopg[binary], so pip install 'graphifyy[all]' provides the postgres driver (#3482, thanks @L4XB).

From the release notes

Changelog 2.9.0 (2026-09-10) Highlights This release focuses on agent resilience and ecosystem integration by introducing automatic model failover, native voice support via LiveKit, and compatibility with MCP SDK 2.x. Models: Safeguard agent uptime by automatically failing over to backup models when the primary model encounters errors. (c300b8d) Integrations: Build rich voice and telephony agents using the newly added LiveKit runner. (f277d91) Workflows: Simplify graph workflow management by authoring and loading ADK 2.0 graphs directly from declarative YAML configurations. (2a9461b) MCP: Work with MCP SDK 2.x servers as well as 1.x. Installs continue to resolve 1.x by default; install 2.x deliberately to opt in. (856acf2) Breaking changes Workflow node resumption: A node that failed now runs again when the workflow resumes, where before it replayed as though it had completed. Make node bodies idempotent: a node that performs an external side effect and then fails will perform that side effect again on every resume. GCS tool local paths: The GCS tools now read and write local files only inside the directory named by local_file_root, and refuse local file access entirely when that setting is absent. Paths may be absolute or relative, and are judged by where they resolve. In-memory sessions only: InMemorySessionService now raises SessionNotFoundError when an event is appended to a session it does not hold, instead of accepting the event and discarding it. The database-backed services already behaved this way, so this affects in-memory sessions alone. Two further changes are marked breaking in the commit log but do not affect a default installation. Nested workflow retry changes behavior only for a node that already sets retry_config. The MCP SDK 2.x field handling applies only if you install MCP 2.x deliberately, which this release does not resolve to. All changes ⚠ BREAKING CHANGES retry a nested workflow when a node inside it fails confine GCS tool local file paths to a configured root rerun a failed node on resume instead of replaying it as complete raise SessionNotFoundError when appending to an unknown session mcp: under MCP SDK 2.x, fields a server sends that the SDK does not declare no longer reach the caller. 2.x closed its models, so an unknown key on a CallToolResult, or on a tool declaration read via raw_mcp_tool, is discarded during validation before ADK sees it. _meta is unaffected: it is a declared field and still arrives. Nothing in ADK can restore the others; a server that relies on vendor extensions should move them under _meta. On 1.x they still arrive as before. Features add a2a_state_forwarding sample (1b28de4) add FallbackModel for automatic failover between models (c300b8d) add get_authenticated_url and get_signed_url to GcsArtifactService (10ac52b) add ignore_args option to tool trajectory evaluation (e2a213f) add transfer_reason to transfer_to_agent to prevent loops (22ab4d8) agent evaluations pri

