Skip to content
Agents tracked: 492 Releases (7d): 545 Agent status: 1 with issuesSince last limit reset: Claude 9d · Codex 7d For AI agents: Markdown Updated Nov 20, 2026

AI agent releases: Oct 26 – Nov 1, 2026

628 stable releases from 145 tracked agents and frameworks in the week of Oct 26, 2026, newest first, with the headline changes from each set of release notes. Our lists keep each project's latest releases, so a very active project's older releases can be missing.

Nov 1, 2026

Loki Mode
  • A run that does not publish a PR keeps its loki/<run> branch whatever its outcome, so blocked or failed work is never thrown away; your checkout still returns to the branch you started on.
  • When only Loki's own acceptance tests conflict with the change, the work is delivered as PARTIAL only if the repo's own tests (a test check that ran at least one test, or the full suite) run and…
  • Loki's own acceptance tests that cannot compile are discarded before they reach your tree: a Go test file whose package does not match its directory, or a Rust call whose argument count matches no…
  • feat(i18n): Add Russian localization (ru)
  • chore(lint): turn noExplicitAny and noNonNullAssertion into errors
  • Blog: GUI overview
  • Extracted bundled tools into standalone plugins: wait, notify_user, search_engine, a2a_chat, scheduler, behavior_adjustment, and the scheduler Tasks UI are now standard plugins, each with…
  • Unified memory tools: The four separate memory_load, memory_save, memory_delete, and memory_forget tools are consolidated into a single memory tool with an action argument, matching the…
  • Per-action tool permissions: Profiles can now allow or deny individual operations across all 12 bundled multi-action tools…
AgentOS
  • name the code a host reads on a configuration refusal, and say a blank persona id is refused (2ed2dec)
Traycer
  • feat: inline agent pages, MCP Apps, and shared task files (885fd24)
  • fix(desktop): keep a support report's feedback event under Sentry's size limit (3cc2ccf)
  • feat(gui-app): keep late dictation finals, warm the recogniser from a focused composer, and name the model set (10e345a)
  • Developer commands resolve the acting user and selected project consistently. dev connected-accounts link now uses the developer project, while root link keeps its consumer-project behavior.
  • Normal commands interrupted by SIGINT, SIGTERM, or SIGHUP now exit 130, 143, or 129. Subscribed listeners can stop cleanly on Ctrl+C or SIGTERM.
  • listen and dev triggers listen support --fail-fast for sink failures. Toolkit filters correctly match slugs containing underscores.
  • Row-level security's system bypass now needs a key that only the app holds, so SQL running for one Space can no longer give itself access to every Space.
  • Hosts hardened with harden-host.sh now drop inbound IPv4 TCP packets that carry IP options and turn off multipath TCP, and production Caddy no longer listens with multipath TCP.
  • Dropping a connection partway through a WebSocket upgrade no longer crashes the server.
  • Add standalone LiteLLM catalog PR before version bump
  • Record where crewai run stops before kickoff
  • Add crewai skill search to list the organization's skills
  • Decision-model category scores for every answered check.
  • Project aliases routed by decision-model tags.
  • The tag-rules router policy authoring command.
Sim
  • feat(billing): require a paid plan or prepaid credits for Chat
  • fix(billing): fail closed on legacy Chat payer reads and cover the Chat plan gate
  • fix(public-shares): stream shared files with byte ranges instead of buffering them under 100 MB
HyperFrames
  • CLI: Keep browser sign-in open through checkout (34ddced2d, #6879).
  • Studio: A clip's thumbnail strip stays put while you trim it (0ab493115, #6942).
  • CLI: A stale mask takes its frame's colours instead of delogo's streaks (6b0abb4eb, #6950).
Ruflo
  • contextDelegation mod option (off | observe | enforce). ruflo-mods automates ADR-521
  • What qualifies: a main-loop, whole-file Read of a 14–120 KB file, for a prompt that routes to research or review and asks for no change.
  • What happens: it is answered by a no-tools haiku completion on the session's own credentials, so no API key is needed.
Claude Mem
  • The repo root drops from 60 entries to 36, so the README sits much higher on the GitHub page.
  • Old plans, reports, playbooks and generated files are gone, along with the superseded cursor-hooks/ and ragtime/ directories.
  • Docs, test assets and Docker files moved into docs/, tests/, .github/, .claude/ and docker/. The changelog now lives at docs/CHANGELOG.md, and the security policy at .github/SECURITY.md.
Mika
  • Mika-v0.1.33-arm64.apk(约 79 MB):近几年的安卓手机基本都选这个,更小。
  • Mika-v0.1.33.apk(约 92 MB):通用包,所有手机都能装。不确定就选它。
  • 两个包功能完全一样,可以互相覆盖安装;App 里「关于 › 检查更新」会自动选合适的那个。
  • Google Drive: files with the same name, even in one folder. Items are now keyed by file ID
  • S3, Azure Blob and OCI: object keys that differ only in a doubled or leading /, a ./ segment, or \ versus /; and, with prefix="docs", docs/a.md and docsa.md.
  • Kafka: the same message key on two topics of one topic_as_map feed. A tombstone on one topic deleted the other topic's outputs
Kortix
  • After an automatic compaction, the summary shows once, inside the "Context automatically compacted" marker, instead of a second time below it.
  • A worker service that scales up from zero waits for its previous deployment to leave before it rolls, so deploys no longer collide.
  • The macOS desktop installer builds on macOS 15 runners, so a release no longer loses it during GitHub's macOS 14 retirement.
  • Honor hosted base CI before starting work or merging, and pause merges after a factory merge leaves the base red.
  • Stack mode routes planner requests to split an oversized issue through the slice gate; missing information and contradictions still park.
BrowserSkill
  • Shared-control default: new CLI and DSH sync calls select target policy.
  • CLI/daemon and extension now use protocol 1.8. The new target policy requires
  • DOM references remain attached to the same nodes across supplementary
  • test(host-service): fold git-status-store adversarial suites into regular tests
  • feat(desktop, chat): per-turn changes card, turn snapshots and Undo (SUPER-2736)
  • fix(host-service): bound the git status cache
  • Download the provider that runs a saved custom endpoint
  • Index remote sessions for the session sweep and membership cascade
  • Show the OpenCode Go monthly limit as a third usage bar
  • Fixed cookie attributes so cookies set honors --http-only and --same-site, preserves the existing aliases, and rejects unknown arguments
  • Fixed page error clearing so errors --clear removes captured errors
  • Fixed MCP selector waits so bare integer selectors return an error instead of becoming timed waits
  • Renamed to DS Bot. The desktop app and its downloads are now called DS Bot; your team, chats, settings and browser data carry over when you update.
  • Group chats: no relay limit; @ calls in one message start all the named members at once, and the calling Bot waits until all of them finish.
  • How Bots work: Bots hand parallel work to other Bots instead of starting DSH subagents. In a private chat, a message you send while the Bot is working reaches it right away.
  • Fixed OpenAI remote compaction (/responses/compact) replaying backend-only internal_chat_message_metadata_passthrough metadata that Codex rejects as an unknown parameter
  • Fixed Codex conversation replay failing with an unknown internal_chat_message_metadata_passthrough.content_token_count parameter by omitting backend-internal response metadata from replayed input
  • Fixed service_tier: "ultrafast" being dropped for OpenAI-family models on custom providers even when the model's serviceTiers lists it
  • Validate the maintained Codex 0.159.0 compatibility profile for Open
  • Preserve legacy feedback delivery alongside optional signed attachment
  • 不打扰模式:「Shizuku 可用」改为看接口本身,不再要求装 Shizuku App——只装了 Sui(Magisk / KernelSU 模块)的设备也能直接用,状态里写明接口来自 Sui 还是 Shizuku;两者都没有、和有 Sui 但还没授权,提示分开说。
  • 不打扰模式:新增 root 通道,给已 root(Magisk / KernelSU / APatch)但没有 Shizuku 的设备。只有你在页面上选了 root 才会第一次执行 su,授权走 root 管理器自己的弹窗;「自动恢复无障碍」的一键授予也能走 root。
  • 不打扰模式:通道选择改成一个面板——关闭 / Shizuku / Sui / root / 自连 ADB,各自写明适合谁、现在什么状态,并标出推荐的那个(只推荐,不替你选);失败时显示真实原因,不会悄悄换到别的通道。Sui 和 root 两条通道还没在真机上验证过,遇到问题请开 issue 反馈。
  • [PR] test(e2e): fork-adoption relay echoes plain text, not rendered ref tags — unblocks v0.1.210 release
  • [PR] docs: New blog post — What to do when billion-context is frequently compressed? | docs: 新博客文章——billion-context频繁压缩该怎么办?
  • [PR] fix: frozen compression state becomes visible, diagnosable, and healable
  • npm: bash npm install -g base44@latest
  • Homebrew: bash brew install base44/tap/base44
  • Version: 0.2.7 NPM Tag: latest
  • 角色资产支持上传、试听、替换和移除音色样本。
  • 分镜视频提示词支持在对白处使用 #角色名 引用音色,例如「0-3秒:@林小雨抬头,#林小雨轻声说:「早上好。」」。
  • 首次分镜拆解、单条和批量视频提示词生成会依据角色绑定和可用音色生成引用。
  • Blank window fixed. After updating to 0.1.136, a window whose Cloud hadn't updated yet could show only the rail of places and an empty page.
  • Updates: a failed update check counts once, and its notice stays up during the quiet hourly checks.
  • Goals: a one-on-one goal whose turn keeps losing its slot now stops as blocked instead of retrying forever.
Qwen Code
  • feat(managed): Add bounded SQL history queries
  • feat(generated-ui): keep chatting with a card's values and turn Generated UI on by default
  • test(web-shell): follow the renamed workspace copy in two e2e specs
  • 适配 DSH 0.2.1-rc.5,0.2.1-rc.4 列为兼容线 — v0.17.0 只声明支持到 0.2.1-rc.3;npm 上的 DSH 升到 rc.4 / rc.5 后,DSH 的插件兼容检查会拒绝加载它,表现为升级 DSH 后 TUI 打不开。本版把 rc.4 / rc.5 加入验证线与 peer 范围,dst 自动化适配器的验证版本同步加入。按发布包逐文件比对,TUI…
  • 首次启动引导可以直接登录或加 key — 引导第一步现在能登录订阅账号、添加模型提供方(API key)、重新检查或跳过这一步;订阅登录带回的新模型按 0/1/N 处理:只报结果、直接设为默认,或在内联选择器里挑一个。「添加模型提供方」沿用 /provider 向导,向导的切换题会逐项列出本次启用的模型,并保留「保持当前模型」。
  • 官方 Git 工作树实验包 — /plugin list 里可以直接开启 DSH 官方的 create_worktree;工具卡标题显示分支名,结果用两行写明检出位置和起点。Minimal 与随包的 liangshen 预设不提供这个工具:Minimal 的持久 shell 不跟随目录,liangshen 为跨平台统一规则一并隐藏。工作树不会自动清理,分支需要另外删除。
No release notes published.
Reasonix
  • CLI archives built from the Studio release studio-v2.56.1.
  • Zoom into shown files on the card. In the Shown viewer, pinch on a phone, or use Ctrl or Cmd with the scroll wheel or a trackpad pinch on desktop, to zoom an image, video or HTML page up to 8x…
  • A new mark, and status icons that move. Glimmervoid's mark is now four corner brackets around a small diamond light. It replaces the play triangle in the favicon, the site nav and the loading screen.
  • Desktop tabs show the same icons as the phone. Every view tab in the desktop header now leads with the line icon the phone bar and More menu use, so Focus, Boards, Settings and the rest look the…
  • fix(grafana): cap Loki query requests at server limit (579006aca) — Sri Vishnu
  • Fix: ids a type and a function whose names differ only in case (Foo the class, foo the function) get distinct node ids instead of colliding (which dropped one node and minted a bogus self-loop)
  • Fix: php a Scope::MEMBER access in a class-level initializer (a const value, a property default, or a parameter/promoted-constructor default) emits a references_constant edge.
  • Fix: hooks git rev-parse output is decoded as UTF-8, so a linked worktree under a non-ASCII path resolves correctly on Windows instead of mojibake
  • Interface follows your system language on first start. With no language saved yet, SolidCraft starts in the system's language: the OS locale on the desktop and the browser's languages in the web…
  • Sketch Offset keeps its corners joined. Offsetting a chain whose corners only meet in position, such as separate lines and arcs held by Coincident constraints or ends within 0.0001 mm, offsets it…
  • Joints under each component, and show or hide many Browser rows in one drag. Each component placement in the Browser has a Joints folder listing the joints that attach it, with the placement each…

