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 19 – Oct 25, 2026

614 stable releases from 144 tracked agents and frameworks in the week of Oct 19, 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.

Oct 25, 2026

Noetic
  • refactor(types): keep the tool approver with the harness judgment calls (caef046), closes #129
Solidcraft
  • Section views are solid again: bodies behind the cut no longer show their colour through the cap, and hovering the cut no longer lights up what it hides.
  • Cosmetic threads are drawn on internal and external threads.
  • Text parameters: sketch text built from a parameter follows every change.
  • feat(library): AI Coding Agents vs Code Completion Tools: Which Should Engineering Teams Choose?
  • feat(mothership): run scheduled jobs as durable unattended chat turns
  • improvement(docs): build feature-branch docs previews only when docs change
Glimmervoid
  • Delete my data: Settings > Privacy has a Delete my data button. After you confirm, Glimmervoid asks for everything its usage data recorded under this install to be erased and switches to a new…
  • Usage data says plainly what it collects: the first-run banner, now with a cat, and the terminal notice read "Glimmervoid collects usage data to help us improve it.
  • Anonymous usage data marks a first start: the app start event now says whether this is the first time Glimmervoid has started on the machine (no config file yet), so a new install can be told…
Claude Mem
  • npm publishing now uses trusted publishing (OIDC). The "Publish to npm" GitHub Actions workflow authenticates to npm with a short-lived OIDC token issued for this repository's workflow instead of…
  • playground: refresh MiniMax text model presets (a49a47b)
  • add star history chart and rank badge to READMEs (3123853)
  • integrations: add Microsoft Agent Framework tracing page (a772dad)
AI Memory
No release notes published.
  • hooks: wrap unambiguously-fish scripts as rtk run --shell fish (eb8bd76)
  • ng: compact Angular build tables safely (c375c3c)
  • analytics: align weekly spending joins and aggregate project rows (5f6111b)
  • Restores the GitHub release record for experiential 0.7.220, published to PyPI on October 24, 2026.
  • The original annotated GitLab tag is preserved exactly: v0.7.220 points to source commit 29b51ed85df6e129256cb1c47e1b93ff61969ea8 (tag object c57e2d599d6d53042e1c0715788a207e42461c2b).
  • This restores release metadata for the GitHub publishing cutover. Source code, package versions, and existing PyPI artifacts are unchanged.
  • Nothing to do. The browser extension updates itself once no chat is busy.
  • Phone notifications. Chat On Steroids can tell you on your phone, through the free ntfy app, when a chat needs you, something fails, or a long chat finishes. Settings → General → Phone notifications.
  • Settings backup. Settings → General → Backup exports your settings to a file and imports them on another computer.
  • npm: bash npm install -g base44@latest
  • Homebrew: bash brew install base44/tap/base44
  • Version: 0.2.3 NPM Tag: latest
Rakazo
  • Render emoji and stop browser save prompts on bot computers
  • Support isolated Slack workspace installations
  • Guide authentication recovery after blocked shell substitution
  • Continue a channel request when a lead is set, and name the lead
  • Make the Linux remote desktop runtime start on Ubuntu 26.04
  • Show a storage card in the sidebar when the disk is almost full
  • 改进视频任务恢复:保留上游任务与结果,支持重新下载、续查和中断恢复,减少重复生成。
  • 修复默认聊天模型选择时的配置传递。
  • 改进桌面启动体验:显示启动页面,增加健康检查超时处理和脱敏诊断日志。
Billion Context
  • fix(ci): resolve step clobbered the real PATH with its 'publish|cut|noop' variable — jq/git/gh all died exit 127 (runs 40193290049/40193606788/40193898967/40194217250); rename to PMODE
  • fix(ci): always install our own jq — system jq-1.7 vanished between steps of the same job (run 40193898967)
  • fix(ci): release-bugfix.yml must bootstrap its own jq — 2026-10 runner images dropped it (dispatch 40193606788 died at resolve, exit 127)
Nocobase
  • [Workflow: CC] Restrict workflow task reads to assigned users and supported associations.
  • Bots can drive a browser: one Playwright profile per Bot, with sign-ins kept between runs. Posting, uploading, downloading and deleting need your approval first, and payments always go back to you.
  • /btw is now Side chat: type /side or pick Side chat from the Bot panel's ··· menu and ask follow-ups in the side panel. The panel menu also opens a terminal or the file tree.
  • The desktop app's About page shows whether DS Bot is up to date, downloading or ready to restart, with a Check for updates button.
  • GitHub polling needs an allowlist. New host-only github.polling_repositories = ["owner/repo", ...]; empty (the default) ignores every notification, so polling deployments must set it.
  • Jira project_keys = [] now disables lookups. The default is now ["*"]. If you set [] to mean "all projects", change it to ["*"] or remove it.
  • litellm.extra_headers is host-only. Repository .pr_agent.toml and comment arguments can no longer set it.
  • Hermes: a plugin gives each Hermes conversation its own address on the network. It turns on with a Hermes that has the plugin APIs (NousResearch/hermes-agent#143310).
  • Skill: one task per session, and conversations start at person/session@machine.
  • Gemini CLI: one session per window, not two.
LiteLLM
  • Deployment: Fixed the proxy sometimes exiting on startup with an ImportError when it cannot reach the remote model cost map, broken since v1.104.0 <sup>Also in 1.104.7, 1.105.5</sup>
  • Deployment: Fixed image scans flagging brotli 1.2.0, the emSign ECC Root CA - G3 and unused lockfile packages
  • #4386 fd2e2c7 Thanks @RealKai42! - Handle tasks directly by default instead of proactively launching subagents.
  • #4384 c9a4836 Thanks @RealKai42! - Agent swarms now start only when you turn on swarm mode: the agent can no longer launch AgentSwarm by itself, so swarms are no longer available in kimi -p or ACP…
  • 5bf0641173af1deef3eb979a2b88e1d5d2d00870 feat: add default record retention days to global settings
  • 266a0c0a88a430da47a98e0517f1eabca9edcae1 feat: brand the description, og tags and manifest of sign-in pages
  • a866fe91a75f21b3210aabdb0262807e3e7d58a5 feat: support custom TableColumns mapping in Lark syncer (e.g. enterprise_email)
  • The record: every conversation, on every device, whichever key answered. nanoMuse Cloud keeps each chat turn it answered since relay 0.29, in full, with the pictures as files named by their…
  • The face's clips travel with the face, and each device says what answers it. A clip drawn on one device is put up and named in the profile's clips ({mood: {sha256, mime, bytes}}).
  • nanoMuse Cloud answers the nanoMuse apps only. A sixth switch on the operator's Controls, nanoMuse clients only, is on by default: a chat request on an account key from software that is not one of…
  • Conversations that stay with you: Claude chats now show turns made outside Orca. Retained messages return after reconnect without reopening the chat, and crash recovery no longer guesses that a…
  • A browser with a better memory: Search your browsing history, reopen pages, remove unwanted suggestions, and clear history with Undo.
  • Remote work with fewer interruptions: Slow connections survive missed wake probes, reconnect notices explain what is happening, SSH terminals wait for sign-in, and new splits stay on the paired…
  • 建用户时可以同时创建一个默认专家。权限按页面 Tab 拆开,也可以按组折叠
  • 用户组可以限制技能包、知识库、连接器、模型、存储和专家谁能看见
  • 技能包可以共享,并用文件树编辑。创建专家、团队等资源时可以选头像
  • Managed cloud agents: dynamic workflows, controls and verified advisory learning (ruflo-agent 0.4.0, ruflo-workflows 0.7.0, ruflo-console 0.49.0; ADR 0003 and 0004). Everything is opt-in.
  • User guide: a "Managed cloud agents" section in docs/USERGUIDE.md.
  • memory_search_unified dropped an entry whenever the same key existed in two namespaces; it now deduplicates on namespace and key
  • bash curl -fsSL | bash -s -- --release --version 0.1.2026.10.25
  • Start OpenSRE
  • bash opensre
  • The golden image of the inputs d27023f0394d, built by the golden image workflow from ef69f5e19c1b46e41e743234fa5019e39442adcb.
  • Sign in with GitHub Copilot or Grok in Settings → Providers, without entering an API key.
  • Cleaner connection controls across ChatGPT, GitHub Copilot and Grok, with quieter account management and recovery actions shown only when appropriate.
  • macOS: prevent the backend process from creating a second, generic Dock icon. Refs #46.

Oct 24, 2026

Solidcraft
  • Render workspace (first release): a new Render tab with GPU rendering, studio lighting, materials, and image export.
  • Frames: a profile library and frame members along sketch lines.
  • Holes: place a hole by distances from reference edges, snap to face and round-edge centres, and drag placed holes. Tapped holes gain metric fine, UNC/UNF and G/Rp catalogues.
Ruflo
  • ruRE: CUDA launch plan (ruflo-rure 0.6.0, ADR-501). New read-only MCP tool rure_cuda_launch_plan turns what rure_analyze_cuda found into a plan for measuring: per architecture and kernel, how many…
  • rure_analyze_cuda reported the wrong architecture (sm_5) for a standalone .cubin built with CUDA 12 for sm_70 to sm_90; the SM number sits in a different byte depending on the cubin's OS ABI.
  • The worker daemon let a maxConcurrent or workerTimeoutMs saved in daemon-state.json override the same setting in config.json.
Noetic
  • feat(core): error model fixes and replay follow-ups for #111 and #115 (d85827e), closes #111 #115 #119 #117 #115 #115 #111 #115 #111
  • feat(react-core): collect reply ratings with comments for Automatic Learning (4930320)
  • feat(react-core): collapse tool calls made in a row (f0a2490)
  • fix(mcp-apps)!: give MCP App widgets an opaque origin by default (b0c1736)
Reasonix
  • CLI archives built from the Studio release studio-v2.51.0.
  • delegation: let the own-mail hold follow the safety level (297218b)
  • orchestrator: set the safety level per channel and check all outside content (f0ae974)
  • ui: explain the safety levels with one worked example (fa5fb15)
  • A queen's soul. Her profile says who she is meant to be and her memory holds what she knows.