From the release notes

v2026.09.10 — 三語正式版 / Trilingual release 繁體中文 maintenance 發布前掃描發現 notebooklm-skill 已封存，三語資源表與 Cookbook 改列歷史範例，不再推薦新流程使用。Graphthulhu 原入口回傳 404，移除失效連結並保留歷史名稱與評分；其授權目前無法重新核對。原編輯星級不變，沒有新增替代專案。 來源 1 reader-ux 首頁頂部 Banner 保留原版構圖、配色、文字與圖示，加入 18 秒循環動畫。先走共用基礎，再依序展示 A、B 路線；每語言 13 個圖示區域會做對應的小動作，包括 CLI 游標輸入、工具輕轉與 Hub 箭頭旋轉。文字與卡片固定，最後 2 秒保持靜止。 來源 1 accessibility 三語文件站提供停止／播放控制；系統偏好減少動態或停用 JavaScript 時使用靜態圖。README 保留靜態圖入口，三語 PDF 也使用完整 PNG。SVG 不載入外部圖片、字型或 JavaScript，圖片容量上限未提高。GitHub 嵌入圖片是否遵守真實系統的減少動態偏好仍未確認，需要靜態閱讀時請使用靜態圖入口或文件站控制。 來源 1 known-limitations 首頁下方的舊學習地圖仍待修正，不屬於本次頂部 Banner 更新。Track A 的正確順序是 A1 → A2 → Stage 5 → A3 → Stage 8；Stage 7.5 是進階選讀。修圖前請以正文中保持展開的路線表為準；這張舊圖尚未加入動畫。 來源 1 简体中文 maintenance 发布前扫描发现 notebooklm-skill 已归档，三语资源表与 Cookbook 改列历史范例，不再推荐新流程使用。Graphthulhu 原入口返回 404，移除失效链接并保留历史名称与评分；其授权目前无法重新核对。原编辑星级不变，没有新增替代项目。 來源 1 reader-ux 首页顶部 Banner 保留原版构图、配色、文字与图标，加入 18 秒循环动画。先走共用基础，再依次展示 A、B 路线；每语言 13 个图标区域会做对应的小动作，包括 CLI 光标输入、工具轻转与 Hub 箭头旋转。文字与卡片固定，最后 2 秒保持静止。 來源 1 accessibility 三语文档站提供停止／播放控制；系统偏好减少动态或禁用 JavaScript 时使用静态图。README 保留静态图入口，三语 PDF 也使用完整 PNG。SVG 不加载外部图片、字体或 JavaScript，图片容量上限未提高。GitHub 嵌入图片是否遵守真实系统的减少动态偏好仍未确认，需要静态阅读时请使用静态图入口或文档站控制。 來源 1 known-limitations 首页下方的旧学习地图仍待修正，不属于本次顶部 Banner 更新。Track A 的正确顺序是 A1 → A2 → Stage 5 → A3 → Stage 8；Stage 7.5 是进阶选读。修图前请以正文中保持展开的路线表为准；这张旧图尚未加入动画。 來源 1 English maintenance The pre-release scan found notebooklm-skill archived. All three catalogs and Cookbooks now keep it as a historical example, not a recommendation for new workflows. Graphthulhu's original address returns 404; its dead link was removed while retaining its historical name and rating. Its license cannot currently be reverified. Editorial ratings are unchanged, and no replacement project was added. source 1 reader-ux The homepage's top banner keeps the original composition, colors, text, and icons, with an 18-second animation loop. Shared foundations come first, followed by tracks A and B; 13 icon regions per language make small, meaningful movements, including typing CLI cursors, gently turning tools, and rotating Hub arrows. Text and cards stay fixed, with 2 still seconds at the end. source 1 accessibility All three documentation locales provide stop/play controls and use a static image when reduced motion is preferred or JavaScript is disabled. Each README links to its static image, and all three PDFs use the complete PNG. SVGs load no external images, fonts, or JavaScript; image budgets have not increased. Whether GitHub's embedded image follows the actual system reduced-motion preference remains unverified; use the static-image link or documentation controls for still viewing. source 1 known-limitations The older learning map lower on the homepage still needs correction and is not part of this top-banner update. The correct Track A order is A1 → A2 → Stage 5 → A3 → Stage 8; Stage 7.5 is an advanced option. Until that image is corrected, follow the always-visible route tables in the text. The old map has not been animated. source

From the release notes

0.10.16 (2026-09-10) fix(providers): correct GPT-5.6 cache retention and add the shipped 5.6 siblings (7f55072d02ee1d73383539312ebba2f5c13fbeb6)

From the release notes

0.10.15 (2026-09-10) feat(providers): support GPT-6 (gpt-6-astra) across the OpenAI param guards (f33595810842c5a0e8a411d104b655c07ae938ed)

From the release notes

Patch release. Three fixes from the issue backlog, each confirmed in source before any change and each with a regression test proven to fail without the fix. A curate can no longer destroy a user's memory index (#3224) AutoMemoryBridge.curateIndex() rebuilt MEMORY.md from its own topic files and overwrote whatever was there. The #1556 guard only skipped the write when nothing matched — but the bridge writes the first topic file itself, so the guard protected exactly one sync and then stopped applying. The reporter measured a hand-curated index of 75 lines and 49 links reduced to a 6-line stub. A curate may now grow or reorder the index and never shrink it. When the generated view is smaller, the existing file is left untouched, the generated view is written beside it as MEMORY.generated.md, and the bridge emits index:preserved. The old failure was silent; a silent refusal would only be a quieter version of the same problem. autoStart: false is honoured (#3278) Two mismatches were stacked: what init writes what the check read file .claude/settings.json claude-flow.config.json key claudeFlow.daemon.autoStart daemon.autostart A different file and a different capital S, so the setting a fresh project ships was never consulted — under a comment reading "Opt-in only — prevents unintended token consumption (#1427, #1330)". Both files and both spellings are now read. Reads disclose the store they did not search (#3228, in part) 3.41.1 shipped the sibling-store disclosure in memory list alone. retrieve, search and an empty list kept returning a clean negative while the rows sat in the file they do not read — on the reported Windows install, 31,673 rows answering found:false. Still not fixed: the Windows 0xC0000409 abort on a ~4.4 GB allocation with the native bridge enabled. That is #2948 and this release does not touch it. On Windows the split is now visible, not healed. Install npx ruflo@3.41.2 Merge commit df87b0db338a.