Oct 31, 2026

OpenMausBot
  • What's new, in one popup. Once after this update: the new MausBot Cloud, and Team mode's shared bots. Each card has a button to try it.
  • Share a bot with your team, bot by bot. From a bot's Share button, choose who can view it, chat with it or edit it. A teammate without access can ask for it in one tap.
  • Everything in your language. Run and routine notices, connect prompts, notifications, room lines, and teammates' hand-offs and results now come in each person's language.
Ruflo
  • Prompt-injection detection runs the AIMDS pattern packs (@claude-flow/aidefence 3.1.0, ADR-517).
  • aidefence restored to source (3.0.4 → 3.1.0, ADR-516). npm latest 3.0.3 carried ADR-118 detection that was never committed; it is ported back alongside main's #3177 ReDoS bounds.
  • A fresh install's dependency tree converges (ADR-513). agentdb 3.0.0 is stable, so the optional peer ranges that excluded 3.0.0-alpha.20 now match it.
AgentOS
  • library: a label's end separators, a sigma's place in its run, and a key for every tag (d45850c)
HyperFrames
  • Engine: Nested video clips on a frame boundary export every source frame (47fe68562, #6893).
  • CLI: Billing top-up opens the pricing page (4d44c0d0a, #6887).
  • CLI: Fill --mask boxes from the picture behind and crop generated clips to the requested aspect (919e138fc, #6892).
Mika
  • 网页可以在任务结束后接着运行。 说「播放,然后保持这页别关」「挂着」这类话时,Mika 会把这一页留在后台接着运行,回答下面、悬浮胶囊和通知栏都看得到、一键能关;最多留一页,2 小时后自动关。没让留的照旧随任务关掉。
  • 让 Mika 建技能,真的能存下。 说「帮我建一个技能:……」,或者把别处写好的技能贴给它,Mika 会写好存成你的个人技能,马上出现在「设置 › 记忆与技能」(标着「按你的要求写下」)。改已有技能也行。没做成的事,Mika 不会再说「已经做好了」。
  • 任务中说的话可以撤回、编辑。 已发出、模型还没处理的那句,长按可以撤回或取回修改。
  • build(nix): install OpenBot 0.56.0
  • Show notification levels as one card for each server
  • Let the agent answer page dialogs that its action opened
  • Skill groups: with 30 or 50 skills, the resume's Skills section was one long paragraph.
  • Jev by TypeSafe for scoring: the new Settings → Scoring section chooses what Score in Find jobs and the job agent use: your AI model, as before, or Jev, TypeSafe's model for typed judgements, with…
  • The MCP connector, for other apps too: Settings → Claude connector is now Settings → MCP connector.
  • Choose Files or Folder from Import notes to import multiple files through a continuous queue, with per-file results and support for adding more files while imports are running.
  • Import PDFs with selectable text alongside Markdown and text files, up to 25 MB per file. PDF text conversion also gives the tutor's PDF reader more readable words, lines, and prose-column ordering.
  • Organize sources in persistent folders: rename sources and folders, and drag them between folders.
Oh My Pi
  • The soft steering signal a tool sees as toolCall.steeringSignal now aborts with the source of the queued message ("user", "irc", …) as its reason, so a tool can yield only to the sources it should
  • Reject incomplete local compaction, turn-prefix, handoff, and branch summaries ending with stopReason: "length" instead of using truncated text as complete context.
  • Replaced isLoopGuardedModel with loopGuardLevel, which returns "off", "exact", or "semantic" from the model's thinking-loop-guard catalog rule
  • [PR] release v0.1.209
  • [PR] ci: 恢复紧急快速发布入口(fast 模式加双信号防误触门)
  • [PR] Update QQ Group information in README
  • Godot 4.2+
  • Active Ziva account
  • [Agent Files] Comment on files for the agent. In a text file, hover a line and click '+' (or drag across several lines) to write a comment.
  • [Updates] View skipped stable-version release notes in a collapsible timeline; single-version updates show their notes directly. Markdown footnote navigation now works correctly with backlinks.
  • [Proxy] Test proxy connection in General Settings before saving, with support for PAC proxy/direct fallback reporting.
  • Validate the maintained Codex 0.159.0 compatibility profile for Open
  • 💻 Windows:双击下载的安装包,按提示完成安装。
  • 🍎 Mac:打开下载的安装包,将 toonflow.app 拖入“应用程序”,再从“应用程序”打开。
  • 📱 Android:在手机上打开 APK,按系统提示允许此来源安装应用。后续版本可直接覆盖安装。
dsh-Go
  • 排障:引擎起来后,App 会把一份脱敏的本机诊断写进引擎($DSH_HOME/dsh-app/diagnostics.md):版本、设备、运行时、启动方式、连接、性能档位、最近错误和日志尾部。新增内置技能 dsh-go-support,教 AI 先读它再排查,并按症状给出可直接粘贴的 issue 草稿。
  • 设置 → 诊断新增「让 AI 看诊断」:手动刷新这份快照,然后问 AI「看下诊断」就行。
  • 更新:修了「已经是 1.0.4 却提示最新」——某个更新源停在旧版本时,不再就此收工,会继续问其它源,取更新的那个;公测包也不再带内测通道令牌,免得一个过期的内测清单盖住 GitHub 的新版本。
  • Plugins: plugin providers now use the full menu-bar layout (inheriting your global layout, with countdowns and conditionals), can show compact multi-quota summaries and a built-in provider logo…
  • Codex: cost history survives Codex archiving old sessions as .jsonl.zst and switching between Codex homes, GPT-6.1 Sol Fast requests are priced at Fast rates, accounts can have custom names, and…
  • Providers: new OpenDesign subscription quotas; Kiro works with Kiro CLI 3; Perplexity organization credits, WorkBuddy reward credits, LiteLLM team budgets, MiniMax China, Copilot monthly pace and…
  • Team Organization for multi-expert-team collaboration, enabling multiple long-lived professional Teams to coordinate on complex cross-team tasks.
  • Scheduled dispatch mode for Agent Teams, introducing verifier/challenger review roles with binary voting, traceable review, and timeout escalation.
  • Swarmflow execution engine enhancements for Design mode, including expert routing for the canvas-based multimodal creation pipeline.
  • Nothing to do. The browser extension updates itself once no chat is busy.
  • Message times. Settings → General → Show message times shows a small time above each message. Point at it for the full date. It is off unless you turn it on.
  • The log says why a message waits. When a ChatGPT page holds a queued message back, the app log now says why: an answer or a tool still running, a covered message box, or a question the page can't…
Sim
  • feat(favicons): one server route for every site icon, with parent-domain fallback
  • improvement(favicons): resolve platform tenants no provider can see, retry timeout-derived icons, no miss flash
  • improvement(docs): simplify workflow lightbox and retire migration artifacts
  • Cursor keeps queued conversations when login or usage limits interrupt saving.
  • Claude Code and Codex retain those conversations until access returns.
  • Saved conversations become available after the agent can process them again.
  • The golden image of the inputs 9c2b3c5b93ef, built by the golden image workflow from ef69f5e19c1b46e41e743234fa5019e39442adcb.
  • cc-haha 手机 App(#1688,#1698,#1710,#1715,#1724,#1725,#1727,#1729,#1738,#1741,#1743):在桌面「手机连接」里扫码配对自己的电脑后,可以在手机上看流式回复、审批工具权限和回答提问、发新任务、创建定时任务、浏览文件和改动。语音输入由电脑上已配置的语音服务转写。Agent…
  • 手机连接改用 Cloudflare(#1735,#1740,#1742,#1728):默认的「快速连接」不需要 Cloudflare 账号,自动生成二维码,手机扫码后在桌面上批准。也可以用自己的固定域名:直接粘贴 Cloudflare 给出的安装命令,只提取其中的 Token,不执行命令;Token 被拒或路由配错时直接给出原因。同一 Wi-Fi…
  • 运行任务时防止系统休眠(#1730):在 设置 → 通用 中开启后,有任务在跑时电脑不会休眠,屏幕仍可以关闭。默认关闭。
  • [Collection: SQL] Fixed the SQL collection preview crashing when a query returns JSON or JSONB objects and arrays
  • [Record history] Fixed record history blocks still requesting history data after form submit when the block is hidden or the current role has no permission to view it
  • [Data source: External Oracle] Fixed Oracle external data sources failing to create records when a table primary key is defined as an IDENTITY column, because the primary key and auto-increment…
  • experiential 0.7.225 (exp-gateway-native unchanged at 0.3.173). Adds GatewayRungDispatchPolicy.access_programs: a deployment-authored OpenAI access-program opt-in injected into the Responses…
  • Updated bundled Claude CLI to version 2.1.316
  • Print face: pick the face a part sits on, and Send to Slicer uses it. Mark a flat face as a body's print face with Ctrl+B, Print Face on the face's context menu, or the Print face input in the…
  • The toolbar shows which analyses are toggled on. A surface analysis that is on shows as pressed on the ribbon, and a status item above the view names it with its setting, its legend and an × to…
  • Point clouds: color and filter by any property in the scan. A point cloud keeps every property its PLY file gives its points.
  • Added CLAUDE_CODE_DISABLE_ULTRATHINK, which turns the ultrathink keyword off
  • Added pooled group budgets to the Claude apps gateway: one monthly budget that a group's members share, on top of each member's own limit, set with pooled_budget on the spend limits admin API
  • Added CLAUDE_CODE_DISABLE_LOOPED_CROSS_CERT_DROP to keep system cross-certificates that certify each other in a loop, for servers whose private CA needs one
  • fix(llm): align deployment token contracts across setup and transports (ffe05112c) — Anwesh
  • fix(llm): stop terminal LiteLLM retries and refresh the safe patch lock (89a6611e6) — Anwesh
  • fix(llm): refresh retired model defaults and preserve provider model IDs (278d0cc27) — Anwesh