  • Her profile is a page of her own (/queen/<id>/profile), in place of the side panel. Her card
  • Persona: her text (core_traits in profile.yaml) in the chapters it is written in
  • A reconnect never delivers a message twice (test)
  • README links the license files
  • DESIGN: the inbound policy covers messages from other machines
eve
  • 8329939: A parent agent's message to a subagent no longer answers the subagent's pending tool approval or session-limit prompt.
  • 084b2bf: When compaction runs partway through a turn, notices the framework sends again, such as the deferred tool catalog notice, now go before the turn's input instead of after the latest tool…
  • c37d279: The connection sign-in page now says "Authorization not completed" instead of "Authorization complete" when the provider redirects back with an OAuth error, such as the person declining…
OpenMausBot
  • Moving to My Cloud no longer waits forever. The move used to wait on anything unfinished, even a card waiting for you to sign in, connect an app or answer a question, so it could sit there for hours.
  • It waits only for a bot that is answering right now, and names it: "waiting for Scout to finish". A bot that's answering is never cut off.
  • Other unfinished work holds the move for at most 30 minutes.
  • Add and switch between saved accounts, identified by email.
  • Disconnect an account to remove its saved credentials from this device.
  • Control background use with one setting shared across ChatGPT accounts. It is enabled by default; background work counts toward your ChatGPT plan usage.
  • 5475347: Fix non-streaming Responses API structured output failing when multiple assistant messages contain JSON. Providers can supply separate structured-output text for parsing, allowing OpenAI…
  • Updated dependencies [5475347]
  • Updated dependencies [e729d58]
  • For the installer path, run CodeWhaleSetup.exe; it installs
  • Extract the archive for your machine: codewhale-windows-x64.zip or
  • Double-click codewhale.bat (not the raw .exe) to launch
  • Emad Mokhtar (@EmadMokhtar), for the updated optional evaluation and resume case (PR #338).
  • Thanut Jongteerathanachote (@winxtxrgit), for the Copilot context schema fix and regression tests (PR #340).
  • Parman Mohammadalizadeh (@MannXo), for the Cursor installation and hook-merge guide (PR #341).
Qwen Code
  • feat(serve): add daemon token persistence
  • fix(release): cover every spawned vendored binary in the tarball exec-bit gate
  • test(managed-agent): reuse a same-key cancel (FG11d)
Glimmervoid
  • See who is next in line: every session waiting on you, failed or finished unseen now carries a small numbered tab in the corner of its row in the Focus rail and on the phone, in the same order as…
  • Phone bottom nav usable on first run: before your first session, the Next session banner sat on top of the phone's bottom nav and blocked taps on More.
  • evals: apply template_format override when copying a PromptTemplate (7121ade)
  • evals: let sync/async client kwargs override the shared LLM kwargs (27361d4)
  • refactor(deepwork): drop the speculative claims path; unify the tombstone flip predicate
  • fix(agents): deny question for subagents by default
  • fix(companion): stabilize multi-monitor popup and native dragging
Omnara
  • ghcr.io/omnara-ai/omnara-web@sha256:c3bec21bb0df1a17d4ec8ccca4dc742bdc61ee2688250cd12f505c5ad6e0d32e
  • ghcr.io/omnara-ai/omnara-api@sha256:4ef7be855384f42a2fb3fdd490a0b730e5afebf38235a1d23a6332654a67eec4
  • ghcr.io/omnara-ai/omnara-worker@sha256:7e04f0f71041d9054c9b194e3cd5545899f7f2391071074c68f6bdd5847734d6
  • feat(public-api): add unstable read-only Topics endpoints
  • feat(ai-gateway): update Claude Desktop setup steps and merge Claude tabs
  • feat(web): warn ClickHouse staff viewing data
  • The golden image of the inputs 67532b5680e8, built by the golden image workflow from ef69f5e19c1b46e41e743234fa5019e39442adcb.
OpenClaw
  • Safer agent file edits: apply_patch now validates an ordered envelope before mutating the workspace, so a later invalid hunk cannot delete files changed by an earlier operation.
  • More reliable managed upgrades: app-owned updates now converge official plugins alongside core and retain repair diagnostics. Fixes #174120. Thanks @kazuyuki-eguchi and @RomneyDa.
  • Memory Wiki data safety: OKF concept re-imports now preserve human-authored notes. Fixes #175700. Thanks @YangManBOBO and @RomneyDa.
  • 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.
  • Thinking models on Ollama: with a model that thinks before it answers, such as qwen3.5:4b, reading a posting and writing documents ended in "The model's answer was cut off" or "didn't come back in…
  • New providers and Usage & Spend: abliteration.ai and Chatbox AI join the bundled providers (110 in total), and Usage & Spend can search and sort sessions and open one to inspect its requests.
  • Reliability: a guarded file-descriptor crash during CLI discovery is fixed, passive Antigravity (and CodeRabbit) probes no longer trigger multi-gigabyte background CLI updates, Claude OAuth…
  • Menu bar and plugins: Auto tokens name the window they picked (5h 84%, W +7%), stacked text uses a heavier weight, and user plugins gain window titles, reset text, usage hooks, and visible refresh…
  • 独立 headless CLI(dst) — dst run、-p、dst cli 不加载 React / Ink / channel / TUI 偏好即可被脚本驱动,dst dsh 原样转交官方参数;随 npm 包分发,src/ 零改动。自动化适配器当前只验证 DSH 0.2.0-rc.2,使用官方 headless 的持久化与恢复规则,不承诺与任意 TUI preset / fork…
  • Windows 离线安装器 — Release 增加 dsh-tui-setup-win-x64.exe:自带轻量 Node 运行时(系统已有满足版本约束的 Node 则复用)、免管理员、把 npm 全局路径追加到 HKCU 的 PATH、profile 落在 %USERPROFILE%\.dsh\profiles\dsh-tui,卸载不触碰 .dsh…
  • /settings 分页 — 设置屏分成「常规 + 插件」页签;不声明 page 的旧插件区块并入常规页,只有一页时不显示页签条。同时修掉 dsh-tui 自己的设置区块会因注册表竞态从屏上消失的问题。
LiteLLM
  • Deployment: Verify image signatures with the cosign.pub from commit e82a3c6, the previous key no longer matches <sup>Also in 1.103.8, 1.104.6, 1.105.4</sup>
  • Deployment: Removed the eMudhra root CA from every certificate bundle the SDK, proxy and Docker images trust <sup>Also in 1.103.8, 1.104.6, 1.105.4</sup>
  • Deployment: Changed the cosign key that signs Docker images and Helm charts to the one from commit e82a3c6 <sup>Also in 1.103.8, 1.104.6, 1.105.4</sup>
  • Let an hourly routine window cross midnight
  • Change the server key without hosting the server again
  • Keep older chat history when a capped snapshot arrives
  • nanoMuse Cloud is one model, for chat and for the screen. The account serves Qwen/Qwen3.6-27B-FP8, a multimodal model that reads pictures, so the chat and the hands run on the same deployment.
  • Pictures and clips come from a provider of your own. nanoMuse Cloud stopped funding image and video generation: the Making pictures and Making clips rows on Android, the iPhone, the desktop and…
  • The relay's sign-in gives nothing away, and a credential is one device. A wrong password, an unknown account and an account without a password get the same answer.
  • Embedded runtimes can coexist in one process with independent network policies, including background maintenance and provider requests.
  • Add SQLite/PostgreSQL schema equivalence checks, sharded CI tests, and a web production-build check.
  • Consolidate provider error factories and timeout defaults, extract credential resolution, and add a provider scaffold with safe failure cleanup.
  • Skills, ready wherever you work: Install Orca’s skills in one click from setup screens, and install or update them on paired servers and SSH hosts with Orca’s own installer.
  • A browser that fits your projects: Choose shared, per-repo or per-workspace profiles for new tabs.
  • More fluid agent conversations: Split into a fresh chat beside your current one, continue with a terminal agent from the pane header, and keep queued messages and multiple image attachments intact.
  • harness: Codex adapter for the provider error classifier (usage limit as usage_cap with try-again reset time)
  • unpark queues at the back of the lane (max order + 1), never at an old or front order; keeping the old order needs --keep-order
  • review rework 2/3: a queued review rework runs BUILD rework on the same PR branch with only that review's comments, then re-runs CHECK
  • Added support for foldable phones such as iPhone Duo, which switch to the wide layout when unfolded
  • Added connected hosts, so agents can create, prompt and monitor agents on other machines and see them in the Subagents track
  • Added paseo.sh in Simplified Chinese at paseo.sh/zh-CN
  • bb desktop 0.47.1
  • Security updates from advisories published Oct 23.
  • undici 8.12.3, now a direct dependency, so a fresh install no longer gives every package the pinned 8.10.2 below (GHSA-xjj6-f33q-929c, GHSA-4qv8-2p9j-2685, GHSA-f72v-6w45-q472).
  • body-parser 2.4.0, fast-uri 3.1.9 and zod 4.6.7, through Alpha Hub's MCP SDK.
  • Updated Node.js to the 24.22.1 security release.
  • Hardened Twilio webhook verification and CSS sanitization.
  • Added business hours and withheld caller ID support for Twilio calls. [Enterprise]
  • [PR] feat(ci): patch-train backport bot — backport label routes merged fixes onto the tag-based bugfix-release train
  • [PR] Fix: control-plane system messages collapsed as regular messages + position validity fallback (#4457, #4292 family root | 修复:控制面 system 消息按普通消息折叠 + 位置合法性兜底(#4457,#4292 家族根修)
  • [PR] fix(ci): derive the open patch train from refs — the variables-API pointer structurally 403s under GITHUB_TOKEN
  • Warn when a provider-specific thinking setting from a broader scope overrides thinking
  • Send the least thinking a model accepts for thinking=False on models that can't turn thinking off
  • Raise NotImplementedError for a file in a CohereModel tool return instead of dropping it
  • Added lastUserInput to the JSON that hook commands receive on stdin, so a hook can inspect the latest message you sent, for example to screen prompts before the agent acts on them.