From the release notes

Patch release. Two fixes, both reported by people hitting them rather than found in-house. Memory path identity (#3196) An explicit database path now selects that database. The bridge cached one registry globally, so the first caller to touch it decided the file for the whole process and every later caller's --path or CLAUDE_FLOW_DB_PATH was accepted and then silently ignored. A CLI write and an MCP write landed in two files, both reporting success, and neither interface could see the other's rows. memory list also stopped printing a confident total for one store while rows sat unread in its sibling — it now names the store it did not read. Reported downstream by pacphi/agentic-kit#213. Not done deliberately: merging the two database files. #2786 gave AgentDB its own plaintext file so native better-sqlite3 never opens an encrypted memory.db. Unifying them would re-break encryption at rest. Seraphina truncation (#3267) Seraphina no longer passes a truncated reasoning dump off as coordination guidance. The routed model spent its whole token budget reasoning and never emitted its JSON, and the resulting empty proposal list read as the swarm needs nothing — the one conclusion a coordinator must never reach by accident. Install npx ruflo@3.41.1 Merge commit cda8b04f6ae1.

From the release notes

Release v4.7.7

From the release notes

Changes since langchain-anthropic==1.7.1 release(anthropic): 1.7.2 (#40387) fix(anthropic): preserve invalid tool use blocks (#40372)

From the release notes

No content.

From the release notes

Patch Changes 5e418dc: Document that onResume / on_resume needs a control plane that knows the option: an older self-hosted or BYOC control plane drops the memory field and restores memory while reporting success, instead of rejecting the request. 9136603: Retry control-plane HTTP requests up to three times after 429 responses using the server's delta-seconds Retry-After delay. Retries can be configured or disabled with retries, and stop when waiting would exhaust the request timeout. Envd requests, including filesystem operations, and volume-content requests are not retried.

From the release notes

Patch Changes 90e3fc5: Reject invalid Sandbox.create lifecycle options and Sandbox.connect onResume values before requiring an API key, matching the Python SDK. 5e418dc: Document that onResume / on_resume needs a control plane that knows the option: an older self-hosted or BYOC control plane drops the memory field and restores memory while reporting success, instead of rejecting the request. 9136603: Retry control-plane HTTP requests up to three times after 429 responses using the server's delta-seconds Retry-After delay. Retries can be configured or disabled with retries, and stop when waiting would exhaust the request timeout. Envd requests, including filesystem operations, and volume-content requests are not retried.

From the release notes

Covers the 24 commits since the v0.5.69 tag. The package version already moved to 0.5.75 in 4a39068 (CLI model selector), so the release commit is the changelog alone, matching the convention in eb712ca/4eda76e2a. Co-Authored-By: Claude Code noreply@anthropic.com

From the release notes

Retain the shared handoff runtime source for installed plugins; not a…

From the release notes

The PageIndex SDK, local or cloud — vectorless, reasoning-based RAG, end to end. Much faster indexing — the PageIndex Flash engine gets the tree from layout stats: no LLM involved for the structure generation itself, LLMs only write the node summaries, and tree expansion proposes a wave of nodes concurrently. client = PageIndexClient() client.submit_document("report.pdf") client.chat("What does the report conclude?") Index, to chat, to agent integration, one client. Local mode needs no server, no vector DB, no PageIndex API key. Highlights Flash engine: the local default (mode="standard" keeps the classic LLM pipeline). Embedded bookmarks are consumed when trustworthy, and tree optimization is on by default — optimize="merge" for the deterministic LLM-free pass, "full" (default) adds LLM expand, which runs a wave of nodes concurrently instead of one round-trip at a time. One complete surface, local and cloud: PageIndexLocalClient(storage_path=...) is the same client as cloud — submit, tree, page content, chat, and agent tools all present in both modes, so code moves between them unchanged. Cloud documents, your own model: api_key decides where your documents live; a configured chat model decides who answers — and the two combine. PageIndexClient(api_key="pi-...", chat_model="openai/gpt-5.2") runs the same in-process document-QA engine over the live cloud tool set. Page content flows through your process to your provider on your credentials; doc_id targets at the prompt level; enable_citations stays with the managed chat. Agent integration: the cloud MCP tool contract, in-process — client.agent_tools() (plain functions), as_openai_tools(), as_anthropic_tools(), as_claude_mcp(), plus one-call openai_agent_config() / anthropic_runner_config() / claude_agent_config() bundles and agent_instructions() for the system prompt. Cloud clients get the live server tool set over the MCP bridge (read-only endpoint by default); local clients get the in-process subset with the same schemas and envelopes — agent prompts port unchanged. Chat surfaces: chat() — question in, answer out, on any backend; chat(stream=True) shows the run as it happens, thinking and tool calls woven into the text, or as typed events via .events; chat(protocol="responses" | "messages") drives the OpenAI Responses or Anthropic Messages API natively with that protocol's own shapes, and chat_completions() keeps the OpenAI-compatible envelope — all with doc_id targeting, streaming, honest usage accounting, and prompt-cache continuity across turns. Transcripts append verbatim: a protocol lane's output goes back into the next request unchanged. Model & connection knobs: index_model / chat_model, index_backend / chat_backend (and per-call backend) passed verbatim to each lane — LiteLLM-routed providers, keyless OpenAI-compatible servers, Azure/Bedrock/Vertex included. index= / chat= slots: the grouped spelling of the flat arguments — a string shorthand or a mapping (index={"model": ..