Oct 30, 2026

Graphify
  • Feat: path when several routes tie on length, graphify path and the MCP shortest_path tool report the most conservative one: among all shortest paths, the route whose weakest hop is weakest is…
  • Fix: go a call in a package-level var initialiser (including inside a func-literal closure, the common swappable-hook pattern) is credited to the file node instead of being dropped
  • Fix: go a method value passed as a function argument (Wrap(s.Health)) is linked as an indirect_call, bringing Go to parity with Python/JS for callback-by-method-value.
  • Raise an error status a retrying transport gave up on as ModelHTTPError, so fallback moves on
  • Send stored response, item, conversation and container IDs back only to the endpoint that created them, via Model.accepts_server_state
  • Make json-repair an optional dependency, needed only by Coder and RepairToolArguments
OpenCode
  • Worktree and Shell Mode Shortcuts: Press ⌘⌥O on macOS or Ctrl+Alt+O on Windows and Linux to choose a worktree, and ⌘⌥B or Ctrl+Alt+B to choose its base branch.
  • Send Quotes From the Comment Popup: Hold ⌘ or Ctrl in the quote popup to turn its checkmark into send, or press ⌘Enter or Ctrl+Enter, to send just that quote and comment while leaving the rest of…
  • Percentage Compaction Buffer: Set "compaction": { "buffer": "20%" } in opencode.json to keep a share of each model's context window free, so one setting fits models with small and large windows.
  • Route a channel with many long member descriptions
  • Say that the OpenBot source code is open, not that OpenBot is open source
  • Explain why a Workspace only Claude agent cannot start its sandbox
  • release(langgraph): 1.2.19
  • release(checkpoint-sqlite): 3.1.3
  • fix(langgraph): keep trace_policy from rewriting streamed node payloads
  • macOS (-sea assets): unsigned (Apple secrets not configured)
  • Linux (-sea assets): SHA256SUMS-sea published with the release
AI SDK
  • Updated dependencies [cbc9ac0]
  • @ai-sdk/gateway@4.0.132
  • agent-framework-core: Add an explicit default approval mode for local MCP tools and prompts outside the per-tool approval lists, and warn when never_require_approval leaves the default implicit…
  • agent-framework-core: Add the shared microsoft.agent-framework telemetry resource attribute across logs, traces, and metrics while preserving explicit resource overrides
  • agent-framework-core: Add experimental agent-triggered in-run compaction and skill unloading with AgentTriggeredCompactionStrategy and UnloadSkillsCompactionStrategy
Antigravity CLI
  • Added "optional": true for API key providers in a plugin's auth.json, for services that work without a key but raise rate limits with one: when you haven't set the key, the plugin's commands now…
  • Fixed Markdown tables in long streaming replies leaving misaligned borders and repeated text in your terminal scrollback: a table's rows now stay in the live view until the table is complete, so…
Mika
  • 插上数据线也能用。 新增「用电脑启动」:手机连上电脑后运行一条命令就能启动 Mika 的手机助手,不用配对无线调试。适合配对一直不成功的手机;重启手机后要再运行一次,或者再配对无线调试让它自动恢复。
  • 让电脑上的 AI 帮你装。 README 顶部新增「安装」:把一句话发给电脑上能执行命令的 AI 助手,它会照着安装文档检查手机、下载安装、打开需要的权限;API Key 由你自己在 Mika 里填。
  • 操作更像人,减少被 App 当成机器人。 打字改成一段一段地输入,动作之间有长短不一的停顿;小红书、微信、抖音等容易风控的 App 节奏更慢,一个任务里浏览的屏数有上限。设置 › 干活时 › 操作节奏:快 / 标准(默认)/ 像人。
  • Dylan Duan (@Kiddo4993), for the command-prompt updates and selection-contract tests in PR #351.
  • briefing: dismiss a Right now point and drop mail whose sender you wrote to (24e577b)
  • chat: attach, split and save any file by one handle (6cd7d4c)
  • chat: search the speaker's own earlier chats before looking things up again (1ff8263)
dsh-Go
  • 正式下载统一为 arm64 安装包,发布页明确标出安装包和辅助更新文件的用途。
  • DSH Remote 默认自动选择连接方式,优先尝试局域网 / P2P,再尝试 TURN,最后回落 Relay。升级后不再需要打开直连开关,旧的关闭设置也会迁移到自动策略;相应路径需服务端与主机支持。
  • 新增网络详情:点击连接卡或主机旁的网络标签,查看实际路径、连接耗时、可测的往返延迟及回落原因。重连后清除旧测量,改善 Wi-Fi 与蜂窝网络切换时的连接恢复。
Strands Agents
  • feat(models): default Bedrock maxTokens to the model info output limit (Python + TypeScript) (0a50a80ad)
  • feat(models): add Model.model_info with capabilities and limits (Python) (bcd704b4b)
  • fix(python): summarizer fallback truncates the largest text blocks first, not every oversized block (930fcadac)
Omnara
  • feat: support multiple organization owners
  • feat(machinepool): add Agent37 sandbox provider
  • feat: process notifications
Ruflo
  • http_fetch SSRF via DNS (CWE-918, CWE-350).
  • Before: the MCP tool refused loopback, RFC-1918 and link-local hosts only by the URL's hostname text, then let fetch() resolve the name itself.
  • Now: every connection resolves through a lookup hook. The hook refuses the host if any resolved address is private, and the socket connects only to addresses it checked.
Glimmervoid
  • Dashboard updates for npm installs: if you installed Glimmervoid with npm install -g on macOS or Linux, Settings > Updates now downloads a new version in the background on your channel…
  • glimmervoid doctor exits 1 when node-pty fails to load, so scripts and the dashboard updater can tell a broken install from a working one.
  • Board wires run at right angles through channels between columns, not as curves. Wires of one style into the same side of a box merge into one trunk with a single arrowhead.
  • A refused answer now ends that refresh on the agent that gave it. The next scheduled run tries the same agent again, with the existing failure backoff.
  • Feedback button in the My Work header, next to Star. It opens a short "My Work feedback" issue form.
  • Activity-driven background rebuilds of My Work are coalesced, so an open My Work page no longer rebuilds on every database commit (#5133, mitigates #5124).
  • [Diagrams] Mermaid diagrams now render with a faster engine in the background, match the app's light/dark theme, and open as previews from .mmd/.mermaid files.
  • [Agent Files] Open files in tabs with a breadcrumb, resize the tree, and configure autosave, indent, and word wrap in pane settings. Right-click to attach files or folders as context.
  • [Agents] Archive all conversations of an Agent in one action, with a confirmation showing the count. Delete conversation branches from the Branch Management panel.
  • session.clear() now cancels a running turn and every working task before it clears history, instead of waiting for the turn to finish.
  • Hooks can start a session's next turn with ctx.continue() from turn.settled and * handlers.
  • The bash tool now keeps only the last 50 KiB of each output stream while a command runs, so a verbose command no longer holds its whole stdout and stderr in memory.
  • #1437 94e6e76 Thanks @okwasniewski! - Breaking: a secret fill no longer turns off screenshots for the model.
  • #1282 7518ae5 Thanks @MarcosCamara01! - Add drag_at to the agent's pixel tools, in agent.act and e2e mcp, and dragAt(from, to) to ExecutorActions: a drag from one screenshot point to another, for…
  • #1447 82d5095 Thanks @21prnv! - A replay no longer fails its end check on an alert that shows an ISO 8601 timestamp, such as Saved at 2026-10-24T15:41:27.955Z, when the time differs from the…
Kortix
  • Discuss documents and comment on selected content in artifacts (8f1877d).
  • Load models from the live catalog instead of a baked sandbox list, without delaying boot or prompts.
  • Continue older OpenCode chats on pi and simplify the daemon to Kortix names.
Noetic
  • ci: publish prereleases from next to the next dist-tag (0b247bc), closes
  • Docs: Track Changes now follows Word end to end — a Reviewing Pane, Show Markup filters and balloon modes, per-author colors, a Track Changes Options dialog, moves recorded as move pairs, and…
  • Sheets: spelling check for workbooks, What-If data tables, combo-chart Select Data, PivotCharts, x14 data validation, persisted table filters, the remaining Show Values As modes and passwords on…
  • Slides: JPG page export (@tokg-venv); slide edits stop re-sending inline media and full decks to the AI (@HThanh-how).
  • feat(indexeddb): replace JSON storage with SQLite #2141
  • refactor(indexeddb): use Tokio owners and defer result loading #2153
  • feat(indexeddb): add transactional SQLite schema renames #2156
  • Sessions side by side — Show up to three Sessions next to each other in a Space, or open a Session in its own window or browser tab.
  • Session menu — Click the Session title to pin, rename, end or delete it, or turn the Session into a skill or an Automation.
  • Message actions — Copy a message, edit and resend it, or expand a long one. A divider shows when the model changes between tasks.
Opik
  • [issue-8590] [SDK] fix: accept repeated identical JSON objects in TS LLM-judge parser
  • [OPIK-9051] [BE] fix: a thread search picks threads and keeps all their traces
  • Consistent Chromium identity, and Chrome mode for agents in the user's own Chrome
  • Let the new chat page take pasted images and files
  • Clearer Connect mobile pairing panel
  • One stack engine for run-issue and run-brief (--size-gate takes off or stack, file and slice are deprecated aliases for stack, and --run-children is removed), PR bodies whose risk receipt no…
  • All seven published @on-par packages move to 0.9.25 in lockstep.
  • bash npm install -g @on-par/factory-cli@latest
Trueforge
  • 48efe38: Show context compaction in chat agent steps as Summarizing/Summarized context with expandable event JSON, and mark Summary on the session event timeline.
  • 00f5d15: Close the environment delete confirm dialog before the request so error toasts stay visible above the modal.
  • cc28923: Stamp tfg.source: ui metadata on UI-created sessions and filter Chat History on it so scheduled sessions are hidden by default.
  • feynman mcp adds MCP servers. Pi has built-in MCP support, but its docs say pi mcp add, and feynman mcp list used to be sent to the model as a chat question.
  • /release-notes (alias /whatsnew) browses the release notes of every Vibe version in a searchable, full-screen view.
  • Provider option preserve_api_base = true keeps a custom api_base that /whoami would otherwise overwrite (split-horizon / proxied deployments)
  • VIBE_TERMINAL_KEY names the terminal --continue uses; an empty value makes it ignore the terminal
  • A new notebook already has one thing in it. Every evening at 21:00 it looks back over the day and
  • The Mac has the CodeFlying market. market_search and market_call now exist on the Mac as they
  • A new face for the default Andee. Periwinkle body with a peach rim light, bigger and darker eyes
  • #4553 9140ea0 Thanks @RealKai42! - A single file read by the agent now returns at most 100,000 characters or about 25,000 tokens, whichever comes first, instead of up to 500,000 characters.
  • #4521 dbb65af Thanks @7Sageer! - Shorten built-in tool and subagent prompts.
  • #4521 dbb65af Thanks @7Sageer! - Shorten repeated goal and plan mode reminders.
  • 890628511e423864d21c694231277805ce9f113c MUL-7132: fix(issues): resolve comment trigger avatar URLs
  • f2a6398d6ea742cc10c96a8d460b4ecc8d95db2a MUL-7257: refactor(server): remove unused compatibility queries and capabilities
  • e7dfbe83bdee703e806214d46cd7dc85f90a7480 MUL-7995: fix(openclaw): report actionable state ownership lock error
  • #299 3a54152 Thanks @PazChen! - Chat: @ is the one menu: moments, files and skills. Every skill is there with what each does, and picking one names it in the message, as the Skills button does.
  • #278 1ed02d7 Thanks @rick-mingyu-liu! - Settings › Agents says when Claude Code is older than the one OpenFilm was tested with (2.1.308), whose model menu may then miss the newest models, and…
  • #283 de5a98b Thanks @WEBGRS! - Chat: when a DeepSeek Harness model needs an API key this computer doesn't have (only a DeepSeek account is signed in), the chat says so in the app's language…
n8n
  • core: Fix $('Node') expressions returning null for legacy Code node output (18c89cd)
  • Nothing to do. The browser extension updates itself once no chat is busy, now also while a message waits in the queue.
  • Goal step reviewer. Settings → Agents & automation → Review each Goal step's changes.
  • Same as system theme. Settings → Appearance → Theme → Same as system follows your Mac's or Windows' light or dark mode, also when it switches while the app is open.
  • Full Changelog
  • commit: 727afb1ad6902176e2ec55254154b900c489a998
  • source tag: beta2026-10-30_10-41-44
  • regions: cn global
  • ⚡ A faster Orca: Launch smoothly with hundreds of chats, with less lag as you type and switch tabs.
  • 🧩 Customize Orca with icon themes, syntax highlighting and custom editors.
  • 🖱️ Keep working while agents use Chrome, Blender and other Mac apps in the background.
  • Import your Zotero library. The literature library gains Import from Zotero: bring a personal library or selected collections from Zotero 10 or later on the same computer, carrying bibliographic…
  • PDF reading order is automatic. Opening a linked PDF no longer shows persistent reading-order controls above the document.
  • Windows setup is faster and safer. Installation drops from a ~21 s median to ~14 s with a third fewer payload files and parallel extraction, updates no longer delete research data stored next to…
  • Merge pull request #266 from palmier-io/docs/in-app-changelog-0.12.0
  • [docs] Update in-app 0.12.0 changelog
  • Merge pull request #265 from palmier-io/feat/adjust-controls-feel
Solidcraft
  • SolidCraft 95 theme: an active toggle on the view's bottom bar, such as Orthographic, showed as a blank dark blue block.
  • System theme on macOS: with Preferences › Interface › Theme set to System, the app follows a change of the macOS Appearance while it runs.
  • Older designs: a sketch with a projected face seen edge-on, saved by an older release, keeps its line when it is rebuilt, so an offset that uses that line no longer fails.
Phoenix
  • annotations: add DECISION and AGENT kinds (9e7a797)
  • api: expose prompt version creation timestamps (87a9934)
  • client: expose trace and session filter expressions (dab09f1)
  • Standalone pages: each page gets its own link, visitors keep their chat history, and the name, icon, theme color, welcome message, and suggested questions are all customizable.
  • Embedded chat window: a single code snippet embeds the page in your website, with a chat button in the bottom-right corner.
  • Data isolation: the Agent serving visitors can only use the tools, skills, and references enabled for that page.
  • 317188bcedf34ae70ec5d2261fa9073030febc17 feat(auth): redesign the sign-in, sign-up and forget pages and fix dark mode
  • f231a93ab3cffd1641f680659b447c7ffe5a0afb feat(dcr): inherit EnableSigninSession and EnableWebAuthn from default application
  • 20564f455bd2c9fb0434af5b0e476422dcec7fc5 feat(helm): probe /api/health with a startup probe and configurable timings
  • core: use to_arrow_table for DuckDB 2.0 compatibility (6c48e83)
  • Updated bundled Claude CLI to version 2.1.314
  • CI: raised PyPI project size limit guard from 50 GiB to 150 GiB to match the new quota granted by PyPI
  • fix: distinguish pre-dispatch refusal from uncertain delivery
No release notes published.
  • Added permissions.blockWritesOutsideWorkingDirectories: Write, Edit and NotebookEdit refuse paths outside the working directories in every permission mode
  • Added an optional groups list to Claude apps gateway upstreams: only developers in the listed IdP groups are sent to that upstream, on failover too
  • Added guardrail_trace to Claude apps gateway Bedrock upstreams, to log which Amazon Bedrock guardrail filter fired on a request, without the matched text
  • feat(cli): manage third party provider model connections through login (54fc4777e) — Anwesh
  • Revise README for clarity and project focus (d2116c9d4) — 0xpaulx
  • fix(cli): let update --check tell a new release from a failed lookup (1de50e8da) — Yathiskumar
  • bb desktop 0.48.1
  • "Tell me when…" without an AI running. A bot can promise to tell you when something happens, such as a reply arriving, a page changing or a time passing, and then stop.
  • Saved steps on a Mac. When a bot finishes a computer task on this Mac, it can save the steps. The next time you ask for the same thing, it repeats them in a few seconds with no AI and no usage.
  • Korean joins the app's languages.
Reasonix
  • CLI archives built from the Studio release studio-v2.56.0.