  • Added a temporary opt-in Halloween theme: turn on Seasonal Themes in /config, a row shown only while the theme is in season, for a Halloween startup banner, a moon-phase spinner, and a themed…
  • Improved Auto mode to ask for approval less often on routine commands, such as cleaning up temporary files, stopping a local process, or running a code formatter.
  • ThunderPhone: new realtime speech-to-speech plugin for saved-agent and inline sessions, including bolt.
  • OpenAI: Decisions API support directly and through Inference with openai/gpt-6-luna-decisions.
  • TypeSafe Jev and LiveKit Clef Omni: decision evaluation through LiveKit Inference.
  • Fix AGENTS.md to start the login process while evaluating and stop after three rounds.
  • Update snapshot and changelog for v1.15.29.
  • Added PerplexityAgentHttpLLMService for Perplexity's Agent API, which answers with web search and supports function calling. It defaults to openai/gpt-5.6-luna with web search on.
  • Settings adds preset, web_search and max_steps. preset and model are independent, and at least one must be set.
  • With neither preset nor reasoning set, each request sends Perplexity's fast preset, its lowest-latency setup, and the first request for each model logs it.
  • fix: remediate GitHub code scanning finding #2424 - Fixed #2424 with a o (6479872be) — Vincent Hus
  • feat(llm): add subscription adapter registry and gated Grok OAuth (4e3e85515) — Anwesh
  • fix: remediate GitHub code scanning finding #2421 - - Updated surfaces/ (bc23d1562) — Vincent Hus
  • release(langgraph): 1.2.17
  • fix(langgraph): apply an explicit recursion_limit equal to the default
  • fix(langgraph): keep the caller's namespace in RemoteGraph.stream(subgraphs=True)
  • Switch agents mid-session in chat and terminal
  • Move the chat's switch to terminal into the session tab menu
  • Show agent html_render pages in chat, with tap-to-expand and pinch zoom
  • bd purge no longer deletes the closed root of a chain that still has live work under it
  • Auto-backup never claims a commit the snapshot does not contain (#7044, fixes #7032). The
  • The fan-in recheck no longer reverts concurrent writes (#7146, backport of #6876, fixes
  • Updated bundled Claude CLI to version 2.1.310
  • Operators can now edit the company. The Operator role preset grows from one grant (tasks:assign) to fourteen: agent creation and configuration, skills, environments, invitations, task assignment…
  • GitHub review bots move behind their own experimental setting. The GitHub Code Review Bot catalog entry, its setup, and its management screens appear only while Settings → Experimental → GitHub…
  • Budgets hold on unpriced usage by default. Each budget policy gains an unpriced-usage setting that defaults to block: while a run's usage has no reliable price, the policy counts as at its limit…
  • Three packages. @layerfs/sdk is now @layerfs/agent (Projects,
  • The layerfs command. One @layerfs/agent call per command, with JSON
  • Explicit Sandboxes. sandbox.create() starts a Sandbox and

Oct 23, 2026

  • Added the Monitor tool and plugin monitors on Bedrock, Vertex, and Foundry, and when telemetry or nonessential traffic is disabled; long foreground sleeps are now blocked there too
  • Fixed requests failing behind some gateways and compatible servers that reject the effort-2025-11-24 beta value or mid-conversation system messages; Claude Code now retries without them
  • Fixed background cleanup deleting some recent files, such as pasted images, when cleanupPeriodDays is set to a very large value
  • macOS (-sea assets): unsigned (Apple secrets not configured)
  • Linux (-sea assets): SHA256SUMS-sea published with the release
  • bb desktop 0.47.0
  • Your edits stay together: Tabs of the same file share unsaved text across workspaces.
  • Chats that keep up: Codex and Pi show turns made outside the chat. Long pastes can become text attachments, Codex questions get answer cards, and OMP gains image sending and actionable plan-review…
  • More control over agents: Pick chat or terminal when creating a workspace, set that choice for your default agent, add Claude and Codex accounts on supported remote servers, and install shared…
Deep Agents
  • Avoid importing langchain_anthropic and langchain_openai at SDK import time.
  • e85c6b9 release(deepagents): 0.7.32
  • 492a184 perf(sdk): stop importing langchain_anthropic at import time
  • Works with the new HyperFrames skills. Update yours with npx hyperframes skills update.
  • brag-slim treats website content as data and asks before installing anything.
  • Stricter contrast check before rendering.
AI SDK
  • Updated dependencies [e4469a3]
  • @ai-sdk/gateway@3.0.221
  • #1219 f589a32 Thanks @okwasniewski! - Add onAction to agent.act: a hook that runs after each action once the screen settles, for replayed actions too. Use it to take a screenshot per action.
  • #1326 04acca1 Thanks @okwasniewski! - app.screenshot() returns the absolute path of the file it wrote, not a path relative to the attempt directory.
  • #1218 58a5cec Thanks @okwasniewski! - Keep screenshots after a secret fill. app.screenshot(), toHaveScreenshot(), and assertion and failure screenshots no longer stop once a test fills a secret.
Ruflo
  • ruflo cleanup --force --keep-config deleted the config files it is meant to keep (claude-flow.config.json, the hooks/claudeFlow blocks of .claude/settings.json): the parser stores the flag as…
  • ruflo-console 0.45.1: ADR init, saves and exports work on Windows. The write layer shelled out to dd, install and sh, which do not exist there.
  • ruflo-console: the autopilot envelope accepts Windows paths (C:\work\app, UNC) and compares them ignoring case and separator style; a POSIX-rooted path containing a backslash is still refused.
  • An empty reply with finish_reason=length is logged and retried once with reasoning forced off. That retry counts against the existing paid-send budget and is bounded by the per-attempt timeout.
  • Billed usage keeps the retry's prompt size and adds completion tokens and cost from both attempts.
  • Health status names the active OpenRouter credential, including a custom OPENROUTER_BASE_URL.
  • Every new session runs pi. The pi_harness flag and runtime: opencode no longer select OpenCode. A session that already runs OpenCode keeps it. OpenCode is deprecated.
  • Share one response. Share response, beside Copy response, publishes one finished assistant response through a public link: the response only, or the prompt and the response.
  • Weekly changelog. /changelog groups what went live in production by week, and each week has its own page with every change.
  • Set up a remote host from Settings: Settings > Sessions has a Set up remote host button.
  • Pick your own jump key: Settings > Shortcuts and about now records the key that, held with Cmd on a Mac or Alt elsewhere, jumps to the next session needing you, in place of J.
  • Agents are pointed at glimmervoid spawn: an operator session's agent is now told to delegate work that edits files or runs long with glimmervoid spawn, which gives the task its own dashboard card…
Phoenix
  • factory: link session IDs in factory comments to the Claude Console (b022cfd)
  • factory: nominate factory candidates in triage and dispatch trusted ones (2dcd97c)
  • server: add service.version to the server's OpenTelemetry resource (6933827)
  • fix(v3): set additionalProperties false on nullable act action schema
  • fix(v3): honor CAPTCHA awareness options for local attachments
  • fix(v3): avoid forced done tool use for newer Claude models
  • Stability: Access 12 mature experimental features now promoted to stable status. See the unit guide for more details. (ea298d4)
  • Classifiers: Route workflows with TypeSafe's Jev models via the new TypesafeClassifier and ClassifierNode. Jev returns calibrated probabilities in a single forward pass, much faster than an LLM.
  • Live Audio: Enable natural real-time voice interactions with barge-in support alongside Gemini speech-to-text and text-to-speech integration. See the unit guide for more details. (2e7ef92)
  • WU Leizhi (@MbappeWU), for the OpenCode status fix and root/named-plan regression coverage (PR #331).
  • @kuei51307-hub, for the DSH selector notice and command-level regression coverage (PR #333).
  • Emad Mokhtar (@EmadMokhtar), for the skill-lens evaluation, configuration and measurement guide (PR #334).
Learnharness
  • Fix OpenCode Go requests omitting the required conversation identifier. LearnHarness now supplies application and conversation identity through Pi’s native APIs, reusing existing chat IDs and…
  • Refresh the provider model catalog by upgrading the integrated Pi libraries to 1.4.0.
  • Upgrade all active KaTeX rendering paths to 0.18.2, addressing GHSA-238p-pmpm-9mq7 while preserving math rendering and equation copying across messages, questions, practice problems, and Mermaid…
  • Fix: security the skill install stages every write (version stamp, SKILL.md, references) through a randomly-named mkstemp/mkdtemp file created with O_EXCL and restrictive mode instead of a…
  • Fix: python an explicit module-level type alias (X: TypeAlias = ..., type X = ..., NewType) is extracted as a type node, so a field or parameter typed by the alias resolves to it instead of…
  • Fix: go a type A = B alias is extracted as a type node (tree-sitter parses it as a type_alias, not a type_spec), so references resolve to it; a re-export (type Field = pkg.Field) mints no node
  • [PR] test: verif harness paths via fileURLToPath — fixes doubled-drive ENOENT on Windows CI
  • [PR] test: skip the #4079 egress-fuse e2e pair on windows CI runners
  • [PR] compress: accept flat single-range summary-in-content calls, guard unrecognized refs separators, unmask reject reasons
Reasonix
  • CLI archives built from the Studio release studio-v2.50.0.
  • chore(desktop): bump version to 1.39.0
  • fix(host-service): keep trailing whitespace in the git top-level path
  • fix(cloud-workspaces): stop box restarts from reordering the list
  • Add support for LiteLLM-compatible kwargs.
  • Implement platform application search in CLI.
  • Highlight action picker search in CLI.
  • chore(docker): upgrade Node.js to 24.22.1
OpenMausBot
  • Meet MausBot. MausBot is a liquid-chrome helper that knows the app. Ask it how to do anything, and it sets up bots, apps and routines for you. It has a Chief of Staff's powers over your bots.
  • If you don't have a Chief of Staff yet, MausBot becomes yours and keeps an eye on your day. If you do, your Chief stays exactly as it is.
  • A welcome card introduces it once. Tap its picture to rename it, pick another colour or look, upload a picture or generate one with AI. Add MausBot slides it into the top of your list.