From the release notes

RuFlo 3.41.0 — Join the open swarm in four commands Any machine running RuFlo can now join a shared, signed, invite-gated swarm at x.ruv.io with a key only it holds. No Tailscale, no VPN, no account. Step-by-step guide with troubleshooting: https://gist.github.com/ruvnet/6d3202cc7030e32169e8d5ac73b4e870 Illustrated explainer: The Swarm Wall (Cognitum Media plates) Try it npx ruflo@latest federation join --code v2.… # 1 · get in with your own key (code comes to you privately) npx ruflo@latest federation roster # 2 · you are listed npx ruflo@latest federation claims # 3 · who owns what npx ruflo@latest federation sync # 4 · read what the swarm posted Then in Claude Code: claude mcp add --transport http --scope user x-ruv-io https://x.ruv.io/mcp What's new ruflo federation command (#3256): join, sync, roster, claims, registry, plus admin-gated invite, admit, publish. Self-service join generates ~/.ruflo/nostr.key (0600), redeems the invite with a NIP-98 signature, and proves membership with NIP-42. Public and private swarm channels (#3261, ADR-386): ruflo federation channel --action create|grant|accept|publish|read|list. Private channels are NIP-44 encrypted with keys the gateway never sees. x.ruv.io gateway plugin (#3255): MCP at https://x.ruv.io/mcp, WebSocket proxy, ruv://federation/registry, ruv://swarm/roster, ruv://claims/board, ruv://swarm/channels. Admin writes are token-gated and fail closed. Seraphina (#3256): swarm coordinator via the cognitum meta-llm. seraphina_guidance { goal } returns guidance, proposals and risks. Advisory only. Canonical relay wss://relay.ruv.io. Sign the NIP-42 relay tag with this exact address; the old *.run.app host still works as legacyRelay. 15 federation MCP tools ship in the RuFlo MCP server. CI: @metaharness/darwin pinned to 0.10.2 so the pin-drift and clean-install gates are green again (#3262); skills and gateway docs cover channels (#3263). Validated before this release 67 CLI tests and 12 gateway tests green; all 23 workflows green on main. Fresh npm install ruflo@latest on a clean cache: federation join with a new key, registry, roster, sync, and a public-channel publish/read round trip against the live relay. Relay E2E on three paths: direct wss://relay.ruv.io, the legacy host, and proxied through wss://x.ruv.io. Private-channel E2E from a second, independent key: plaintext for members, ciphertext only for non-members. Known ruflo verify reports the same one stale witness marker (#1697 rvf-wasm overrides) as 3.40.0; manifest refresh is tracked separately. Packages @claude-flow/cli, claude-flow, ruflo → 3.41.0 (latest, alpha, v3alpha).