Oct 29, 2026

  • feat(marketing): list press coverage on the media page
  • fix(api): mint YC Bookface codes in Autumn, scoped to the Pro seat
  • chore(plugins): bump the superset Claude Code plugin to 0.3.1
Reason Network
  • feat: deliver Antigravity messages natively with durable receipts
  • Cheaper than vector RAG: for both index and query.
  • Faster: trees are created from layout stats.
  • Unified SDK: vectorless RAG. Local or cloud, for index and chat.
  • macOS (-sea assets): unsigned (Apple secrets not configured)
  • Linux (-sea assets): SHA256SUMS-sea published with the release
Codex Security
  • GitHub Action for Codex Security 0.8.2.
  • Installs @openai/codex-security@0.8.2.
  • yaml uses: openai/codex-security@400aab4fa5c3376df00bf0d79b65222953170f0d # action-v0.8.2
OpenCode
  • opencode-pty 0.4.0 (protocol 7) from packages/process-daemon, published to npm as @opencode/pty@0.4.0.
  • New durable API for fact conflict resolution so you can resolve data disagreements reliably from the API.
  • Conflict review is enabled by default and many fixes ensure 'choices' (user decisions about conflicting facts) are retired correctly on edits, deletes, and rebuilds.
  • Copy stored vectors between vector stores without re-embedding to save time and cost when moving/replicating embeddings.
  • Long agent turns use less CPU. A paired local benchmark with 1,000 tool steps used 50–59% less CPU than the previous implementation.
  • fx update now works as an alias for fx upgrade, including --channel and --json.
  • Tool results keep their original timestamps when a turn is rebuilt or resumed.
Ruflo
  • Verified-outcome learning loop (#4581, #4583, ADR-502), on by default (ruflo-mods learningLoop; set it to false to turn it off).
  • The mod records each main-loop turn. Its outcome is decided only by the exit status of a test, build, typecheck or lint command run after the turn's last edit.
  • || true, pipes and ; verify nothing, and the model's own words are never read.
  • api: expose prompt version creation timestamps (87a9934)
  • evals: support TypeSafe Jev decision models in Python classification evaluators (9089efc)
  • factory: grade PR reviews with one verdict and report only that verdict (2dcaa0b)
OpenMausBot
  • OpenRouter and other API-key models work again on Windows. Every turn there failed with "MCP tool count exceeds the 128-tool limit". The built-in tools now fit.
  • Saved Low, Medium and High Gemini Flash picks run again on Google Gemini (Antigravity), instead of failing every turn with "did not switch".
  • A slow local model gets to finish its answer. Each server under More servers has a Wait choice, from 5 to 60 minutes, so a local model on a slower computer is no longer cut off at 3.
OpenClaw
  • Failed updates stay recoverable: a failed npm rollback no longer leaves the installed package missing, and update checks keep Doctor recovery guidance, stop recopying private rehearsal databases…
  • Workspace restrictions reach Gateway-built tools: with tools.fs.workspaceOnly, openclaw delegates built through Gateway loopback and skill dispatch now require approval as intended, and media…
  • Microsoft Teams allowlists stay restrictive: name-keyed teams or groupAllowFrom entries that do not resolve now grant nothing instead of opening the bot to every team or group.
  • Added keyboard-driven transcript selection with word, line, and page motions, integrated search, and configurable shortcuts.
  • Rendered Markdown consistently in async questions, choices, answers, and transcript replies.
  • Added GPT-6.1 Terra to the bundled, Vertex AI, and Amazon Bedrock model catalogs.
eve
  • GET /eve/v1/info now serves only operator principals and returns 403 to end users.
  • Tool-stub grants on Vercel OIDC subjects move from vercelOidc to the new eveVercelOidc helper in eve/channels/eve.
  • A connection's headers and toolCall.providedArguments callbacks now receive ctx.approval.responder while an approved tool call runs, so an MCP or OpenAPI server can act for the person who approved it.
Strands Agents
  • feat(sdk): add structured output attempt limits (9cca7b2ed)
  • feat(stop): use agent.cancel() to stop the event loop (TypeScript) (0334ab309)
  • feat(vended-tools): add graph tool in TypeScript (36e6a4409)
  • #2752 34cbd30 Thanks @mattzcarey! - Add experimental_registerOpenTelemetry() to agents/observability, which records spans from OpenTelemetry-instrumented libraries as Workers trace spans.
  • #2771 54ae3e8 Thanks @mattzcarey! - Breaking: agents/skills now loads Agent Skills from Computer Workspace files (workspace()), R2 (r2()), or code (manifest()), checks SKILL.md against the Agent…
  • #2763 df85805 Thanks @mattzcarey! - Agents SDK spans (agent lifecycle, wrapAISDK, Think) now set span status error on failure, where the Workers runtime supports Span.setStatus().
AI SDK
  • Updated dependencies [7313ce2]
  • @ai-sdk/gateway@3.0.225
  • Blog: ThinkRail 0.2.0 release post
  • docs: add GitHub stars badge and star call-to-action to README
  • build: pin Bun 1.4.3 for development, CI, the CLI and the desktop runtime
  • Removed the Bun event-loop workaround utilities from @oh-my-pi/pi-agent-core/utils/yield; agent loops no longer insert cooperative sleeps while streaming or between tool calls.
  • Added AgentTool.approvalRestrictionPaths so tools can report the paths they read, allowing restricted-folder approvals to cover those paths without remembering approval separately for each path.
  • Added Agent.logContext, which adds custom context fields to every log record written during an agent run.
  • FreeHire as a job source: choose FreeHire in Settings → Job search to search its open catalogue: millions of postings read straight from companies' own career pages…
  • A line about the company or role: each role in Profile → Experience has About the company or role, for a line like "Series B fintech, 40 people" that would otherwise sit in the first bullet, where…
  • Importing a resume with Anthropic: with Anthropic chosen, it always failed with "The AI provider ran into a problem": Anthropic refuses the shape of answer Dossier asks for as too complex, and…
Mika
  • 默认只在花钱前问你。 发消息、发帖、提交、删除、改设置都直接做完,不再一步一问;付款、下单这类花钱的操作仍然先问你,确认卡片上写清在哪、买什么、多少钱。想连付款也不问,可以在设置里的「开发者选项」打开「自主模式」。
  • 悬浮胶囊默认一直显示。 在任何 App 里都能随时双击胶囊,把这一屏发给 Mika。不想要可以在设置里的「悬浮胶囊」改回「有任务时显示」或「关闭」。
  • 从胶囊、通知、定时任务发起的任务,在 App 的对话页里实时可见。 以前要等任务做完才出现,现在运行中就能看到你说的话和每一步,也能随时停止、插话。
Claude Mem
  • Reload and Back/Forward work. The page, focus area and open request drawer are in the URL (?view=, ?focus=, ?request=).
  • Your list view, filter and sort are remembered between visits. The theme still follows your system or ?theme=. If browser storage is blocked, preferences last for the current visit only.
  • Live updates are steadier. A stopped source can't publish. An older read never overwrites a newer live model. When a live update overtakes a save, you see the current model and your save is kept.
  • Reson8: new STT plugin for batch and realtime transcription, including server-side turn detection.
  • Mistral Voxtral 3: adds voxtral-mini-transcribe-3, voxtral-small-transcribe-3, voxtral-mini-transcribe-realtime-3, and voxtral-mini-tts-3, plus 36 TTS voice presets.
  • Inworld: adds inworld-tts-2-flash through the plugin and LiveKit Inference.
BrowserSkill
  • fix(geometry): unify cross-frame coordinates and screenshot conversion | 统一跨 frame 坐标与截图转换
  • refactor(observation): decouple DOM processing from capture orchestration | 解耦 DOM 处理与采集流程
  • refactor(observation): unify cross-frame capture coordination and data ownership | 统一跨 frame 采集协调与数据归属
No release notes published.
  • Host themes: the host app can theme the TUI over the server API, so Crush matches your applications.
  • Host context: the host can tell Crush what's in view, like open files, and those files appear in completions.
  • Append to the next prompt: clients can add text the agent reads with the next prompt, without starting a run or cluttering the chat.
  • feat(ingestion): attribute AI gateway spans to their gateway API key and service
  • feat(telemetry): report AI gateway request counts from self-hosted instances
  • feat(pagerduty): send alerts, project notifications and prompt changes to PagerDuty, with the routing key on the automation
  • Amazon Q for VS Code 2.9.0
Qwen Code
  • Managed memory availability now respects the memory.enableManagedAutoMemory setting.
  • Size-triggered microcompaction now clears old tool results toward a low watermark, so it no longer rewrites another part of the conversation prefix on every subsequent turn.
  • chore: bump version to 0.8.2
  • Privacy Bot: a Bot can be marked private; other Bots cannot read its folder, memory or chats, and sharing needs your approval.
  • Bots can read, check and edit Word, PowerPoint and Excel documents.
  • A Bot can propose a new skill or an AGENTS.md update after long work; applied only when you approve.
  • PR descriptions stop listing changed files and stop naming local briefs. This release also adds stacked run-brief and removes the ship-it/ prefix and the simulator.
  • All seven published @on-par packages move to 0.9.24 in lockstep.
  • bash npm install -g @on-par/factory-cli@latest
nanoMuse
  • The desktop starts again when you are signed in. The loopback door now asks for the port it had last, so the provider row is usually right as it is.
  • The relay's console loads on a large record. The relay keeps the figures of a turn run on a key of the person's own in one small row…
  • A task can be given a budget: tokens, money, time, and the size of one request. Off unless set.
  • Telegram waits for input left sitting: a session alert no longer reaches Telegram the moment it happens when no dashboard is open.
  • Spawned sessions stop chiming: the dashboard chime now plays only for alerts the server decides to send, so sessions a lead spawns and the machine shell stay silent, as their notifications already…
  • Alerts that land together sound once: alerts from any sessions within notifyDebounceMs (3 seconds by default) of the first one now fold into one alert instead of one per session.
  • build(nix): install OpenBot 0.53.2
  • Mark the mobile app 1.4.3 available on iOS, and turn on its iOS flags
  • Start on a Mac without a host manager when a parent folder is writable
  • Native Approvals for Harness Agents: Claude's and Codex's own permission requests now reach the AgentOS approval UI.
  • Initially Disabled Company Routines: Routine(enabled=False) lets a template ship an inactive Routine that a user turns on later from the UI, with True still the default.
  • Headline Numbers on Charts: ChartTools.create_chart() takes an optional preformatted value — a total or latest value such as "$22.3K" or "94%" — emitted as "value" in the chart block and rendered…
  • pi-subagents 0.84.0.
  • Security fix: workflow scripts can no longer reach the Feynman process outside their sandbox (GHSA-6v8j-6w28-2w8r).
  • Finished background subagent runs are now deleted from disk after 30 days (90 if their session still exists) instead of being kept forever.
  • One connector catalog. Customize → Connectors is one catalog for every project, with no flag: Kortix's own apps, the MCP, OpenAPI and GraphQL servers of the integrations.sh registry, and the…
  • Agent-led onboarding. A new account's first project opens on "Your first chat with Kortix" instead of the three-step setup wizard.
  • Kortix Drive on by default. Every project that never turned Volumes off gets Files: a private folder per person and the shared Company folder. New sessions mount /drives/me and /drives/company.
Solidcraft
  • rustls-webpki 0.103.16 and webpki-root-certs 1.0.10 for RUSTSEC-2026-0382 and RUSTSEC-2026-0383, which could let certificate checks accept distrusted intermediates.
  • Trackpad pinch zooms the view on Linux under Wayland and XWayland, as it already did on macOS and in the browser.
  • Play History: the History controls and Inspect → History have a Play button that rebuilds the history one feature at a time.
Opik
  • [OPIK-8748] [FE] feat: online-eval rules pick the OpenAI and OpenRouter reasoning effort
  • [OPIK-8809] [FE] feat: simplify the annotator view card and preview, and continue the same Ollie chat
  • [OPIK-9002] [BE] fix: let LLM clients fail over to the next address without replaying requests
  • npm: bash npm install -g base44@latest
  • Homebrew: bash brew install base44/tap/base44
  • Version: 0.2.6 NPM Tag: latest
  • chore(docker): upgrade Node.js to 24.22.1
  • [AI employees] Improved AI employee chat usability on mobile and in narrow blocks, with a compact toolbar, bottom-sheet options, touch-friendly selectors, and automatic options dismissal when…
  • [Feishu / Lark] Fix the Feishu workflow notification form not rendering on legacy URLs (no /v), and message fields being dropped from the saved node config on modern URLs
  • A notebook you can read. 小本本 has five new tabs beside 记忆 / 待办 / 打算, on both bodies
  • A heartbeat. About one look every 20 minutes at whether anything it is keeping an eye on has waited
  • 动态 (Mac) / the timeline (phone) — a quiet line for everything that is not conversation: a todo
  • Local model settings: The Viewer settings page now exposes mirror URL, model path, cache directory, offline mode, and load timeout for local embeddings.
  • Runtime compatibility: The plugin now uses SQLite 13.0.3 Node-API prebuilds and requires Node 22 or newer in the actual host process.
  • Memory Recovery and Capture Boundaries: Fixed dirty-data recovery, orphan trace writes, and LLM empty-response retry stability issues.
n8n
  • core: Require a signed token to accept an invitation (4580f22)
  • Google Drive Trigger Node: Read watched item instead of listing changes (caa8c82)
  • Let plugins give tools to specific agents
  • Added Chat display settings that group tool calls or collapse completed conversation turns
  • Let plugins add MCP servers whenever an agent session opens
  • A Helm chart for Sim - the open-source AI workspace where teams build, deploy, and manage AI agents
  • Add crewai skill search to list the organization's skills
  • Pass crewai run --inputs to a Python flow
  • Add deploy create --zip and print the commit a zip is built from
Omnara
  • feat: deduplicate compiled agent configs into agent_config_definitions
  • fix: sharpen error logging and cancellation handling
  • fix: raise default API rate limits 10x
  • Improve Markdown list formatting when copying and exporting
  • Support / in document titles
  • Improve editing, database view, and search performance
  • Added a custom_configuration field to NvidiaTTSService.Settings, which forwards model-specific key/value parameters to NVIDIA TTS — Magpie's flush, chunk_len_threshold and max_chunk_threshold…
  • Added ClassifierUserTurnStopStrategy and ClassifierUserTurnStrategies, which let a classifier, such as JevClassifier, decide whether the user's turn is complete.
  • Added number_pronunciation_language to SmallestTTSService settings, which reads numerals, currency, times and dates in a language of its own, independent of the synthesis language.
  • 「外观 → 尺寸 → 行宽」新增自定义百分比,可在 30–100% 之间按 1% 调整,对话与输入框一起生效,不影响已有预设。关联 #14649 fixed in
  • 设置页顶部的标题块(分类名、设置项数量)去掉了,导航已标明位置,第一张卡片直接出现在页面顶部。关联 #14625。
  • 已安装的程序位于项目文件夹内时,「用其他程序打开」会列出它并标明原因,悬停可看是哪个文件夹,不再悄悄消失。关联 #14627。
  • Claude Code integration — connect an installed and configured Claude Code CLI, with an explicit opt-in in Settings → Providers. Authentication and billing follow the CLI's native configuration.
  • Custom and local endpoints — configure one OpenAI-compatible endpoint with multiple models directly in Setup or Settings.
  • Desktop drag-and-drop file references — drop files into the content area to insert editable path chips into the focused learner editor, or the dictation destination when no editor is focused.
  • Attachments: multi-attachment messages and localized BYOA attachments/image inputs are now supported.
  • BYOA engines: added the local OMP engine, provider model discovery, and per-engine model catalogs.
  • Agent reliability: preserve Codex startup diagnostics, harden engine triage permissions, and remove off-boarded agents from conversations.
  • Four new languages: German, Russian, Turkish and Traditional Chinese join the language picker, and many more dialogs, menus and shortcuts are translated.
  • Faster workspace switching: Hovering a workspace in the sidebar gets it ready, the workspaces next to the current one stay warm, and the switch itself does less work. Dragging a row never switches.
  • Remote files that work like local ones: Play audio and video from SSH hosts and paired servers, save remote text files up to 50 MB, open folders too big for one reply as you scroll, and never open…
  • Improved streaming of agent replies: lines that scroll off the top of the screen are now scrolled into your terminal's scrollback instead of repainting the whole screen, and text wrapping writes…
  • Improved memory use shortly after startup: the CLI now returns about 100 MB of memory left over from startup to the operating system 15 seconds after the interactive UI opens.
  • Improved the after-compaction size shown in the compaction marker and in /context to also count what the summary carries into the next prompt, such as your earlier requests and artifact, subagent…
  • commit: b0c217c98b2346bb32a34821241992ba52918a86
  • source tag: beta2026-10-29_10-44-16
  • regions: cn global
  • @agent-browser/eve tools now come from the MCP tool catalog. The extension exposes the CLI's MCP tools instead of 21 hand-written ones, so tool names and inputs change: navigate is open, evaluate…
  • Added agent-browser mcp call <tool> '<json>' to run one MCP tool without a stdio session and print its result as JSON with the agent-browser version, for runtimes that cannot keep a process open…
  • Added @agent-browser/sandbox/mcp-tools.json, every MCP tool with its title, description, input schema, annotations and profiles, generated from the CLI and checked by a test
  • fix(deps): require dnspython>=2.9.0 (GHSA-45c3-73f7-pv4m) (f2f467bcc) — Vaibhav Upreti