Software Factory
  • factory status and kpis show the count of open factory draft PRs
  • Baseline cache entries expire, are re-checked when GitHub CI disagrees, and can be cleared
  • review rework 1/3: a trusted "changes requested" review on a factory PR queues its issue for review rework, once per review id
  • 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.
Ziva
  • Godot 4.2+
  • Active Ziva account
  • Automatic thinking level operation for improved model reasoning
  • Resume failed or cancelled delegate subagents
  • Subagent session link in UI with view-only subagent sessions
  • #2684 9c79ba1 Thanks @ben-reitz! - Fix PiHarness with onWake not resuming a sleeping Pi task, such as a compaction retry, while another task is running. See How recovery works.
  • Desktop app for Linux: one AppImage that runs on every distribution and updates itself, and a deb for Debian and Ubuntu (x64). See Install.
  • Side questions with /btw: in a Bot's own chat, type /btw <question> to ask while it works.
  • Message layout in Settings → General → Appearance: a Bubble width slider (50–100%), and a flat layout that shows a Bot's words without a bubble, under its avatar, name and time, with a Text width…
  • cd5f15d366f516c992ec8331f0bc158708bb69ec: feat(openrouter): send app attribution headers and session_id for sticky routing (@Lee-Si-Yoon)
  • a8a8ec8aaeef8464ede0461407621f1e081a29c1: feat: add a built-in keanu theme (@andreynering)
  • 78b6edb353f3fb11178c7e38bbac762772dfb8b0: feat: publish optional signed Packslip release manifests (@jdx)
  • Focus the composer when the user opens an agent
  • build(nix): install OpenBot 0.48.1
  • Add a Set all menu to the agent list of a role
  • 9c8a3514a15a702f81bfda456d80c7ccaa4d2712 Update package versions for release
  • c3b9f22d64fd86e8133395555a1fa4aa22f745e0 Add ACGS to the Python community projects
  • 71cf48cb4019bc503c7ceb61420cc3d836250790 Python: Honor allow_multiple_tool_calls in AnthropicClient when tool_choice is unset
Feynman
  • pi-subagents 0.81.0.
  • A research subagent asked for structured output that forgets to return it gets one more turn to do so, instead of failing and losing its work.
  • The main session can tell subagent completion notices apart from text you typed.
  • feat(ai-gateway): show per-tool setup instructions after creating a gateway key
  • feat(topics): write rules as target trace transcript
  • feat(blob-storage): add an outbound host allowlist for blob storage endpoints
OpenHands
  • feat(verify-openhands): launch --mars-web and --lock-to-cloud
  • feat(digitalocean): keep the web build signed in across restarts
  • fix(automations): render the shipped current_phase on live runs
  • Updated dependencies to address recent software ecosystem vulnerabilities.
  • CodexAgent: Now signs in automatically with OPENAI_API_KEY, and keeps an existing ChatGPT or other login.
  • OS Manager: OS Manager now lists and runs external agents such as ClaudeAgent and CodexAgent registered in AgentOS(agents=[...]), by name or ID.
  • Knowledge Upload: POST /knowledge/content now returns a clear 400 message when an upload has no file, text_content or URL, or when url is not a single URL.
  • [server] Allow the snowflake worker id to be rebound while the application is running.
  • [Workflow] Improved workflow execution detail performance by loading only the latest job per node and loading node history on demand with cursor pagination and lightweight fields.
  • [AI employees]
  • 隐私与数据治理:新增 Session 隐私自动提取、恢复和隐私树相关 MCP 工具;支持查看当前隐私配置树。
  • Agent / Memory 稳定性:改进 Session 记忆读取与提取重试、Phase 2 解耦、轨迹去重、插件 pending 写入回放、跨分支捕获和 recall 上下文记录。
  • Compile 工作流:新增增量更新、调度与版本历史,并修复 workspace 根目录及 Windows 沙箱命令处理。
  • Recon toolbox baked into the image: nuclei, httpx, subfinder, katana, ffuf and dalfox at release-pinned versions, plus a sqlmap checkout that tracks the default branch.
  • Recursive recon skill: bbot v3.0.2 with a policy gate. Passive, zero-touch scans when a program bans scanners; active modules only when it allows them.
  • php2shell skill: upgrade a file-inclusion finding to verified RCE, lab first and gated on program policy.
  • Publications: a new, optional section in your profile for papers, articles, books and chapters.
  • A photo on your resumes: add one in Profile → Basics and crop it: drag it or use the arrow keys, and zoom with the slider.
  • A theme for each application: the Resume and Cover letter tabs have a Theme menu.
Opik
  • [OPIK-6842] [DOCS] docs: explain how threads get their environment
  • [OPIK-7791] [QA] e2e specs for release 2.2.106, 2.2.108, 2.2.111, 2.2.116
  • [OPIK-7532] [QA] Proposed taxonomy changes from app surface
Reason Network
  • fix: a link inside an off folder no longer takes sessions out of it
  • A fixed UDP port for direct connections, set per machine
  • An off folder also holds a session reached through links above and inside it
  • Download installers / 下载安装包
  • 0fc67948219a0cdc8e4213dee18e75fb2716a8f5 MUL-7035 feat(cli): add comparison operators to issue list --property
  • 53c0088f1789c37e13846031eb85749383461e8c MUL-7847 fix(task): comment terminal failure on issue when sweeper fails task
  • 6b0f8315f6774949cd367a3ae414b10a6239ee54 MUL-7960 fix(mobile): align lint and TypeScript with Expo SDK 57
WrenAI
  • core-py: ship the wren-core 0.5.1 DISTINCT ON fix in the bindings (17f7ed9)
  • aa5de6438acacf2223b27235fc9a08c9c573a789: chore(electron): upgrade Electron from 42.11.8 to 42.11.16
  • <!-- This is an auto-generated description by cubic. --> <a href=" target="_blank" rel="noopener noreferrer" data-no-image-dialog="true"><picture><source media="(prefers-color-scheme: dark)"…
  • #4348 c569bab Thanks @liruifengv! - Fix workspace trust failures when Git Bash is unavailable on Windows.
  • #4348 c569bab Thanks @liruifengv! - Fix sessions occasionally remaining occupied after an application crash on macOS and Linux.
  • #4356 0e3980e Thanks @7Sageer! - Show the holding client's name in session-in-use errors.
  • Port agentic RAG to GO
  • Revert "fix(ingestion): allow re-parsing a document whose parse was canceled at progress 0"
  • fix(agent/tool): translate MCP tool inputSchema.properties in Info()
  • Published triggers: the draft-sync fix does not repair subscriptions already removed before upgrading. Rebind missing subscriptions and republish affected workflows as needed.
  • Document extraction: re-import/re-index affected content to apply extraction fixes to existing documents.
  • Built-in document metadata: toggling built-in fields rewrites only enabled, non-archived documents whose indexing is complete.
  • Microsoft Store app: updates through the Store.
  • Desktop app installed from the installer (.exe / .dmg): no change in this release; it updates with the next bundled release.
  • CLI: hermes update, or re-run the installer one-liner.
  • Signed Caveman runtime companions for macOS, Linux, and Windows. The binaries are licensed under the Apache License 2.0 (LICENSE, NOTICE); see LICENSING.md and the attached third-party notices.
  • Save reusable prompts and slash commands in ten numbered /presets slots, with a keyboard menu for quick access.
  • Set model, reasoning effort, speed, and permission profiles directly with inline /model and /permissions arguments.
  • Preview viewed and generated images in the fullscreen transcript on terminals supporting Kitty or Sixel graphics.
  • Added a Configure MCP servers... row for installed plugins in /plugin, so you can set or change a plugin's MCP server variables at any time without disabling and re-enabling it.
  • Improved terminal sandbox defaults to protect more credentials: sandboxed commands can no longer read or modify ~/.docker, ~/.kube, or ~/_netrc unless you grant access explicitly.
  • Improved context compaction, including /compact, to keep images, PDFs, and other files you attached in the conversation summary, so the agent can still see them after older history is summarized.
  • feat(knowledge): import knowledge base bundles through the API and sim knowledge import
  • feat(monitoring): log event-loop lag, stalls, and in-flight route handlers
  • feat(ec2): add Amazon EC2 integration
  • No notable changes
  • commit: 6e12261af3121a1fc8f56a23481f3b99bb32787c
  • source tag: beta2026-10-23_10-31-52
  • regions: cn global
  • feat(triggers): add 15 provider integrations and shared source support
  • feat(minimax): default text actions and discovery to M3.1 Flash
  • fix(cloudflare): show the granted scopes on the MCP consent page
  • New providers and CLI login: Soniox and Trae CN join the bundled providers (108 in total), and codexbar config login --provider copilot signs in to Copilot from the CLI through GitHub's device flow.
  • Pace and spend: monthly allowances show a daily target (what's left divided by the days until reset), Claude Cowork usage is no longer counted twice, Claude's priced subtotal survives unpriced…
  • Menu bar and Settings: iCloud fleet usage appears in the menu bar for providers signed in only on another Mac, stacked meters share one column, user-plugin quotas draw both meters in the merged…
  • Raise model error categories like ModelRateLimitError from embedding models
  • Add usage_limits, deadline, run_id and conversation_id to ModelSelectionContext
  • Default BedrockConverseModel's maxTokens to Claude's maximum output, streaming such requests behind the scenes
  • feat(auth): add browser ChatGPT subscription login (db77c0e17) — Anwesh
  • Stream replies live and continue cut-off replies (0db7d5ced) — Vaibhav Upreti
  • perf(tools): keep the Kubernetes SDK out of the tool registry load (4faf8d52a) — Vaibhav Upreti
Omnara
  • feat: limit agent git credentials to chosen repositories and read-only access
  • feat(machinepool): add Superserve sandbox provider
  • feat: rename file tools to storage_* and view uploaded images
  • File view: the home page is redesigned as the File view.
  • Preferences are redesigned with the shared UI components.
  • Instancing and level of detail: repeated parts are drawn as instances, and far-away bodies with fewer triangles.