From the release notes

New features Agentic RAG: Refactored the Agentic RAG retrieval framework, significantly improving reasoning speed and benchmark performance. (#19046, #19112, #19172, #19209, #19424, #19459) Knowledge compilation runtime settings (#19254) Enforce a rate limit on the knowledge compilation pipeline (#19156) Knowledge graph shows total and current node counts (#19142,#19095, #19114) Graph search highlights the node when pressing Enter on an entity name (#19320) New Sitemap data source for sitemap.xml-based web ingestion (#19344) WebDAV connector supports custom CA certificates (#16901) MonkeyOCRv2 parsing support (#18887) Self-hosted PaddleOCR-VL deployment support (#19011) New Sofya search tool for agent (#19323) EPUB document preview (#19242) Excel citation sources located in document preview (#18853) Model Support New Hubris model provider (#19341) New llmman model provider (#19030) Improvements New RERANK_TOKEN_LIMIT_MODE configuration (#19404) Highlight in retrieval testing and search (#18981) Similarity threshold displayed as a percentage (#18974) UI/UX Completed Turkish locale (#19009) Search match highlighting in Team settings tables (#15985) Bug fixes Fixed images uploaded in chat being ignored by vision models (#18884) Fixed non-UTF-8 encodings in CSV/TXT preview and document decoding (#19222, #19253) What's Changed fix(agentic_rag): include configured system prompt in RAGTools.sys_prompt() by @chakshu-dhannawat in #18927 fix(web): make sidebar i18n reactive and refine template card/icon UI by @yooxdb in #16646 fix(dataset): align parser methods order with python by @6ba3i in #18877 feat(webdav): support custom CA certificates by @taek105 in #16901 Fix Azure embedding endpoint handling by @zcxGGmu in #16993 fix go: migrate missing table chatchannel by @myf-bee in #18944 Python && Go: fix parse status by @JinHai-CN in #18841 fix(ollama): preserve native API base URL in OllamaCV by @iwasaki1112 in #17013 fix(file): stop is_parent_folder_exist raising on a missing parent by @marmar9615-cloud in #18930 fix(agent): enforce MCP task deadlines in worker by @Lem0nTea2002 in #18869 fix: upgrade starlette to 1.3.1 (CVE-2026-54283) by @anupamme in #18947 refactor: delete the unused discord_svr.py bot script by @noron12234 in #17682 Refactor: make all url inside tooltip as link by @wangq8 in #18943 Py: add update_time in select list when list data sync logs by @JinHai-CN in #18867 fix: include filename in file download Content-Disposition header by @huang-aoqin in #17105 fix(embed): drop_params: True was rejected by Together AI and other strict providers by @Harsh23Kashyap in #18627 fix(rag): key Infinity field_map by original column names, not pinyin by @Taranum01 in #18318 feat(go-syncer): add Zendesk connector by @mkaaad in #18961 Fix: Remove the empty response field from the retrieval operator. by @cike8899 in #18963 fix(agent): reject fractional integer DSL parameters by @6ba3i in #18964 Fix: update t

From the release notes

Warning Self-hosted deployments using the bundled Weaviate must complete a manual, staged upgrade before starting 1.17.1. The bundled Weaviate server moves from 1.27.0 to 1.39.2 — 12 minor versions — and skipping minors is unsupported. Pulling and restarting can silently and permanently break vector search. See Upgrade Guide below. Fresh deployments, external Weaviate, and other vector stores are unaffected. New Features Dataset-Scoped Knowledge Base API Keys Knowledge base service-API keys were scoped to the whole workspace: one key could read and write every knowledge base in the tenant, so giving an integration access to a single knowledge base meant giving it access to all of them. Keys can now be bound to specific knowledge bases from the knowledge base's API access panel, with a workspace / this-knowledge-base scope selector at creation time. A bound key returns 403 for any other dataset — both documents and retrieval — and for endpoints that carry no dataset ID, such as list-all. Existing keys are unbound and keep their workspace-wide behavior, so nothing changes when you upgrade. (#37569) Keyboard Movement for Workflow Nodes and Comments The workflow canvas required a mouse to reposition anything. Nodes, multi-node selections, comment markers, and draft comments can now be moved with the arrow keys — hold Shift for larger steps — including Note, Iteration, and Loop nodes. Continuous movement is committed on key release or blur, so it produces one undo entry rather than one per keypress, and read-only mode, container boundaries, and collaboration updates are all still respected. (#41967) Marketplace Creator Profiles and Homepage Redesign Plugin authors now have public profile pages at /marketplace/creator/<handle>, listing their creations with a publisher sidebar, reachable from a new Creator Center entry in the sidebar help menu. The marketplace homepage was rebuilt around a hero, a banner carousel, sticky search, catalog tabs, and tag/language filters, with a stacked-and-swipeable mobile layout. Bundles are no longer mixed into plugin categories. (#41538) Configurable Markdown Form Field Name Length The Markdown form field-name limit was hard-coded at 128 characters, with no way to accommodate longer generated field names. It is now configurable through MARKDOWN_FORM_FIELD_NAME_MAX_LENGTH (Docker) or NEXT_PUBLIC_MARKDOWN_FORM_FIELD_NAME_MAX_LENGTH (direct Web deployments), defaulting to the previous 128. Character validation and prototype-pollution protections are unchanged. (#41697) Bug Fixes Knowledge Base This release fixes a cluster of extraction bugs that changed document content without reporting an error. Re-import affected documents after upgrading; indexed text is not retroactively repaired. CSV cells were passed through pandas type inference before indexing. Leading-zero codes became floats (00123 → 123.0), empty cells became the literal string nan, and cells containing NA or NULL became missing values. The extract