Oct 28, 2026

LiteLLM
  • The cosign claims were validated
  • The signatures were verified against the specified public key
  • fix(proxy): backport #46712 gzip scoping and off-loop compression to stable/1.106.x for v1.106.6
  • docs(self-host): restore a production snapshot into a new stack
  • Agent import follow-ups: Back from a dropped agent, shared helpers, logged Undo failures
  • fix(runs): deny an open approval when its run is stopped
Openbot
  • Count each email that the account service sends in OpenPanel
  • build(nix): install OpenBot 0.53.1
  • Move a Windows install out of the OpenBot home folder on update
  • macOS (-sea assets): unsigned (Apple secrets not configured)
  • Linux (-sea assets): SHA256SUMS-sea published with the release
  • Connected apps work the same on every AI. A bot given Gmail or Calendar gets their tools on Claude, Codex, OpenCode and the rest, under the same names (search, read, draft, send emails; list events).
  • Answer a sign-in request any way you like. When a bot asks you to sign in, saying "done" in the chat counts like pressing Done.
  • Hold ⌃⌥Space anywhere, speak, and let go. MausBot answers in a small box at the bottom of the screen while you stay in your app. Click the box to open the Quick asks chat, where every ask lands.
  • Fixes an unauthenticated, webhook-triggerable crash of the whole API process: a Return Response header name containing a character illegal in an HTTP token (e.g.
  • Nothing to do. The browser extension updates itself once no chat is busy.
  • Scheduled runs. A chat can send a message by itself on a schedule: every few hours, daily, on weekdays or weekly (chat menu ⋯ → Schedule…).
  • Changed files under each answer. An answer that edited files shows a short "Changed files" line. Open it to see each file's changes, or go to Review.
Software Factory
  • base-branch for ship/run-issue/run-brief and multi-file run-brief.
  • All seven published @on-par packages move to 0.9.23 in lockstep.
  • bash npm install -g @on-par/factory-cli@latest
  • briefing: open and reply to Right now's email points from Home (ddc4429)
  • briefing: show your own inbox in Right now and message you when it changes (428456a)
  • delegation: open reply cards as a conversation and redraft on newer mail (bd293a7)
E2B
  • Updated dependencies [9fdfa21]
  • Updated dependencies [d58ef55]
  • @e2b/python-sdk@2.62.0
Sim
  • feat(library): Best AI Agents for Drafting Social Media Posts: 4 Platforms Compared
  • feat(library): Best AI Agent Platforms for Customer Onboarding: A Buyer’s Guide
  • feat(library): 7 Best AI Assistants for Inbox Management and Reply Drafting in 2026
Ruflo
  • @claude-flow/memory 3.1.0 adds ControllerRegistry.set(). The CLI's memory bridge already called it to register controllers it builds itself, but the method did not exist, so those registrations…
  • agentdb_pattern-store and agentdb_pattern-search use the same backend when that local ReasoningBank is registered.
  • Allowlist: require HTTPS and validate redirects, block on policy failure
  • Allow callers to supply Goose session IDs
  • Linux window icons restored in desktop app
eve
  • Ship current release notes at dist/CHANGELOG.md with a version-pinned link to the full changelog.
  • @chadhietala
  • Updated dependencies [9f19c79]
  • @ai-sdk/gateway@4.0.129
  • Preserve existing Codex hook registrations during setup and avoid nested hook directories (PR #345, #343).
  • Keep Copilot PreToolUse output neutral while using supported context events (PR #346, #344).
  • Honor PLANNING_DISABLED=1 across Gemini lifecycle hooks, including SessionEnd (PR #347, #329).
  • bb desktop 0.48.0
Qwen Code
  • feat(managed): add session permission rules and Auto approval
  • fix(cli): harden serve token resolution against project env injection
  • feat(core): add an experimental shared agent loop and run headless on it
OpenCode
  • opencode-pty 0.3.2 (protocol 7) from packages/process-daemon, published to npm as @opencode/pty@0.3.2.
No release notes published.
Opik
  • [OPIK-9002] [BE] fix: keep provider error bodies out of retry and error logs
  • [OPIK-8927] [FE] fix: a model switch reads the effort levels the panel showed
  • [OPIK-8897] [BE] [SDK] fix: price every prompt-size tier and 1-hour cache writes on every provider
  • Invite a teammate into the pod you're working in. The invite card and link sit right in the pod, and a new teammate sees the earlier messages.
  • Everyone brings their own agent. Each person can connect their own Claude Code or Codex from a Mac, or start a hosted one.
  • The pod team panel shows who's here, which agents are working, what's waiting on you, the pod's files, and who can join.
  • queryTraceThreads() thread rows no longer include model or provider, and page.limit is now capped at 100 (default 25); larger limits fail validation.
  • @mastra/connect connect-session creation response changed: POST /connect/:provider/sessions now returns launchUrl (server-side verified launch route) instead of connectUrl.
  • @mastra/code-sdk no longer registers built-in subagents by default; you must pass subagents: [...builtinSubagents] to keep prior behavior.
  • experiential 0.7.223 / exp-gateway-native 0.3.173: automatic prefix caching on the Gemini Developer API (AI Studio): Studio candidates in the automatic Google prefix planner…
  • Improved ACP configuration defaults to apply only to new conversations, preserving session settings when reopening existing conversations.
  • Improved fuzzy searching in agent model and configuration option pickers for Zed Agent and third-party ACP agents, while keeping search preparation off the UI thread. (#64147; thanks JacobNWolf)
  • Improved OpenCode compatibility when replaying DeepSeek reasoning and disabling thinking for Google models.
  • [PR] fix: external summary mode stops injecting the summary-writing rules
  • [PR] feat: Codex Desktop Plugin (Git marketplace + MCP tools + version sync) | feat: Codex 桌面版插件(Git marketplace + MCP 工具 + 版本同步)
  • [PR] fix: restore native ACP injection across DSH and Codex
  • pi-subagents 0.83.1.
  • Stopping, interrupting or resuming a research subagent can no longer act on the wrong run: an interrupt with no id is refused, and resuming a worktree child while it is still cleaning up no longer…
  • Agent files with a blank line inside an indented frontmatter block are read in full, so settings below that line are no longer dropped.
  • .Net: Bound element nesting depth in XML prompt parsing
  • .Net: Improve OpenAPI HTTP connection handling
  • Python: Avoid logging function argument values
CopilotKit
  • Includes security fixes for Teams and other Channels integrations.
  • Render static JSX as images in managed Slack and Teams messages.
  • Combine rendered cards with native buttons and carousel layouts.
  • Confirm Cmd+Q with a "Quit AO?" dialog on macOS
  • Remove per-device push mute
  • Draw the orchestrator's messages in a worker chat like its worker rows
  • Updated Bun engine requirement to ≥1.4.4
  • Fixed chat-completions context estimates counting past thinking that the request drops (no reasoning replay), which triggered compaction far below the threshold
  • Updated Bun engine requirement to ≥1.4.4
Base44
  • npm: bash npm install -g base44@latest
  • Homebrew: bash brew install base44/tap/base44
  • Version: 0.2.5 NPM Tag: latest
  • #4322 44cdf8f Thanks @Grapedge! - Allow expanding thinking during streaming with Ctrl-O
  • #4342 4a24ae1 Thanks @RealKai42! - Add /import-sessions and kimi import-sessions to import Claude Code and Codex sessions from the last 15 or 30 days. Run either one and choose a time range.
  • #4284 366fafb Thanks @7Sageer! - Sessions in web and desktop now unload from memory after 30 minutes idle and resume transparently on next access.
  • One update flow, with your confirmation — Check releases in Settings → About, choose whether to receive Preview updates, review the exact version, and confirm installation and restart.
  • Recovery that protects current data — Installer-managed updates can attempt repair and, when safe, restore the original program and service.
  • Durable conversation history — SQLite brings model history and WebUI display history under gateway ownership. Existing JSONL history is imported with source backups retained.
  • Scientific connectors reach native source APIs. Europe PMC gains direct Boolean, core and abstract search with explicit pagination, alongside new HGNC gene metadata, ChEA3 enrichment ranks…
  • Track reading progress in the literature library. References can be marked Unread, In progress, or Finished reading.
  • Session packages work in the browser. Local Web deployments can export a .science package as a download and import one through an upload picker — no CLI fallback needed when working entirely in…
n8n
  • Google Drive Node: Fix infinite pagination loop in v1 API request (3faa762)
  • MCP Server Trigger Node: Evict idle sessions to prevent memory leak (54eaf65)
  • It taps by describing, not by calculating. tap_target takes a sentence like *"the 发送 button
  • A drag that is a drag. drag presses, moves while holding, and releases — for sliders, splitters,
  • The Mac got the guards the phone already had — seven of them, each fixing a failure we had measured
  • Added image pasting through your terminal's own paste shortcut, such as Cmd+V or Ctrl+Shift+V with only an image copied, and by dragging image files into the prompt, which attaches them instead of…
  • Improved the /plugin Discover tab to group marketplace plugins into the same sections as the Antigravity plugin marketplace instead of one flat list, listing a plugin under every section it…
  • Improved agy plugin validate to warn when a plugin's name isn't lowercase kebab-case, such as Neo4j or foo_bar, which the marketplace can't publish, and to suggest a valid name, failing under…
  • 🔀 Handle a pull request without leaving the Pull Requests tab: merge it or turn off auto-merge, edit its title and description, add or remove labels, request or withdraw reviewers, and post…
  • ☁️ Databricks Sandboxes can now be managed hosts. Each one is created and run as the user through their Databricks account connection, with no SSH and no shared credential inside the Sandbox…
  • 🧩 Type /mcp in the composer to see a session's MCP servers and whether each is connected, with an inline Sign in for services and host servers that need one (#11128, #11338, #11378, @hzub, @jsonmp-k8)
Phoenix
  • Updated dependencies [86ab99c]
  • @arizeai/phoenix-client@7.16.3
  • explore run is remote by default — runs are pushed, checked and launched on HyperExecute; local-only flags say so in explore run --help. --claim-run is what the VM uses.
  • Up to 20 cloud browsers at once — explore --browser remote drives LambdaTest Chrome over CDP, opening browsers only as the explorer asks for them, and never leaves a VM running after release.
  • One clean stop — explore stop <id> aborts a queued run and stops a running one cleanly. At the terminal, the first Ctrl+C stops, the second acknowledges, the third kills immediately.
  • Chrome extensions (experimental): Add extensions from the Chrome Web Store with Add to Orca, so password managers like Bitwarden and 1Password can fill logins in Orca’s browser. Off by default.
  • Convert a terminal agent session live into a chat ui: Once you quit Claude or Codex in a terminal, the same conversation can continue in Chat while the terminal tab stays open.
  • Worktrees sooner after a restart: Local worktrees load before terminals restore, and the sidebar says discovery is still running instead of showing an empty list.
  • MY FOCUS rail. Your focus areas are listed in priority order, with a needs-you count on each. Every request sits in exactly one area, or in Other. On phones the areas appear as a chip row.
  • Area header. Each area page shows its goal sentence, combined progress, your own quotes, the asks you haven't started yet, and a link to Past focus. The My Goals page is retired.
  • Suggested. New areas appear here by themselves. You can Keep, Rename, Merge or Dismiss each one, with Undo. Your decisions sync across computers and a refresh never overrides them.
  • commit: 0694ed7e001e027a031ffed21a088bd1975132de
  • source tag: beta2026-10-28_11-03-58
  • regions: cn global
  • 修复 npm 0.4.3 误带旧版 Linux x64、macOS Apple Silicon 与 Intel worker 的问题,统一提供版本匹配的 0.4.4 运行时及 provenance。 @T-Auto
  • 修复 Linux 与 macOS worker 在安装包中缺少可执行权限的问题,确保解包后可直接执行。 @T-Auto
  • Replace the stale Linux x64, macOS Apple Silicon and Intel workers accidentally included in npm 0.4.3 with version-matched 0.4.4 runtimes and provenance receipts.
Codex Security
  • GitHub Action for Codex Security 0.8.0.
  • Installs @openai/codex-security@0.8.0.
  • yaml uses: openai/codex-security@4d454dae905c0655d0aaf5633dde8ce9fff4ebe6 # action-v0.8.0
  • [Authentication] Fixed authenticator options being rendered before their auth type plugin finished loading, which made LDAP sign-in fail with "must either provide a buffer via raw or some value"…
  • Bodies in different components no longer merge. Since 0.24.1, an Extrude to Object set to Join also joined the body it ran up to, even when that body belonged to another component, so separate…
  • New designs save as .sc. .solidcraft files keep opening everywhere, and saving one keeps its name; only Save As offers .sc.
Omnara
  • feat: prompt-first onboarding with agent system diagram
  • fix: drop prefix-mismatched Claude thinking server-side
  • docs: describe Ubuntu 26.04 on Omnara-managed machines
  • feat(cli): unify provider login and model selection (f686634e9) — Anwesh
  • build(deps-dev): bump pre-commit from 4.6.2 to 4.7.0 (64182e516) — dependabot[bot]
  • build(deps): bump fastapi from 0.141.1 to 0.162.0 (473c0c1a0) — dependabot[bot]
  • improvement: update SDK snapshot for Copilot CLI 1.0.105-1
  • Default CodeMode to dynamic_catalog=True, so a revealed tool is appended instead of rewriting the cached prompt
  • Pass the Gemini Live reconnect replay to GoogleRealtimeConnection's dial as a keyword-only replay argument
  • Give Harness LLMReminder a default interval=10 instead of generating on every request
Reasonix
  • CLI archives built from the Studio release studio-v2.54.0.
  • Selecting Copilot Fusion High or Max in /model now works from long-context sessions
Glimmervoid
  • Settings save again on a new install: in 0.46.0, a config file with only a few settings, as on a new install, refused every settings save, with a warning that config.json looked wiped.
No release notes published.