  • 12a1e90: DefaultSandbox now selects its provider once, when eve prepares the sandbox, and records the choice in the build: Vercel Sandbox on Vercel, then Docker when a daemon is reachable, then…
  • 62112b5: The just-bash sandbox provider now runs on the public defineSandboxProvider() API, and customCommands and filesystem use eve-owned types from eve/sandbox/just-bash, such as…
  • 8b68a92: Remove the microsandbox sandbox provider. The eve/sandbox/microsandbox export and the microsandbox option on DefaultSandbox.environment() are gone, and DefaultSandbox now selects Vercel…
No release notes published.

Oct 22, 2026

Feynman
  • Two more model errors explain themselves. When a run ends on an error, Feynman names the cause
  • A 404 Not Found from the provider says the model id or endpoint doesn't exist, and to check the model with /model and the baseUrl in models.json.
  • "The model refused to complete the request" says the model declined, and to rephrase or pick another model.
Kortix
  • Move projects between accounts. A project can be transferred to another organization or account.
  • Default reasoning effort per project. Set it once in project settings.
  • Run a session on your own machine with a ChatGPT subscription, through kortixd.
Claude Agent SDK
  • Updated bundled Claude CLI to version 2.1.309
Glimmervoid
  • GitHub actions start sooner and show their progress: Reviews, Issues and PR session checkout actions now run their GitHub reads ahead of background polling.
  • queue: a heartbeat or claim that fails halfway never leaves a claimed issue without a lease
  • CI: a human-applied max-lines:override label lets a PR raise a max-lines baseline ceiling
  • sandbox: sandbox.writablePaths in config.yaml adds directories the sandbox can write (e.g. ~/.nuget)
  • macOS (-sea assets): unsigned (Apple secrets not configured)
  • Linux (-sea assets): SHA256SUMS-sea published with the release
Claude Code
  • Fixed sandboxed Bash commands on macOS and Linux running the shell's startup files outside the sandbox before each command
  • Fixed macOS sandbox denyWrite glob patterns not applying at a path's real location when the pattern's folder is reached through a symlink
  • Fixed sandboxed commands and auto-accepted edits being able to modify the MCP config file passed with --mcp-config
  • Facts about me: a new last section in your profile for what the AI should know about you that the rest of your profile doesn't say, such as why you're changing careers, a work permit, or when you…
  • Facts for one application: the Job posting tab of an application has its own facts, for what matters to that job only, such as having met the team or a reason you want to work there.
  • Editing in Chrome on macOS: clicking Edit on a part of a tailored resume or cover letter crashed the tab in Chrome and other Chromium browsers on macOS, with error code 5.
AI SDK
  • 8e18d59: feat(ai): add experimentalstreamSpeech
  • Updated dependencies [8202f5b]
  • Updated dependencies [85d84a1]
Langflow
  • feat(admin): add team and user administration API and CLI
  • feat(authz): add enterprise directory provenance and CLI
  • feat: add FunASR transcription component
OpenMausBot
  • Your Chief keeps an eye on your day. Once a day, your Chief of Staff posts one short brief in its chat.
  • See where each job's time and money went. The Usage page shows totals for the period, which bots and models used the most, and every run, newest first.
  • X (Twitter) in Apps. X is now a card in Apps, so your bots can search and read X.
Deep Agents
  • deepagents-talon now requires Python 3.14 (>=3.14,<3.15).
  • Added /effort controls for reasoning.
  • Added the NVIDIA OpenShell sandbox provider and an openshell extra for Talon.
  • Nothing to do. The browser extension updates itself once no chat is busy.
  • Chat groups. A project's ⋯ menu has New group. Move a chat into a group from its own ⋯ menu (Move to group), drag groups into the order you like, and rename or delete a group from its menu.
  • Bookmarks. Every finished answer has a bookmark button next to Copy. Once a chat has bookmarks, its header shows them with a count; pick one to jump back to that answer.
  • Fixed paused network requests being lost when the CDP event receiver falls behind, preventing requests from remaining stalled
  • Fixed domain-filtered redirects so open reports an error when the main document redirects to a blocked host
  • Fixed Eve compatibility with the current tool contract and sandbox execution model
AI Memory
No release notes published.
Strands Agents
  • feat(decisions): add decision usage to the request's invocation context (5adc6a6d3)
  • feat(python): add experimental OpenAI Live provider (420cedca4)
  • feat(skills): add docs-verifier skill (0a99337e2)
Openbot
  • Show provider plan limits as rings in the mobile app header
  • release: prepare the mobile app v1.4.1
  • Route channel messages again when the lead uses a free OpenCode model
Reasonix
  • CLI archives built from the Studio release studio-v2.49.0.
  • agent-framework-a2a: Add allowlisted forwarding of function-invocation and client arguments from A2A request metadata
  • agent-framework-core, agent-framework-orchestrations: Add an opt-in synthetic trailing user turn for providers that return empty text after assistant-ending prompts
  • agent-framework, agent-framework-ag-ui, agent-framework-core, agent-framework-declarative, agent-framework-foundry-hosting, agent-framework-hosting-mcp, agent-framework-lab: [BREAKING] Add the MCP…
  • MCP Tool Runner: POST /mcp/server/tools/{tool_name}/run runs any tool listed on the AgentOS Server Card over HTTP and returns its result, so a tool can be tested without an MCP client.
  • AgentOS MCP Server: /mcp and /mcp/server-card now accept browser requests from origins the AgentOS CORS policy allows, so os.agno.com can reach the MCP server of a local AgentOS.
  • Postgres Metrics: Metrics refresh on a migrated database no longer loads the legacy sessions.runs column into memory.
  • Unified SDK: vectorless RAG. Local or cloud. Index or chat.
  • Faster: trees are created from layout stats.
  • Cheaper than vector index: LLMs only write node summaries.
  • improvement: Updated the bundled Copilot CLI snapshot to 1.0.103
  • Clicking a completed response code block copies its original code when enabled.
  • ACP sessions now honor saved sandbox settings and --sandbox/--no-sandbox flags, including on resume.
  • Cloud replies keep full text when the CLI reconnects after a dropped connection
Opik
  • [NA] [FE] fix: make agent graph nodes clickable again with mermaid 11.16
  • [OPIK-8905] [FE] fix: reload once when a lazy-loaded chunk is missing after a release
  • chore(deps): Update dependency sanitize-html to v2.18.1 [SECURITY]
E2B
  • e5a9fa6: Add dnsServers to the network options (dns_servers in Python): the DNS servers a sandbox can reach when allowOut contains a domain.
  • 5a714f0: Deprecate the apiUrl / api_url and sandboxUrl / sandbox_url connection options and their E2B_API_URL and E2B_SANDBOX_URL environment variables.
  • 61b6b55: Use the unified endpoint for sandbox traffic on e2b-tango.dev, e2b-juliett.dev and foxtrot-2.e2b.dev.
  • Fix: security validate_url rejects a fetched URL that has no hostname…
  • Fix: manifest a Maven pom.xml contributes only its own <dependencies> (including those under <profiles>), not the version pins under <dependencyManagement> or a build plugin's own <dependencies>…
  • Fix: js a member call on an untyped receiver (x.foo() where x has no known type) is deferred to receiver-typed resolution instead of binding to a same-named method elsewhere in the file, so a…
Reason Network
  • Messages sent the same millisecond come out in the order they came in
  • Reason Network 0.12.9
OpenHands
  • feat(automations): hide the GitHub token secret field on self-hosted instances when GitHub is connected (OHE-3699)
  • fix(mcp): install OAuth-only library entries through the MCP server's own OAuth
  • fix: keep automation repositories when editing
  • enable_mouse_capture setting (or VIBE_ENABLE_MOUSE_CAPTURE=false) to leave text selection and copy to your terminal.
  • Added max_subagents and max_concurrent_subagents settings to control how many subagents can be alive and running at once
  • In VS Code, connectors that need no login, such as DeepWiki, show as Not connected and connect in one click, instead of asking an admin to set them up.
  • Fix OpenAI models failing every turn with invalid_request_error due to a rejected tool schema
  • Recover the TUI from a vanished or unresponsive server instead of retrying forever
  • Native Muse subscription login and Venice AI provider support
  • feat(mcp): add listLlmConnections tool
  • feat(web): Treemap and Grid views for Topics results
  • feat(ai-gateway): log per-connection socket writes and output completeness
  • deps: exclude compromised litellm 1.107.0 (f6a0231)
  • deps: require certifi 2026.10.18 to distrust the compromised emSign root (9056a4a)
  • evals: typecheck JSONPath extraction against jsonpath-ng (aa4b05b)
  • Docs: page breaks land where Word puts them across a 50-document production sample…
  • Sheets: VLOOKUP over a table or defined name no longer goes blank after a save (reported
  • Slides: chart data labels, data tables, secondary-axis formats, CJK category wrapping and legend placement match PowerPoint.
  • Added project-scoped agent usage aggregation to the bundled CLI, backed by a pageable agent usage HTTP API.
  • Added custom auto routers with client models and local evals.
  • Added a session host registry and Kilo Cloud session adapter.
  • 新增 iphone_use_app_launched 事件,同时发送到 PostHog 和 GA4。记录校验后的 bundle ID、公开目录名称、单次 / batch 来源及前台核验状态;batch 仅统计已成功返回的启动子步骤。
  • PostHog 增加启动次数、使用安装数、占比、每日趋势四个图表;GA4 发布「App 使用分布」报表,包含名称和 bundle ID 数据表及四张分布 / 趋势图。
  • 仅采集显式启动目标,继续排除设备上的 App 清单、画面和内容;支持通过 IPHONE_USE_ANALYTICS=0 或 DO_NOT_TRACK=1 统一关闭统计。插件与 widget 版本同步为 0.3.16。
Ruflo
  • fix(init): stop advertising a memory path the runtime ignores — the init-generated .claude-flow/config.yaml no longer lists the dead memory.persistPath key.
  • init-dual-native-2636-2637.test.ts: 4/4 passing against the built package.