From the release notes

[1.18.0] - 2026-09-10 Added samples: Add Magentic custom-manager prompts and MLflow observability examples (#7876, #8085) agent-framework-core: Add shared vector-store abstractions, portable filters, and an in-memory vector store (#8014, #8115) agent-framework-core: Add a maximum-duration bound and stop-reason signal to the tool invocation loop (#7772) agent-framework-core: Support mixed workflow invocation keyword arguments (#7963) agent-framework-ag-ui: Add emit_messages_snapshot configuration for suppressing the terminal message snapshot (#7808) agent-framework-ag-ui, agent-framework-core: Add supported MCP Host-history conversion for persisted AG-UI conversations (#8130) agent-framework-azure-ai-search, agent-framework-core: Add an Azure AI Search implementation of the shared vector-store APIs (#8153) agent-framework-redis, agent-framework-core: Add Redis HASH and JSON vector stores (#8156) agent-framework-qdrant: Add the alpha Qdrant vector-store connector (#8154) agent-framework-postgres: Add the alpha PostgreSQL/pgvector connector (#8155) Changed agent-framework, agent-framework-core, agent-framework-foundry, agent-framework-lab: [BREAKING] Isolate Lab dependency resolution, remove Lab from core[all], and support Azure AI Projects through 2.6 with OpenAI 3.x (#8188) agent-framework-core: [BREAKING - experimental] Add ranged file reads and move the line-numbering contract onto AgentFileStore (#7669) agent-framework-core: [BREAKING - experimental] Centralize file, memory, session, and skill path normalization (#8123) agent-framework-anthropic, agent-framework-azure-ai-search, agent-framework-azure-cosmos, agent-framework-bedrock, agent-framework-copilotstudio, agent-framework-core, agent-framework-foundry, agent-framework-gemini, agent-framework-mem0, agent-framework-openai: [BREAKING] Make SecretString a masked value wrapper rather than a str subclass and align provider credential handling (#8127) agent-framework-github-copilot: [BREAKING] Make workspace file hooks opt-in and document enable_file_hooks migration (#7517) agent-framework-foundry-hosting: [BREAKING - beta] Restrict checkpoint deserialization by default and allow applications to register additional checkpoint types (#8045) agent-framework-core, agent-framework-foundry: Move Foundry evaluation serialization into the Foundry integration (#8031) agent-framework-github-copilot: Update the GitHub Copilot SDK dependency to 1.0.11 and align file-operation behavior (#8002) agent-framework-ag-ui, agent-framework-devui: Expand compatibility through FastAPI 0.141 (#8052) agent-framework-core, agent-framework-hyperlight: Update Hyperlight Sandbox to 0.6 and enable supported Python 3.14 installations (#8174) agent-framework-foundry, agent-framework-foundry-hosting, agent-framework-openai: Lazily load Foundry and OpenAI integrations to avoid unnecessary import-time dependencies (#8219) agent-framework-devui: Update frontend transitive dependencies (#8019, #8009, #8175

From the release notes

0.8.0 (2026-09-10) Features core: support ordered cube queries (#2677) (8d23753)

From the release notes

What's Changed ci: cap build-test at 60 min and stream prek hook progress by @georgeh0 in #2378 ci(deps): bump taiki-e/install-action from 2.87.0 to 2.87.5 by @dependabot[bot] in #2384 fix(litellm): align embedding responses to inputs by index by @hardness1020 in #2381 fix(engine): skip child-existence reconciliation on demote-only deletes by @georgeh0 in #2379 fix(core): track component activity explicitly instead of by reference count by @georgeh0 in #2385 ci(deps): bump svgo from 4.0.2 to 4.1.0 in /docs by @dependabot[bot] in #2386 ci(deps): bump sharp from 0.35.3 to 0.35.4 in /docs by @dependabot[bot] in #2388 ci(deps): bump js-yaml from 4.3.1 to 4.3.2 in /docs by @dependabot[bot] in #2387 feat(litellm): bound concurrent embedding requests to the backend by @georgeh0 in #2392 fix(litellm): make embedding timeouts terminal, add timeout param by @georgeh0 in #2394 feat(litellm): retry transcription requests under one timeout by @georgeh0 in #2395 docs(litellm): one shared-behavior section for embedder and transcriber by @georgeh0 in #2396 Full Changelog: v1.0.21...v1.0.22