Oct 27, 2026

CrewAI
  • Fix failing flaky tests
  • Update snapshot and changelog for v1.15.31
  • Choose your dictation language. Settings → Providers now has a searchable language selector below the Deepgram key, including supported languages/locales and multilingual transcription.
  • Retry failed responses without losing your work. A separate Retry action reruns an eligible failed foreground turn while preserving its submitted input, failed attempts and partial output.
  • More reliable new-session opening. Opening text is kept out of the URL, accepted sessions use their normal short chat URL, and failed opening submissions can be recovered without leaving duplicate…
Solidcraft
  • Define parameters as you go: type name = value in a drawing tool's value box and the parameter is created.
  • Sketch dimensions: better placement and editing, switch a circle or arc between radius and diameter, delete dimensions from the right-click menu.
  • Add points to an existing spline, and Repeat Last Command in the command palette, with a key you can change.
  • Wasmtime and TLS security advisories are fixed.
  • The plugin host moves to Wasmtime 48.0.7, with the matching Pulley 48.0.7 and Cranelift 0.135.7 (RUSTSEC-2026-0375 to RUSTSEC-2026-0379).
  • TLS certificate verification uses rustls-webpki 0.103.16, webpki-roots 1.0.10 and webpki-root-certs 1.0.10.
  • macOS (-sea assets): unsigned (Apple secrets not configured)
  • Linux (-sea assets): SHA256SUMS-sea published with the release
CopilotKit
  • feat(runtime): send one request id per user request on the Intelligence lock call (d028e14)
  • feat(react-core,vue,angular): translate the chat's remaining hard-coded strings through labels (4ca3a23)
  • feat(showcase): add runnable Vue documentation (136f515)
Ruflo
  • ruflo-console 0.49.7: Claude can clear Project Anatole alerts through the console again.
  • claims_rebalance no longer hands more claims to the first underloaded agent than its limit allows.
  • @claude-flow/aidefence is declared as an optional dependency of the CLI.
OpenMausBot
  • Simple mode, redesigned: calmer, simpler, nothing lost. The app follows your system's light or dark theme by default, Midnight and Daylight are calmer, and the first window is roomier.
  • New Bot just makes the bot. One click creates it and opens its chat, ready to type. Customize it later, or say "Set me up".
  • A new bot sends your AI one tiny test message first, so a key that doesn't work shows up before you start.
  • @agent-browser/eve now requires eve 0.84.0 or later. eve 0.84.0 rejects extensions built by earlier releases, so the extension is now built with it.
  • Added one MCP server per Vercel Sandbox with startAgentBrowserMcp in @agent-browser/sandbox/vercel, which keeps a single agent-browser mcp running over Sandbox stdin, and connectAgentBrowserMcp…
  • Added --omit-command to batch and omitCommand to the MCP batch tool, replacing repeated raw command arrays in JSON output with zero-based indexes
  • Timeouts for code that can hang: @tool(timeout=...) limits each function tool call, and timeout= on stream subscribers, @observer, BaseObserver, LoopDetector and TokenMonitor limits each callback.
  • Anthropic SDK 2.x support, plus new AnthropicConfig options: a typed speed and open_telemetry to configure or disable the SDK's own OpenTelemetry spans.
  • OpenAIDecisionsConfig accepts question_instructions (strings, ag2 inputs or SDK parts), appended to the agent prompt.
  • Update ROADMAP.md to no longer reference outdated 2025 roadmap
  • Changelog for v0.65.0-preview.0
  • chore(release): bump version to 0.66.0-nightly.20261013.g204db94ae
  • Added oidc.user_id_claim to the Claude apps gateway, to identify people by an id_token claim other than sub, such as Microsoft Entra ID's oid
  • Added click to send: in fullscreen, clicking the "to send now" hint under a queued message sends it right away, as its key does
  • Fixed managed settings and organization policy not being fetched for the rest of the session after signing in from the VS Code extension
  • Added a built-in Google Cloud Vertex AI provider, configured with model_provider = "gcp".
  • Added GPT-6.1 Terra to the model catalogs.
  • Added a repository filter to the session picker and directory grouping to the agent command center, making it easier to navigate sessions across linked worktrees.
LiteLLM
  • Logging and observability: Changed failure logs for rejected requests to keep only core request fields, tags and guardrail details <sup>Also in 1.103.10, 1.104.8, 1.105.6</sup>
  • Logging and observability: Changed the GCS logger to use the bucket and credentials configured on the proxy <sup>Also in 1.103.10, 1.104.8, 1.105.6</sup>
Software Factory
  • Lockstep republish: recover from npm phantom-staged factory-core@0.9.20. Same product content as 0.9.20.
  • In a local synthetic benchmark with 96 MiB of saved images, fx used 69.5% less peak memory (116.5 MiB vs 381.8 MiB) and completed turns 3.3× faster (288 ms vs 939 ms) than 0.0.14.
  • Text sessions resume about 18% faster in local comparisons with 0.0.14.
  • WebAssembly also handles longer image conversations: a separate local benchmark completed 140 turns where the earlier implementation ran out of memory at turn 102.
AI SDK
  • b6b9dca: chore(ai): remove runtime validation of inputs that TypeScript types already guarantee
  • Updated dependencies [57086ff]
  • Updated dependencies [01d7a03]
  • Harness Agents in Teams and Workflows: ClaudeAgent and CodexAgent can now be members of an Agno Team and steps of a Workflow, including inside Parallel, Loop, Condition and Router.
  • Session Search: Added GET /sessions/filter-schema and POST /sessions/search with nested AND/OR/NOT filters on session name, session/user/component IDs, session type and created/updated timestamps…
  • Company Tasks: POST /company/tasks accepts status, so Tasks can be created in Backlog, Ready or Cancelled.
Opik
  • [OPIK-9002] [BE] fix: pin Studio LLM calls to the gateway and add opt-in executor network isolation
  • Security: rationale redaction credential redaction for rationale labels moves to a single pre-cache choke point (_redact_rationale_labels inside _safe_extract), so no extractor can skip it, and it…
  • Security: deps anyio is floored to >=4.14.2 in the mcp extra, above CVE-2026-63374…
  • Fix: markdown ATX headings parse per CommonMark: a trailing run of # is not part of the heading text (## Setup ## is "Setup"), up to three spaces of indent are allowed…
Oh My Pi
  • AgentTool gained the optional handlesToolInterrupt flag: an interruptible tool that sets it (such as wait) still runs, without blocking, when a loop-owned interrupt is already queued, so it can…
  • Fixed OAuth logins shared by several processes being disabled with invalid_grant when they refreshed at the same time or one of them quit mid-refresh
  • Fixed transient Google 429 rate-limit and capacity responses retrying in place without blocking or rotating credentials, while daily caps and zero-entitlement limits still rotate
Openbot
  • Remove the downloaded CLI of a provider that is off
  • Add AI agent comparison and examples guides
  • Manage Bitwarden and the Vault of a joined server from the iPhone app
Omnara
  • ghcr.io/omnara-ai/omnara-web@sha256:5f6549266d21fa4ee02d1de0f973d604d1ad0c0988a589c5c70b89aad1077400
  • ghcr.io/omnara-ai/omnara-api@sha256:2289b74e086eafc516d09b17ac644c10493501ff83469a226baf3bf536857da8
  • ghcr.io/omnara-ai/omnara-worker@sha256:5dd2146b6dbb446712f4baca838acd007f70e0d8dd9b2584f09ee06497ad7572
  • Gateways now refuse to start unless OIDC, mTLS user or operator authentication, or gateway JWTs are configured.
  • New sandboxes on Kubernetes, Docker, Podman, and VM now open the control connection from the sandbox to its supervisor, while running sandboxes keep their existing connection direction.
  • You can now suspend and resume sandboxes with openshell sandbox suspend and openshell sandbox resume or through the SDKs, on compute drivers that support it.
  • Defer QuickJS runtime initialization until evaluation.
  • bea2ad5 release(langchain-quickjs): 0.3.11
  • 12b4a78 fix(quickjs): defer runtime initialization until evaluation
  • Bound receipt canonicalization depth and reject non-finite signed numbers. Evaluator failures stop before invalid scores can enter signed evidence.
  • Clear all positive proof classifications when evidence cannot be digested.
  • Emit default-deny MCP policy for Claude Code and native Codex, Hermes, OpenClaw, Copilot, and OpenCode registrations. Preserve generated overlays and user policy edits during upgrades.
  • 41916a5: Dynamic models, dynamic subagent configs, and ctx.agent(name, { model }) overrides now read context windows from the AI Gateway model catalog recorded at build time instead of downloading…
  • b322096: A compaction that fails during a turn, for example on a provider 429 that outlasts its retries, now fails that turn with MODEL_CALL_FAILED and leaves the session waiting with its history…
  • 4276727: Messages a route forwards with ctx.to() now carry the deployment and request that accepted them, so an idle session reached that way hands off to the new deployment like one reached with…
Qwen Code
  • Managed memory availability now respects the memory.enableManagedAutoMemory setting.
  • Size-triggered microcompaction now clears old tool results toward a low watermark, so it no longer rewrites another part of the conversation prefix on every subsequent turn.
  • chore: bump version to 0.8.2
Langfuse
  • feat(ai-gateway): record model, key owner, byte and upstream error facts on generations
  • feat(ai-gateway): product analytics for setup, credentials, invites, keys and model access