  • audit-umbrella-version-lockstep.mjs, leaf-drift check, audit-leaf-versions-published.mjs: all clean.
  • [embedding].query_max_concurrent defaults to 16 instead of 4. Every
  • Upgraded the bundled Bun runtime to 1.4.4, which includes security fixes for Brotli, TLS, HTTP/1.1, HTTP/2, HTTP/3, WebSocket, and fetch, and updates root certificates to NSS 3.131.1.
  • Patched Node runtime required: OpenClaw now requires Node 24.22.1+ on the 24 line or 26.12.1+ on the 26 line, which reject TLS chains signed with the compromised certificate keys addressed in the…
  • Safer updates and recovery: package swaps now refuse unsafe service states, Windows and macOS package-manager updates recover from platform-specific failures, and frozen Gateway startup owners can…
  • Data-preserving migrations and backups: Doctor and migration paths now refuse malformed or lossy workspace imports, preserve malformed Notes bytes, reject unsafe non-UTF-8 wiki rewrites, keep…
Composio
  • ba7c9a1: Remove the unused @mastra/mcp peer dependency. The provider never imported it, but package managers that install peers automatically pulled in about 234 extra packages.
  • Release: promote main (ad4968a1bb) to stable, 2026-10-22
Codex Security
  • GitHub Action for Codex Security 0.6.0.
  • Installs @openai/codex-security@0.6.0.
  • yaml uses: openai/codex-security@4734c60638bab2ff122953c3773def1d6a844d4a # action-v0.6.0
  • chore: bump the version to v2.2.2post1
  • fix(security): remediate emSign CA trust across dependencies, installers, and builds
  • fix(security): update Node.js to 22.23.4
  • Pi 1.4 support: agents/harness/pi now runs on Pi 1.4.
  • Hibernation while Pi's tasks sleep: PiHarness passes Pi's onWake through your harness factory, so a Durable Object sleeps until Pi's next task is due instead of staying in memory.
  • One wake job per object: PiHarness wakes from Pi's wake(), so a task stopped by Pi's restart limit no longer keeps the object awake.
  • feat(mobile): delete a chat from its long-press menu
  • Send a reply's images and files to Telegram
  • fix: stop sending a custom fetch to Google model adapters
  • A way out that does not need the model. "Exit Andee" / "Take a break" (and the Chinese
  • It is a whole-utterance match, on purpose. "How do I exit Andee", "cancel my subscription"
  • Aiming on an app that hides its element tree is now much better. On WeChat the grid label was
n8n
  • deps: bump sharp to 0.35.5 on 1.x (DEVP-1241) (9fa95f0)
  • [cli-v1] Fixed NocoBase CLI initialization failing with a missing chokidar module during fresh production installations.
  • Security: exclude litellm 1.107.0, which was published to PyPI with a credential stealer and backdoor, from every litellm requirement…
  • Preview works on real pipelines and writes nothing
  • Preview now handles target states under a container (rows in a Postgres / Qdrant / LanceDB table, and so on), so it no longer stops at once on most pipelines.
  • Improved how agent responses read: a blank line now separates the collapsed thought block from the response, and once the response starts, the collapsed thought shows only its Thought for … header…
  • Improved / search in the /diff file view to count matches inside folded unchanged lines, such as match 2/5 · +3 hidden or no matches · +1 hidden, and the f hint now reads show N hidden matches.
  • Improved /diff's Commit Tree to mark the selected commit with the same > cursor and selection color as All Changes and Per Turn instead of a highlighted bar, and made Per Turn start with the…
  • fix(tool): report an empty code_comment batch instead of success
  • fix(mcp): follow tools/list pagination when caching server tools
  • fix(ci): prevent PR policy checkout state pollution
  • commit: 2e3844db004a1a44e9b22e2deaa0d4b264f21e62
  • source tag: beta2026-10-22_11-50-32
  • regions: cn global
  • Compact layout: Preferences › General offers Floating or Compact. Compact joins the workspace panels with single borders, and folded Browser and History stacks give the room back to the view.
  • Show text under ribbon icons: turn it off for a ribbon 24 points shorter, in either layout. Tooltips and command menus stay.
  • Assistant is redesigned: grouped tool results with expandable failures, approval cards, setup guidance and a composer pinned to the bottom. It fills the height of the view on either side.
  • ■ Knowledge Base: Fixed an issue where users with only the workspace administrator role were redirected to the agent list page when clicking Import Document in a Feishu knowledge base.
  • ■ Knowledge Base: Fixed an issue where headings with the same content but different levels in a Markdown file were incorrectly recognized during segmentation.
  • ■ Knowledge Base: Fixed an issue where deleting a document after uploading it to a general knowledge base by a custom role returned an error indicating no permission.
  • Five new Scientific Connectors, plus deeper GDC and Ensembl support. The catalog now reaches HGNC gene nomenclature, PubTator 3 literature annotations, DGIdb drug-gene interactions, RNAcentral…
  • Answer cards tell you the truth about how many answers you can give. Choice questions now correctly show a Single select or Multi select label instead of always implying multiple selections are…
  • Session package dialogs stop jumping around. The .science import/export dialog no longer resizes and repositions itself as steps and counters change — it holds a stable position and sizes to its…
  • fix(security): bump certifi and virtualenv past published advisories (a54daa9a7) — vaibhav upreti
  • docs: move OTel payment pilot from Welcome to Guides (457145176) — Anwesh
  • fix(terminal): meet theme contrast requirements (b6f3f0302) — Anwesh
  • feat(chat): recall sent messages from composer
  • feat(library): 6 Best Workato Alternatives for AI Agents in 2026
  • feat(library): 6 Best OpenClaw Alternatives for Teams in 2026
  • briefing: show the replies you're waiting on in Right now (823b230)
  • delegation: let a workflow answer an email as Handle it for me would, behind an injection check (653d80e)
  • delegation: send a chat draft written as you on your tap (211a484)
  • 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

Oct 21, 2026

  • We have updated our dependencies to mitigate the impact from the recent eMudhra Intermediate CA Compromise.
  • Full Changelog
Qwen Code
  • feat(daemon): support session-targeted Skill and Extension refresh
  • fix(managed-agent): bound initial publication guard waits
  • feat(runtime-broker): Session-create warm for Workspace-isolated bound Sessions
No release notes published.
Claude Agent SDK
  • Updated bundled Claude CLI to version 2.1.307
  • Image input for classifiers: codemode models.classify() accepts images for TypeSafe's Jev and Cloudflare's Clef models on typesafe, cloudflare-workers-ai, and OpenRouter.
  • Models Configuration reference: every provider, model, model override, and compatibility field of models.json, generated from its schema. See Models Configuration.
  • Summaries without a fixed output cap: compaction and branch summaries use the model's output limit, so long reasoning no longer makes them fail. See Compaction.
  • macOS (-sea assets): unsigned (Apple secrets not configured)
  • Linux (-sea assets): SHA256SUMS-sea published with the release
  • #163 0cc33fe Thanks @PazChen! - Desktop: DeepSeek Harness is one of the agents the chat can connect, its dsh program over ACP like the others, found inside its desktop app too.
  • #163 0cc33fe Thanks @PazChen! - The app's own ffmpeg is FFmpeg 8.1.2 on a Mac and 9.0.2 on Windows (was 6.0 and 6.1.1).
  • #163 0cc33fe Thanks @PazChen! - Export › Burn in subtitles: bilingual subtitles show the original and its translation once each, as in the editor. The translation was drawn twice.
Reason Network
  • Relays go through HTTPS_PROXY; a join that reaches no relay says so
  • Relay route follow-ups: one network at a time, per-relay route, LF checksums
  • Reason Network 0.12.5
Claude Code
  • Added /claude-api api-version to move Claude API code to API version 2026-10-21
  • Added the claude_code.safety_event OpenTelemetry event for each tool call a permission rule, auto mode, a safety check, a hook or the sandbox stopped, prompted or sent back
  • Fixed requests failing with a model-not-found error when CLAUDE_CODE_DISABLE_1M_CONTEXT is set and the model is an alias with the [1m] suffix, such as opus[1m]
  • Changes since langchain==1.4.6
  • chore(langchain): update lockfile release(langchain): 1.4.7 chore(deps): bump certifi from 2025.11.12 to 2026.10.19 in /libs/langchain_v1 feat(core,langchain,standard-tests): read tool-call-delta…
Strands Agents
  • feat(decisions): add decision usage to the request's invocation context (5adc6a6d3)
  • feat(python): add experimental OpenAI Live provider (420cedca4)
  • feat(skills): add docs-verifier skill (0a99337e2)
Software Factory
  • draft-pr / --no-draft-pr on factory ship, factory run-issue and factory run-brief, with a repo config default run.pr.draft.
  • macOS sandbox lanes enforce the egress allowlist through the lane proxy.
  • Reconcile and doctor flag lane-labelled issues that carry no state label.
Oh My Pi
  • Added the prompt-cache-persistent catalog axis (isPromptCachePersistent) for providers whose own endpoint never expires prompt caches; logged-in OpenAI Codex declares it
  • Updated the bundled Yolo-Auto catalog to its current models, qwen3.8-27b (now the default), yolo, yolo-small, and yolo-128k, and removed the retired deepseek-flash-v4 and qwen3.8-flash
  • Added the _session/steering ACP extension (advertised as _meta.steering.supported), so clients such as agent-shell can steer a running turn.
  • feat(cdp): support instance-scoped Browser.close shutdown
  • test(cdp): verify Browser.close process exit and profile reopen
  • refactor(cdp): use shared CDP dispatch for browser instance control #1772
  • Added GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5 to the Amazon Bedrock provider. (#64201, #64685; thanks IbrahimKhan12 and alexandergunnarson)
  • Enabled session notices and context compaction for external agents using ACP.
  • Improved the branch picker to make clear that deleting a remote-tracking branch only removes the local ref, and removed the force delete option for remote-tracking branches, since it had no effect.
Ruflo
  • Restored the pair command that the bundled pair-programming Claude Code skill depends on.