From the release notes

Signed-off-by: omnigent-ci[bot] <294685417+omnigent-ci[bot]@users.noreply.github.com>

From the release notes

0.4.0 (2026-09-09) Miscellaneous Chores wren-core-base: Synchronize wren-rust-core versions

From the release notes

0.4.0 (2026-09-09) Features core: support ordered cube queries (#2677) (8d23753) Bug Fixes core: reject collapsed relationship joins (#2697) (4000bea)

From the release notes

0.4.0 (2026-09-09) Miscellaneous Chores wren-manifest-macro: Synchronize wren-rust-core versions

From the release notes

🕸️ Cross-Host Federation + Claims Coordinate agents across machines with signed, verifiable messages and work claims. Two coordinators — use either alone or together: Local mesh (agentbbs): HTTP pull, Ed25519-signed, pinned keys. Works over Tailscale, LAN, VPN, or loopback — no Tailscale required. Slack: coordinate in a channel thread with structured messages. Zero infrastructure. Install npx ruflo@latest --version # 3.40.0 Quick start — join the mesh (run on each host) 1. federation_bbs_identity {} → your nodeId + publicKey 2. federation_bbs_peer_add { nodeId, publicKey, url } → pin every other node 3. federation_bbs_serve { bindHost:"<your routable IP>", port:7777 } 4. federation_bbs_register { roomLabel:"#coordination" } 5. federation_bbs_sync { roomId:"<from step 4>" } → on a 15–30s timer Unsigned, misattributed, oversize, or over-hop messages are dropped and counted — a hostile or looping peer can't corrupt the log. Claims — coordinate who owns what Runtime tools: claims_claim, claims_release, claims_handoff, claims_status. Cross-host, publish claim events: { "type":"ClaimIssued", "from":"nodeA", "resourceId":"deploy-api", "ttlSeconds":3600 } One owner per resource. Claim → wait for the coordinator's ClaimAck → then do the work. Security Every mesh message is Ed25519-verified against a pinned key. Registry-anchored pinning refuses any endpoint presenting a different key for a known node (identity-hijack / MITM protection). Private keys never leave the host. Don't put secrets in message payloads. What's also in 3.40.0 Restored memory search recall (a real regression: the default threshold made keyword recall impossible) — #3252 Full CLI test suite green (69 → 0 failures) 📖 Full step-by-step guide, protocol reference, GCP-secret bootstrap, and troubleshooting: https://gist.github.com/ruvnet/2fbad4d0e3bc02cc7d47b4f6fdd2c205 🤖 Generated with RuFlo

From the release notes

What's Changed Features Add telemetry to track checkpoint runtime and CLI usage. Bug Fixes Fix gateway errors reported inside an HTTP 200 response. Keep deploy push on the AMP create source. Report scrape failures instead of raising IndexError in the Oxylabs integration. Route all DashScope models through the native provider. Persist JSON checkpoints as UTF-8. Drop stray $ in f-string search URLs for BrightData. Support list-form type arrays in JSON schema conversion. Correct gpt-4o-mini context window from 200000 to 128000. Preserve streaming tool call arguments at contentBlockStop. Make pre-commit hooks portable on Windows. Documentation Clarify that tracing is managed separately from telemetry. Fix parallel search reference link. Remove stale parameters from handle_llm_stream_chunk docstring. Fix streaming output docstring examples. Use organization UUIDs in the skill install reference. Contributors @Alphaxiaoteng, @DrewWhittleNZ, @Ghraven, @Shxiao101, @Vaishnavi220506, @Vidit-Ostwal, @danielfsbarreto, @georgeatparallel, @github-actions[bot], @jessemiller, @joaomdmoura, @kimnamu, @kiwoongyoon, @liang0417, @lorenzejay, @oxy-giedrius, @simpleqt, @uoparaji