  • feat(ai-gateway): restore empty-state placeholders for providers, models and gateway API keys
Dsh Ops
  • 将 DSH 产品类型、懒加载 scope 与 profile carrier 入口收敛到项目自有 adapter/dsh 边界。 @T-Auto
  • 将宿主兼容声明扩展为 >=0.2.0-rc.2 <0.2.2-0,覆盖 DSH 0.2.1-rc.4 与 0.2.1 稳定线。 @T-Auto
  • 保持现有 Cordis 工具、提示词、权限、subprocess、PTC、结果与 UI 行为,同时为 adapter 边界增加回归保护。 @T-Auto
  • feat(verify): stand in for DigitalOcean so Managed Agents can be verified
  • feat(canvas): per-conversation Docker runtime and Settings → Runtime page
  • feat(automations): filter and edit automation tags
  • The desktop app keeps running in the system tray when you close its window, so Bots go on with their work. Click the tray icon to open the window again, or choose Quit there to end the app.
  • An avatar library in Bot settings and New Bot: import several photos at once and reuse them for any Bot. If some files fail, the others are still imported.
  • When a Bot hands its browser to you for a sign-in, a payment or a check, a card in the chat lets you take over, skip it, or tell the Bot you're done.
  • 将 SDK 最低依赖版本提升至 0.1.17,修复 Compile 所需 SDK 接口缺失的问题。
  • 修复 MySQL 多服务任务隔离及旧 API 兼容性问题。
  • Raise the minimum SDK dependency to 0.1.17 to provide the SDK APIs required by Compile.
  • experiential 0.7.221 / exp-gateway-native 0.3.171. Tagged on GitLab at 5a55226f3.
  • Internal hostnames: pass up to 8 hostnames to browserRun() as internalHosts. Serve each one from a fetch function on your agent or from another Worker. See Internal hostnames.
  • Hosts the model points: give a host a target schema, and the model chooses what it serves with browser.assignHost(), such as a container port or a folder in its workspace.
  • Lighter installs without MCP servers: @modelcontextprotocol/sdk and @modelcontextprotocol/server are now optional peer dependencies, so apps that don't serve MCP no longer install them.
  • A small silent bubble — the lightest way it can speak first. Nothing is read aloud, the microphone
  • A plan it can be held to. Long-term directions and a day plan, both visible and editable in
  • A morning look and a spaced muse. Once a day it looks at the day ahead; for items with loose timing
  • Auto-pause and wake-on-access for fast sandboxes (OSEP-0024) lands end to end: the ingress records activity and parks requests to paused sandboxes until the restore completes, the server runs an…
  • Credential Vault (OSEP-0022) lands end to end for fast sandboxes: the egress sidecar holds the vault with durable per-subject persistence and sealed pause/resume transfer, the server exposes the…
  • Fast Sandbox snapshots and template builds can publish to and boot from OCI registries (OverlayBD OCI lifecycle), and a reliability campaign removes whole-call timeouts from SDK file transfers in…
OpenClaw
  • Updates that finish and recover: an ordinary post-update Doctor failure restarts the Gateway instead of leaving it stopped, an orphaned maintenance worker no longer blocks every CLI command and…
  • Gateway service installs on every desktop: macOS recovers a loaded launchd registration whose plist was deleted, Linux installs under the Ubuntu default umask create a private state directory…
  • Conversations keep accepting messages: killed runs release their restart-recovery fence, interrupted and just-accepted child sessions recover after a Gateway restart, inbound messages stop…
  • feat(db): add manual Trigger.dev task for the account token sealing backfill
  • feat(tailscale): expand and validate API operations
  • improvement(changelog): lead with knowledge base portability
  • 9543a0c65aad9b2783dfcde174458786050174db MUL-6546 fix(util): ignore mentions inside markdown code blocks and inline code
  • e8fdcfd226d238e6f1f22d5d95245936a795d9fd MUL-6623 feat(agent): add host-local Devin CLI over ACP
  • b8306f975c47be0743e6ee16bdbe4ec67eb440c1 MUL-7130: feat(issues): add overview and sub-issues to the outline
  • Create, install and share Desktop plugins (experimental)
  • Refreshed the desktop app experience
  • Add experimental image and decision generation APIs to plugins and sdk, allowing image generation or decision model (think typesafe, clef, etc) workflows.
  • "Google couldn't find that. It may have been deleted." in the inbox and in Settings: when a message arrived and was deleted before Dossier's next inbox check read it, spam for example, the check…
  • chore(docker): upgrade Node.js to 24.22.1
  • Docker images no longer include .env.enterprise. For Azure OpenAI in Docker, put its keys in .env.
  • A date a model passes without zero padding, such as 2024-7-5, is read as that date. It could compare as later than a filing made after the run date, and SEC EDGAR then served that filing.
  • A decision is no longer scored on the bar of a session still trading, and one made before a market closure longer than a week, such as a national holiday, settles once trading resumes instead of…
  • Downloads are meaningfully smaller. CLI and desktop runtime payloads are reduced by pruning unused generated components, bundling the translation worker against its compact runtime, shipping a…
  • The standalone CLI now speaks MCP. Research tools are exposed through a local MCP server for external MCP clients, under the selected profile's existing permissions — it is not a separate sandbox…
  • Long conversations read cleanly. Once a turn settles, intermediate messages collapse out of the way while running, waiting and interrupted work stays visible.
  • Added [ and ] to the Per Turn tab in /diff to jump to the first file of the previous or next turn, wrapping around at either end, in both the file list and the file view.
  • Improved how quickly /diff opens in large repositories and long conversations: All Changes and Commit Tree appear as soon as tracked changes are ready, showing checking for untracked files… while…
  • Improved /diff in checkouts with large untracked files: binary files and untracked files over 1 MB are listed with a note such as binary file or 7.3 MB, over the 1 MB limit instead of being read…
  • Full Changelog
  • 💻 Windows:双击下载的安装包,按提示完成安装。
  • 🍎 Mac:打开下载的安装包,将 toonflow.app 拖入“应用程序”,再从“应用程序”打开。
  • 📱 Android:在手机上打开 APK,按系统提示允许此来源安装应用。后续版本可直接覆盖安装。
  • refactor: clarify provider, catalog, and execution boundaries
  • feat(providers): add Microsoft Teams Bot provider
  • feat(github): accept the App's client ID as appId for GitHub App installation credentials
  • wren: add wren context export-ossie and follow the OSI → Apache Ossie rename (3a5cac4)
  • wren: add wren context import ossie and wren context export ossie (190ed1e)
  • wren: import a Wren project from an OpenMetadata catalog (3eecec4)
Noetic
  • fix(core): bound held spend by request bytes and save decide() generation ids from headers (976a075), closes
  • fix(core): charge a failed model call's recorded rounds and keep rounds[0] honest (b8eca4c)
  • fix(core): count the bill of model rounds rejected by recovery (e0f2270), closes
  • [DingTalk] Fix DingTalk Stream mode failing on file-based plugin installs; cut full-sync API calls ~90%; show DingTalk quota reset time on sync failure
  • [AI employees] Added an optional stateless request mode for OpenAI Responses services to avoid response-item errors when a gateway routes requests across Azure OpenAI resources.
  • [Feishu / Lark] Fix Feishu OAuth callback URLs missing the APP_PUBLIC_PATH prefix; fix automatic login not triggering and landing on a 404 in the v2 client
  • fix(model): declare real WAV sizes in the final streamed audio block
  • fix(embedding): isolate Ollama embedding caches by host
  • fix(model): await provider stream cleanup on public aclose
  • Change thinking effort mid-conversation via OpenAI configuration_update and Anthropic per-message output_config, recorded as ModelSettingsDeltaPart
  • Add ModelRequestRejectedError, ModelAuthenticationError and ModelNotFoundError model error categories
  • State loaded deferred capabilities' instructions in the system voice, append their changes, and show them to before_model_request
AI Memory
No release notes published.
  • Internal dependency updates only (Copilot CLI snapshot update to 1.0.104-3).
  • Full Changelog
  • > [!WARNING] > <details> > <summary>Firewall blocked 1 domain</summary> > > The following domain was blocked by the firewall during workflow execution: > > - github.com > > To allow these domains…
  • Managed model "fusion" now starts sessions on Auto and applies fusionTier settings
  • Interactive and SDK Autopilot sessions keep waiting for attached background tasks instead of stopping after COPILOT_TASK_WAIT_TIMEOUT_SECONDS; -p --autopilot still honors it
  • When a saved Auto preference is no longer available, Auto falls back to its default preference instead of failing each turn
Reasonix
  • CLI archives built from the Studio release studio-v2.53.1.
  • feat(onboarding): lead the startup menu with recommended CI/CD automation (986df129b) — vaibhav upreti
  • fix(deps): lift filelock to 4.2.4 and setuptools to 84.2.1 for open advisories (2b13af5e1) — Vaibhav Upreti
  • feat(shell): block the shell behind a Connect GitHub screen until GitHub is connected (e43d29662) — Vaibhav Upreti
  • Open Claude subagents from the sidebar: Click a Claude subagent row in the sidebar or dashboard to open its transcript in the same terminal, then click the parent to go back.
  • Chats that ask first: Codex chats in Ask mode now ask before editing your workspace, remote Claude terminals on paired servers can continue in Chat, and chats that can’t start tell you why.
  • Agents resume as you left them: Resumed Claude sessions keep their effort level and launch flags, Codex keeps the permission you last gave it, and Session History finds chats in more folders and…