  • Fixed a host-dependent test (plugins/ruflo-console/tests/write-argv.spec.ts) that asserted a file's mode without forcing it past the process umask — found while independently verifying an…
  • pair-command.test.ts: 18/18 passing, run directly against the built package.
Noetic
  • feat(core): roll child-context usage up to every ancestor at spend time (c5f760f), closes
Solidcraft
  • Fonts: the first window no longer waits for the system CJK, Hangul and Thai fonts; they're found in the background and added as soon as they're there.
  • Start page: sample pictures and the check for recent files run in the background, so nothing on it waits on files or pictures.
  • Windows: SolidCraft comes to the front when its window first appears, and a restored window keeps its title bar within the screen's work area.
  • Mistral voice for ag2.live: offline and realtime speech-to-text (MistralTranscriber, MistralRealtimeTranscriber) and text-to-speech (MistralTTSConfig, MistralStreamingTTSConfig), on top of…
  • OpenAIResponsesConfig accepts a reasoning option for effort and summaries.
  • New search toolkits: FirecrawlToolkit and KeenableSearchToolkit.
  • Haystack now requires certifi>=2026.10.19. certifi is installed transitively through httpx and posthog.
Hermes Agent
  • Desktop no longer shows suggestion pills above the composer. The pills appeared for no clear reason, mostly nudging you to set up MCP servers. They and the helpers that only they used are gone.
  • Plugins that install nothing skip the dependency consent step. A plugin that declares no dependencies, or whose pyproject.toml installs nothing, now enables without an empty consent prompt.
  • Dashboard: WhatsApp can be safely re-paired after logging out.
  • The golden image of the inputs 1cf048d96945, built by the golden image workflow from ef69f5e19c1b46e41e743234fa5019e39442adcb.
Phoenix
  • report uncaught frontend errors to the server log (2c0873c)
  • deps: exclude compromised litellm 1.107.0 (f6a0231)
  • deps: require certifi 2026.10.18 to distrust the compromised emSign root (9056a4a)
  • Full Changelog
  • c543585: Breaking: replace limits.maxInputTokensPerSession, limits.maxOutputTokensPerSession and limits.maxTokenCostUsdPerSession with one limits.sessionUsage function.
  • baf3030: Attach the agent trace session ID to auto-instrumented child spans on Vercel so Agent Runs can include network and other third-party spans in a session trace.
  • 91b3183: The session input-token limit now counts only new input (uncached input plus cache writes).
  • Prompt mode now honors managed Auto model selection for -p and piped stdin runs
  • Subagents honor custom and MCP tool approvals, avoiding duplicate permission prompts and allowing inherited workflows to run
  • Plugin marketplaces accept git-subdir sources and long plugin descriptions
Billion Context
  • release v0.1.204
  • Revert "#3900: reserve output headroom on every wire (drop the Anthropic exemption)"
  • #3900: reserve output headroom on every wire (drop the Anthropic exemption)
  • docs: v1.0.0 changelog, Memlake launch post, and version-1.0 docs snapshot
  • fix(client): stop the export poll loop from overrunning its timeout on Windows
  • feat/webhook event delivery ids
  • feat(mothership): discover every workspace mcp server without tagging
  • feat(mothership): send workspace skill ids and descriptions in the chat inventory
  • feat(helpscout): knowledge base connector, auto-registered triggers, 12 more tools, and validation fixes
Openbot
  • Add MCP servers to the mobile server settings
  • Add Underdog Bark 3.0 Plus as a local dictation model
  • Put the mobile server settings in sections and align settings text size
  • One assistant from the first minute. A new workspace starts with its Chief of Staff.
  • When a job needs a specialist, it adds one itself, says so in one line, and you can Undo it. When your goal matches a ready-made team, it builds that team for you.
  • In Simple mode, New bot opens the assistant's chat with a request already typed, so you just say what the new bot should do.
  • Update harnesses while sessions run (all phases)
  • Show a live setup checklist while a Chat session starts
  • Move Stay Awake to a sidebar toggle shared with Connect Mobile
Opik
  • [NA] [FE] [BE] fix: block playground runs on an empty dataset or test suite
  • [NA] [FE] fix: stop saying no assertions are defined when a test suite item just has no results
  • [NA] [FE] fix: save to the prompt library only once when Save is double-clicked
Vibe-Trading
  • A retried provider error (a connection reset or rate limit) can halt live
  • Reduce agent request overhead with on-demand tool schemas while keeping the
  • Preserve oversized tool results in readable run artifacts and stop Swarm from
Base44
  • npm: bash npm install -g base44@latest
  • Homebrew: bash brew install base44/tap/base44
  • Version: 0.2.2 NPM Tag: latest
Glimmervoid
  • The review panel no longer lists work that already landed as this session's: when the local main checkout lagged behind origin, the Committed list, the line totals and Worth a look's "Also being…
  • async_mode on /add: accept now, extract in the background. A request with async_mode: true is written to a durable
  • [embedding] dim sets the width of every vector column. It used to be
  • cascade status and /health tell an emptied index apart from a healthy
  • Nothing to do. The browser extension updates itself once no chat is busy.
  • Ask for a picture with /image. Start a message with /image (or pick it from the / menu) and ChatGPT draws it with its own image tool, without Chat On Steroids in the way.
  • Keyboard shortcuts. Ctrl+N (⌘N on macOS) starts a new chat, Ctrl+, (⌘,) opens Settings, Ctrl+Tab and Ctrl+Shift+Tab open the next and previous chat, and Esc pressed twice in an empty message box…
  • delegation: attach up to 15 Drive files to a draft (3558823)
  • memory: file each Learning miss under a finer cause (bcdc874)
  • observability: add opt-in Logfire request tracing (3acb995)
  • Agent controller sessions are keyed by sessionId (scope removed): core/controller APIs, server routes, and client-js controller.session()/session.create() signatures updated.
  • Thread-row query contract changed: use timeRange.traces / timeRange.threads instead of traces.timeRange.
  • querySpans() contract changed: removed model/provider filters and dropped cost, inputTruncated, outputTruncated, errorTruncated from responses (requires updated storage adapters).
  • Postgres Sandbox Checkpoint Store: Added PostgresCheckpointStore for storing sandbox checkpoints directly in PostgreSQL, allowing multi-replica deployments to restore sessions without S3 or shared…
  • Framework Managers as Routine Executors: Company Routines can now use the internal Product Manager or OS Manager as executors through declarations, the Routine API and human-chat configuration.
  • Sandbox Checkpoint Retention: Sandboxes now retain the three newest checkpoints per session by default. Configure retain_checkpoints=None to preserve every checkpoint.
  • feat(skills): simplify CLI install and import setup
  • feat(web): add name search and gateway key filter to organization API keys
  • fix(slack): marketplace review – link buttons and private channel visibility
  • Store calls run in your process. fs, branch and project calls open
  • Workspace API with numbered Sandboxes. workspace.mount(branch) places the
  • Better merges. Entries merge by identity, branch.merge merges by unit,
LiteLLM
  • The cosign claims were validated
  • The signatures were verified against the specified public key
  • fix(count_tokens): backport Bedrock Mantle 404 fallback to stable/1.106.x
  • add Claude 4.5 and 5.x models for Amazon Bedrock (a49636c)
  • enable to source customHeaders from the config file (37b87f1)
  • prompt to auto-pull missing ollama models (e977dc5)
  • 6dadfc7216d0fc8e746442ee890b3b89755b1fa3 MUL-7857 fix(wakeup): evaluate --until-pr against self-hosted VCS pull requests
  • 9a8c3dec4e95deed45d6032be2d6ae94b23a32f5 MUL-7879: fix(wakeup): clear a rule's last_error once a pass commits
  • 3f657e29cf6c70cd2723fdee88647c9f114f68bc MUL-7890 perf(inbox): fetch the unread summary only while the page is visible
  • 6121c737a04cf4f3b170b86b181ae4f85cca2852 feat(helm): probe /api/health with a startup probe and configurable timings
  • f74551ca62a81632b1d4449690554ac873d3dd7a feat: add Logo signup item to hide the logo on the signup page
  • 3ad7552b3fe2aff93eead56b913963bfe432061f feat: add RFC 7009 token revocation endpoint
  • core: Key queue job outcomes by job on shutdown resume (1f074d4)
  • editor: Load credential types on startup (6ff3198)
  • More done from your phone: Create and edit notes, save Markdown on headless hosts, follow relative file links, and see embedded images.
  • Chats you can come back to: Codex chats reopen correctly after undo, Claude keeps its rewound context across Orca rollbacks, and screenshots stay with the conversation.
  • Your work stays protected: Discard pauses autosave of the same file across workspaces.
  • commit: cfb326662233d216f11eed2d494cc555254ac3b3
  • source tag: beta2026-10-21_14-50-01
  • regions: cn global
Reasonix
  • [Desktop · CLI] Go 工具链升级到 1.26.10,修复证书验证对不再受信任的中间证书的处理 — Go 1.26.10 修复了标准库 crypto/x509 的证书验证:遇到已知的、不再受信任的中间证书时会报错(GO-2026-6809)。Reasonix 已用它重新构建;升级之后,漏洞扫描没有再报出 Reasonix 可达的漏洞。
  • [Desktop · CLI] OpenCode Go 连接改用官方文档里的接口地址,DeepSeek V4.1 Flash 也能正常使用 — OpenCode Go 文档里新的推理接口地址和 deepseek-v4.1-flash 之前没有收录,所以 V4.1 Flash 连接可能丢失 DeepSeek 的思考和工具历史回放,新地址也缺少 Go…
  • [Desktop · CLI] Go 工具链升级到 1.26.10,修复证书验证对不再受信任的中间证书的处理 — Go 1.26.10 修复了标准库 crypto/x509 的证书验证:遇到已知的、不再受信任的中间证书时会报错(GO-2026-6809)。Reasonix 已用它重新构建;升级之后,漏洞扫描没有再报出 Reasonix 可达的漏洞。
  • Added Auto mode on Windows 11 version 24H2 and later for personal Google Accounts, selectable via --mode auto, Agent Mode in /config, Shift+Tab, or Remote Control Settings (Agent remains the default).