From the release notes

python/v1.55.1 Auto-drafted from commits in python/v1.55.0..python/v1.55.1, grouped by conventional-commit type. Edit on the release page after publish if you want a polished writeup; the canonical release notes live on the website. 🚀 Features feat(vended-tools): port notebook tool to Python (#4132) (545d9ed) feat(bidi): simplify Google and OpenAI model configs (#4189) (6d4f61f) feat(python): add backgroundTasks to Agent (#4226) (42e5a61) 🐛 Fixes fix(session): filter malformed immutable snapshot IDs (#4199) (6d6c8fc) fix(telemetry): use semantic system instructions for agent spans (#4222) (e5ebbd6) fix(models): resolve nested Bedrock model prefixes (#4221) (21b27d4) fix(context): fix various context manager parity items (#4228) (bebf3e8) fix(openai): emit tool results before user text to preserve tool_use/tool_result adjacency (#4234) (5f636bb) 📚 Documentation docs(changelog): sync strands-agents/harness-sdk python/v1.55.0 (#4238) (57d0705) docs(changelog): sync strands-agents/harness-sdk typescript/v1.17.0 (#4239) (1317a43)

From the release notes

Changes since langchain-openai==1.6.1 release(openai): 1.6.2 (#40339) fix(openai): add GPT-6 Astra reasoning efforts (#40330) chore(deps): bump httpx2 from 2.10.0 to 2.12.0 in /libs/partners/openai (#40309)

From the release notes

This release converges React Native's render-tool hooks onto react-core, improves Copilot context timing during page navigation, and adds provider-level agent configuration. It also includes fixes for nullable tool schemas, the Vue human-in-the-loop lifecycle, and the Inspector experience.

From the release notes

Patch release shipping #3250. Fixed federation_bbs_* tools were unconditionally degraded. All four tools (register / publish / watch / human_join) gated on await import('agentbbs'). The published agentbbs package is a CLI-only launcher — its tarball contains only bin/agentbbs.js, package.json and README.md, with no main, no exports, and no root index.js for Node to fall back to. That import could never resolve, so the tools reported degraded: true no matter whether agentbbs was installed and working. Replaced with a subprocess presence probe. Probe shell gated to win32. The follow-up hardening: shell: true was unconditional while the probed binary comes from AGENTBBS_BIN, and under a shell that argument is command-interpreted — a crafted value ran a trailing command (verified locally). POSIX now spawns without a shell; Windows keeps cmd.exe so the agentbbs.cmd shim still resolves via PATHEXT. Matches the existing convention in browser-tools.ts, commands/init.ts and init/helpers-generator.ts. @noble/ed25519 no longer externalized in tests. It is a hard dependency of @claude-flow/cli, not an optional one, and externalizing it broke dynamic-import resolution under Vite's SSR transform. Scope This makes the Phase 1 local room/envelope surface functional when agentbbs is genuinely installed. It does not implement cross-host federation — that remains Phase 2+. Install npx ruflo@latest All three packages (@claude-flow/cli, claude-flow, ruflo) are at 3.39.3 on latest, alpha, and v3alpha. 🤖 Generated with RuFlo https://claude.ai/code/session_013u4pmL9ZUAXb6usVQgNo67

From the release notes

Fix Space image previews and simplify file deletion (#1009) * fix(spaces): restore image attachment lightbox and carousel * fix(spaces): make image preview actions clickable below titlebar * fix(spaces): simplify move to trash confirmation

From the release notes

Changelog b98d421 feat: add APIMart model provider 8dc8a3f feat: refresh Qwen models to the current Bailian lineup

From the release notes

What changed since August 12 MCP server works again. agent-harnesses-mcp 0.5.2 pins the MCP SDK below 2.0. The 2.0 release (July 28) renamed FastMCP, and 0.5.1 crashed on startup for anyone who installed it. Install: claude mcp add agent-harnesses -- uvx agent-harnesses-mcp. Cold-tested: initialize and tools/list complete, 10 tools. 164 harnesses, 12 categories. New in the tables: YYLO (coding agent products), AnythingLLM (personal agent runtimes), ClawBench (evaluation and benchmarking). Flowise stays listed with its archived flag instead of moving to the Graveyard. Roo Code is out of the turnkey coding agent pick and stays in the tables with its archived flag. On the radar. DeepSeek Harness (pinned, not ranked, pending a star-velocity check on 2026-10-07), Google Antigravity SDK, L∞pGate, and eight community submissions below the ranked-table bar. Community submissions now reach the curation queue. Open "Add project" issues feed the weekly discovery run, so the biweekly curation pass vets them with live repo data and closes the issue when an entry lands. Site. The nav shows a live star count that links back to the repo. Weekly rescores on August 16, 23, 30, September 6 and 9 refreshed every star count and the landscape charts. Thanks to contributors HelpMatey (#61), InsightFactoryAPP (#111), reacher-z (#60), rxdt (#69), and danawoodman (#78).