Oct 26, 2026

  • The Stop hook copies what the turn made for you into ~/.claude-mem/deliverables/<project>/, content-addressed (report-3f9a1c2b.md), so the copy outlives the worktree, scratch folder or temp…
  • The worker records and serves each copy at GET /api/deliverables/:id/:filename, with the same loopback, same-origin and sandbox rules as recorded files; markdown and text open inline.
  • Privacy and limits are explicit. A turn whose prompt is entirely <private> leaves no copy and no record. Nothing is copied for a worker on another host, which could not open the copy.
Noetic
  • feat(core): graceful stop with cancel(ctx, { drainMs }) (8fed546), closes
  • feat(core): record paid model dispatches and settle cut-off bills (b64bf61), closes
  • feat(core): report cut-off drain calls by dispatch and generation id (caa45f7)
  • Cursor works again with Cursor 3.27, which now encrypts its stored app token: CodexBar decrypts it through Cursor's Safe Storage Keychain item, prompting at most once on a manual refresh and never…
  • Updates: automatic checks and automatic installs are now separate choices, and saved choices survive relaunches and Settings visits.
  • Usage and cost: OpenCode's local history can optionally feed Codex cost history, Codex usage reported over a fixed 30-day window gets pace, Kimi's monthly Total breaks down into Kimi Code and…
  • #232 2811d4b Thanks @rick-mingyu-liu! - An agent whose process ends before it answers, without a word, is no longer blamed on the network.
  • #175 cf6cd6f Thanks @PazChen! - Windows: DeepSeek Harness installed with npm starts in the chat.
  • #245 4981238 Thanks @PazChen! - Enable Cursor's parameterized model picker and refresh agent options for the selected model before the first message.
  • pipeline: put per-request saved tokens on OutcomeSnapshot (8cedf7b)
  • proxy: let turn hooks rewrite OpenAI Responses requests on HTTP and WebSocket (fbf7bc3)
  • proxy: report the same tool savings on gateway /v1/compress as the proxy path (a904210)
  • macOS (-sea assets): unsigned (Apple secrets not configured)
  • Linux (-sea assets): SHA256SUMS-sea published with the release
eve
  • 4a4951e: Tools and connections now share the Approval policy contract; approvalKey and ApprovalContext.approvedTools are removed.
  • 4a4951e: Session commands now return Delivery receipts with independent stream() and result() reads.
  • 4a4951e: Public APIs now use eve-owned contracts for history, decisions, channel integrations, and sandboxes; legacy aliases and internal helpers are no longer exported.
  • Add OpenAI browser tools and the shared Bash helper.
  • Improve bounded page observations, form state, file access, and browser recovery.
  • Fix temporary empty Cloud tab URLs.
  • Added --hold-idle-file and --max-hold-min to self-hosted-runner, to keep an idle session while your host reports processes it left running
  • Added a tip for the Explanatory output style, shown when you ask Claude for explanations
  • Fixed long freezes at startup and after reading or editing files when a CLAUDE.md file holds a very long list item
  • Skills added or edited on disk are picked up by running sessions without /reload
  • The footer shows the active model and its thinking level next to the token counter, and updates after /model, /thinking or a resume.
  • Clicking the model in the footer opens /model while no turn is running, keeping the prompt being typed.
Deep Agents
  • Fixed PTC prompt caching to cache prompts per slot and tool identity.
  • 9d654b5 release(langchain-quickjs): 0.3.10
  • e72c1e6 fix(quickjs): cache the PTC prompt per slot and tool identity
  • Added SummaryOptions.remotePromptCacheKey and remoteCompactionPolicyModel() so remote compaction can carry a cache key gated against the request it actually sends
  • Remote compaction now sends cache and session headers for the endpoint and model the request actually uses, so a configured compaction override no longer inherits the live turn's routing
  • Aside providers (AgentLoopConfig.getAsideMessages, Agent.setAsideMessageProvider) now receive which boundary is polled, "work" after tool calls or "stop" at the end of a turn, so they can keep a…
  • release(checkpoint): 4.3.1
  • release(langgraph): 1.2.18
  • fix(langgraph): let error_handler recover on every path and stream failed task results
Ruflo
  • mcp start -t http|websocket with no token now uses a per-user token in ~/.claude-flow/mcp-http-token (mode 0600, created on first use), so a loopback server is no longer open to every local process.
  • Federation: VERIFIED and ATTESTED trust now require the key pinned for the peer in the trustedPeerKeys config (node id to hex public key).
  • A file in .claude-flow/daemon-queue/ no longer counts as a manual trigger, so it cannot skip the supervisor gate.
No release notes published.
Glimmervoid
  • Quick session from the one you are in: Cmd+Option+0 on Mac, Alt+Shift+0 elsewhere, starts a new session right away on the same project as the session you are typing in (or the selected one), with…
  • A session closes itself when its agent quits: typing /quit (or any clean exit of the agent) now removes the session instead of leaving a "Session complete" card.
  • Remove skips the confirmation when nothing would be lost: Cmd+K, the card's X, the actions menu, focus view and the phone board now remove a session at once when its worktree has no uncommitted…
AI SDK
  • 47236dd: add onStepError callback to generateText, streamText, ToolLoopAgent and telemetry integrations, and onError and onAbort to ToolLoopAgent
  • c8e8866: Add aiVersion to telemetry events so tracing-channel subscribers and telemetry integrations can identify the emitting ai package version, including in bundled applications and…
  • 1cf95c6: add experimental steering to generateText, streamText and ToolLoopAgent
Openbot
  • Stop the AppImage crash handler before the AppImage unmounts
  • build(nix): install OpenBot 0.51.0
  • Fix local mobile builds in worktrees
E2B
  • bea0fe6: - Added an idempotencyKey (JavaScript) and idempotency_key (Python) option
  • fix: route Bedrock India (in.) inference profiles via bedrock-runtime
  • Remove open-computer (moved to its own repo)
  • Neutralize spreadsheet formulas in CSV chat exports
  • feat(marketing): trim the paid pricing cards
  • fix(marketing): skip the API read for a malformed fight handle
  • fix(desktop): route fire-and-forget openExternal calls through safeOpenExternal
  • fix(release): inherit workspace version for schema dialect (e4228cc)
  • chore(release): stamp v0.2.1 [skip ci] (a98c095)
  • chore(release): publish v0.2.1 --release v0.2.1 (54a6391)
Opik
  • [NA] [FE] feat: hide columns in the playground dataset and test suite table
  • [NA] [BE] fix: leave playground traces out of the home page's trace count
  • [NA] [FE] fix: run the playground with Cmd/Ctrl+Enter, and only the prompt you are in
  • Security: deps the Pillow floor is raised to >=13.0.0 on Python 3.11+ (staying at >=12.3.0 on 3.10, which Pillow 13 no longer supports) and pypdf to >=6.19.0, and the lockfile is refreshed to…
  • Fix: llm a chunk that failed before returning a response no longer has its files reported as model omissions.
  • Fix: watch a Markdown file that detect() classifies as a paper (arXiv markers, an Abstract) is included in the watch quick-scan rebuild instead of contributing zero nodes.
  • pi-subagents 0.83.0.
  • Background research subagents that finished their work always report it, instead of ending as "exited with code 0 before writing a result".
  • Turns after a subagent returns keep the prompt cache on models that take mid-conversation system messages, so they are cheaper and faster.
  • 用户可为自己保存 Team、Agent、Task 的默认组合。新会话会直接使用该组合,不再每次弹出选择
  • 未设置时仍走原来的选择流程;默认组合失效时自动回到原流程,并说明原因
  • 可在会话中锁定或解锁该默认组合,避免后续会话被意外改掉
  • ui: a nonce Content-Security-Policy on every page (85b5945)
  • ui: confirm changes inside a frame once the browser reports the question seen (9d9ca16)
  • ui: let host origins frame the UI and call a short host list (42b626d)
  • Better keyword ranking: fixing prefix-score inflation raised strict recall-all@5 from 65.74% to 84.68%, a gain of 18.94 percentage points, in the documented BM25 development evaluation.
  • Less repeated summary work: summaries process new observations and merge them into the existing summary.
  • Bounded graph work: search shares indexed graph reads and avoids repeated full-graph enumeration.
  • #4414 f36c127 Thanks @RealKai42! - Stop showing a failed swarm for AgentSwarm calls rejected before any subagent starts, such as while swarm mode is off.
  • #4424 f895f30 Thanks @liruifengv! - Update model switch confirmation prompts.
  • #4423 07a34fe Thanks @RealKai42! - Show a high-usage warning when you turn on swarm mode with /swarm.
Paseo
  • Fixed repeated Android animation exceptions for views that have already unmounted
  • Fixed the Android animation run loop continuing while the app is in the background
  • Fixed finished Android chat messages being remeasured during streaming updates
  • Docs: a Mailings tab merges a local data source into letters, envelopes and labels (@tokg-venv, asked
  • Sheets: encrypted workbooks open, save and change their password (@barathrajd); restyling an existing table from the Home gallery no longer overlaps it (@widechaos, reported
  • Slides: A4/A3 paper sizes, portrait and a custom slide size (@DuBearte, asked
Codex Security
  • GitHub Action for Codex Security 0.7.0.
  • Installs @openai/codex-security@0.7.0.
  • yaml uses: openai/codex-security@c1551aa2983456a05c616804f311b2d4636280c4 # action-v0.7.0
Langfuse
  • security: rebuild to pull node.js patches
  • test(worker): skip Azure LLM connection tests until CI key is replaced (v3 backport of
  • Joints follow your parts: a joint placed on a hole rim, an edge or a corner now moves with it when the part changes (thicker, moved hole), for every joint type, in one design and through Update Links.
  • macOS: Undo, Redo, Settings and Help are back in the app bar.
  • Coloured components: each component gets its own colour, and you can change it.
  • fix(mcp): support explicit user confirmation for recording
  • Codex/audit intervention transport
  • fix: drop PascalCase tool projection and tool-call flooding; cancel same-step duplicates
Cline
  • Cloud tasks: Settings > API Configuration has a Local | Cloud toggle when cloud sessions are on. Cloud picks the Cline model new cloud tasks run on, saved separately from your local models.
  • Starting or resuming a cloud sandbox shows a progress card with the repo, branch and steps (start sandbox, clone repository, start agent). It collapses once the agent starts.
  • Cloud sessions hide the Plan/Act toggle and the auto-approve bar instead of showing them disabled.
  • #15472 9d28e36 - Fix copying selected chat text with Cmd/Ctrl+C or right-click Copy in VS Code 1.143. The selection no longer clears and focus no longer jumps to the prompt.
  • #15410 5f2aa0b Thanks @WebReflection! - Update DOMPurify to 3.4.17 to fix sanitization issues flagged by Dependabot alerts.
  • #15412 0377248 Thanks @WebReflection! - Update vite to 7.3.7 to fix the launch-editor editor CLI flag misinterpretation flagged by Dependabot (CVE-2026-112944).
  • More reliable math and code rendering. Corrected escaped percentage signs disappearing from questions, along with common literal-newline and over-escaping cases in tool-generated content.
  • Figures no longer change the rest of the app's styling. SVG figures are isolated from the page, keeping plan labels readable while preserving the figure's own appearance, including fullscreen.
  • PDF paths work consistently with the other file tools. Home-relative and ~/ paths are resolved before reading; absolute paths keep working. Refs #95, #113.
  • [File Editing] Isolate drafts across notes, skills and artifacts, retain failed edits for retry, and support up to 8 MiB text editing and previews.
  • [Language] Add Italian (it-IT) UI support with full translations and emoji picker.
  • [Voice] Hold the mouse button in a text input to start dictation, release to transcribe.
Andee
  • Asset file names gained a platform field — andee-<version>-<lang>-<platform>-<device>.<ext>, so
  • The order a reader sees is the alphabetical order of these names. GitHub sorts the asset list by
  • No Han characters, per above.
  • Queens write to each other. Turn on queen_mail and a queen gets two tools: list_queens
  • Chains in bounds, every hop logged. A letter or reply is one hop deeper than the deepest letter
  • A queen run by Claude Code. Under queen_mail.claude_code, a queen is
n8n
  • core: Release the SQLite write lock when the write connection fails to open (01b5e80)
  • Update certifi to 2026.10.23 in task runner (3012503)
agent-browser
  • Fixed wait --download timing out when the download finished before the wait started, and moving the file to the requested path
  • Fixed --allowed-domains launches failing with "Cannot find default execution context" on pages that embed cross-origin iframes
  • Fixed full-page screenshots on high-DPI displays being twice the page size with the page in the top-left quarter
  • Managed memory availability now respects the memory.enableManagedAutoMemory setting.
  • Size-triggered microcompaction now clears old tool results toward a low watermark, so it no longer rewrites another part of the conversation prefix on every subsequent turn.
  • chore: bump version to 0.8.2
  • Chats that follow your settings: New chats use each agent’s permission from Settings → Agents, renamed chats keep their names in Session History, and a chat won’t write a conversation a terminal…
  • A browser closer to Chrome: Dock DevTools inside the pane, mute tabs playing sound, see a link’s URL on hover, and use Orca shortcuts and mouse back/forward inside pages.
  • Watch agents browse: A cursor and badge show where an agent is clicking or typing, chat jumps you to the page, and links from terminals and chat open beside their source.
No release notes published.
  • Seven new Scientific Connector tools span imaging, pathways, lipids, structures, and cancer models.
  • The Literature library becomes a full research inbox. Semantic Scholar citation discovery surfaces a paper's references and citing papers, multi-source supplementary attachments download with…
  • Session packages work from anywhere. .science export and import now run from headless CLI and local Web sessions, not only the desktop app.
  • bash curl -fsSL | bash -s -- --release --version 0.1.2026.10.26
  • Start OpenSRE
  • bash opensre
Reasonix
  • CLI archives built from the Studio release studio-v2.52.0.

Releases by week

Every stable release of the week, Monday to Sunday, for the last 26 weeks. Each agent's full history is on its own page.