  • Added whole-file toggle (f) and hunk folding (Ctrl+O) to Commit Tree and Per Turn views in /diff, and consolidated Per Turn diffs so multiple edits to a file in one turn appear as a single…
  • Added _meta["antigravity.google/root_conversation_id"] to MCP tools/call requests so MCP servers receive the top-level conversation ID even when a tool is called from a nested subagent.
  • Desktop app for Windows and macOS: DeepSeek Bot runs on its own, with its own DeepSeek Harness 0.2.1-rc.2 and its own data, so it does not depend on the DSH you have.
  • At its first start the desktop app asks where to keep its data and the Bots' files, and offers to copy your Bot team from DeepSeek Harness; your DSH data is only read.
  • The Windows desktop app updates itself: it downloads a new release in the background and asks whether to restart now, or installs it at the next quit.
  • #4294 3f80896 Thanks @liruifengv! - Add a confirmation dialog when switching models in a session with conversation history: start a new session, compact and continue, continue directly, or cancel.
  • #4188 3f0857e Thanks @oocz! - List plugin skills under their qualified <pluginId>:<skillName> name so same-named skills from different plugins all stay visible and invocable.
  • #4214 ef727e5 Thanks @sailist! - Support streaming large requests and responses through remote control, raising the transfer limit from 10 MiB to 100 MiB, with concurrent transfers bounded to…
Omnara
  • feat: add E2B machine pool provider
  • fix: lock machine pools FOR NO KEY UPDATE so pool locks don't block foreign key checks
  • fix: stop duplicating and escaping tool results sent to the model
Pydantic AI
  • High — HTML conversion in web_fetch: GHSA-4whf-38hw-pxxh. When web_fetch_tool fetched attacker-controlled HTML containing many elements, conversion could delay unrelated work in the same process.
AI SDK
  • 11b886a: Stop forwarding buffered source stream parts after a transform calls stopStream(). This prevents text and errors from appearing after the transform emits its finish events.
  • Updated dependencies [d7b0ffa]
  • Updated dependencies [5a96038]
  • One model path: the Kortix LLM gateway. Every project's models go through the gateway, and custom OpenAI-compatible endpoints are set up in the gateway as custom providers.
  • pi is the default agent harness. New sessions in a project start on pi unless the project turns pi_harness off. Existing sessions keep their harness.
  • One Drive with team folders. Files lives under the gear, and kortix drive works from the CLI.
  • [client-v2]
  • Fix vertical alignment of sequence inputs in v2 filter forms.
  • Fixed disabled styling of relation fields controlled by linkage rules in v2 mobile forms.
  • fix(security): validate OpenSearch connector URLs
  • fix(security): scope original-file S3 uploads
  • fix(deps): dependency upgrades for 1.12.7
  • feat(scheduler): run a private loop as its owner, with their own accounts (a5c1acbd4) — Vaibhav Upreti
  • feat(memory): report the Knowledge page mirror within seconds of a memory write (15638f21f) — Vaibhav Upreti
  • feat(slack): make Slack's native agent stop button stop the reply (dd9b48c18) — vaibhav upreti

Oct 20, 2026

Reason Network
  • Fix the Windows relays test: a closed port takes about 2 s to refuse there
  • README, DESIGN: point agents asked to join at the skill
  • README: sign in and you have a network; same account, same network
  • Linux: the CLI starts again on Ubuntu 26.04 and other systems without libxml2, and CI plus the release builds now check the binary's linked libraries so it can't regress silently.
  • Widgets: after an update, CodexBar replaces a widget extension that is still running the previous version, so widgets resume updating without the manual killall workaround.
  • Memory and CPU: Devin no longer loads a browser's entire Local Storage (about 263 MiB down to 2.5 MiB extra on a large store), hidden OpenAI dashboard pages stop at the refresh deadline instead of…
  • Fixed interactive demo verification failing in packaged desktop apps because required browser runtime files were missing.
  • Fixed Windows browser helpers closing immediately instead of running demos.
  • Applied the existing damaged-escape and doubled-backslash LaTeX repairs to practice problems, feedback, and revealed solutions.
Orca
  • Rewind Claude chats: Use Rewind to here on a message that started a turn, edit your prompt, and take the conversation in a new direction. Claude continues from the history you kept.
  • Your chat, your settings: Each conversation remembers its own model, effort and Fast settings.
  • Keep more of your agent's work: Completed replies and tool results are saved in full rather than cut off at the preview limit.
Solidcraft
  • Threads: hovering a long helical edge, such as a STEP thread, takes 29 ms instead of freezing the window for 7.6 s, and a design with such threads no longer keeps loading forever.
  • IGES import: a revolved arc becomes a cap, not a whole sphere; trim curves with reversed pieces no longer drop their faces.
  • Languages: Dutch, Swedish, Traditional Chinese, Indonesian, Romanian and Hungarian.
Claude Mem
  • Resolves the same project read keys for search results, explicit timeline anchors, query-selected timeline anchors, and selected observation reads.
  • Keeps adopted records visible while excluding unrelated or unadopted records and respecting platform-source filters.
  • Preserves the Smart default three-step workflow and the optional dumb_search alternative introduced in 15.0.0. No API or schema migration is required.
  • macOS (-sea assets): unsigned (Apple secrets not configured)
  • Linux (-sea assets): SHA256SUMS-sea published with the release
  • feat(backends): show full organization names in the backend selector
  • feat(launchers): read public skills from the bundled extensions package
  • feat(agent-profiles): choose skills per profile from the server catalog
OpenMausBot
  • Send to My Cloud happens by itself, chats included.
  • This computer keeps My Cloud up to date: your bots, skills, logins, 1Password, settings and now your chats. It is one way, and nothing is ever deleted on either side.
  • My Cloud's welcome shows "Bringing…" until your things have arrived, then names what came over.
Glimmervoid
  • The needs-you box at the top of the rail is easier to read: instead of one purple "1 NEEDS YOU, 1 FAILED, 1 FINISHED" line that wrapped in a narrow rail, it now shows one chip per kind with the…
  • A remote session card says why its connection ended: when a remote session's connection to its Coder workspace ended, the card showed failed with no reason.
  • Restarting a failed remote session gets it a fresh workspace: a remote session whose Coder workspace failed to come up, or was deleted outside Glimmervoid, used to fail again with the same reason…
  • Clarified that include_hook_events does not suppress SessionStart and Setup hook events, which the CLI emits regardless of the setting
  • Updated bundled Claude CLI to version 2.1.305
  • CI: hash-pinned the release build tools (build, twine, wheel, hatchling) so a tampered or unexpected download stops the job
  • fix(cli): enforce read-only workspace settings in untrusted folders
  • fix(core): avoid duplicate tool response turns when resuming sessions
  • fix(core): prevent deletion of resumed session history on quick exit
  • Added sandbox.allowNestedUserNamespaces so sandboxed commands on Linux can create user namespaces (Chromium, rootless podman); this lets them get around the sandbox's file protections
  • Added CLAUDE_CODE_RETRY_WATCHDOG_USAGE_LIMIT_RECHECK_MS so unattended retry mode (CLAUDE_CODE_RETRY_WATCHDOG) notices a usage limit that was lifted before its reset
  • Fixed every request failing on Vertex, Bedrock and other providers when the endpoint or a gateway rejects tool search, and behind gateways with ToolSearch disallowed or left out of --tools…
No release notes published.
Openbot
  • Apply the new mobile layout with tasks
  • Set the agent memory limit only in Server settings
  • Remove the grid from the landing hero
  • Enabled native compaction for supported Bedrock Claude models
  • Added opt-in reclaimAbovePct to auth.accountPolicies: sessions automatically moved to a lower-priority account return to the preferred one once it has that much quota left
  • Added reachesBudgetedHost(): whether the catalog's host request budgets (inline image bytes, official image count) apply to a request, true on a model's official endpoint and on Factory Droid's…
Sim
  • feat(mothership): let an agent chat list its owner's other live chats for history recall
  • improvement(search): return 30 previews by default for a Search MCP query across every source
  • improvement(changelog): more relevant Search across sources
  • improvement: [All SDKs] internal dependency updates only (SDK snapshot refreshed for newer Copilot CLI builds; no SDK API changes)
Codex Security
  • GitHub Action for Codex Security 0.5.1.
  • Installs @openai/codex-security@0.5.1.
  • yaml uses: openai/codex-security@beac2a77ad1411626f77031294b6efdfd8f6b1e8 # action-v0.5.1
LiteLLM
  • Vector stores: Run db_scripts/strip_vector_store_internal_metadata.py to fix stores created before this release
  • Anthropic: Fixed signed thinking blocks and thinking.block_binding being dropped on multi-turn requests <sup>Also in 1.105.2</sup>
  • Chat completions: Fixed tool calls landing on choices[1] with gpt-5.4+ models, broken since v1.97.0 <sup>Also in 1.105.2</sup>
  • Add native structured output for dynamic workflow subagents on supported models
  • Add assistedApprovalInstructions setting to guide Assisted Permissions confirmation decisions
  • Shell commands that create and then enter a directory now request path approval instead of being denied early.
AI SDK
  • 66e74dc: fix(ai): preserve repaired tool names in streamed UI message parts
  • Updated dependencies [9bec2c3]
  • @ai-sdk/gateway@4.0.120
E2B
  • d7ed073: Remove Python 3.10 compatibility shims (Python >=3.11 is already required).
  • Updated dependencies [d7ed073]
  • Updated dependencies [3c4c7cf]
  • 🔌 MCP registry and gateway: on servers configured with an MCP registry, connect managed MCP accounts in Settings, pick their tools for local hosts or sandboxes through a server gateway with shared…
  • 🎚️ Set reasoning effort for Claude SDK and Codex SDK agents from the web composer, not only for native terminal sessions (#8591, @jsonmp-k8)
  • 🔐 Sandboxed agents can use approved Git repositories over SSH without receiving the SSH key, with optional push access per repository (#9462, @mxatone)
Langfuse
  • feat(topics): one gate for Topics visibility
  • fix(sessions): Don't allow expanding empty reasoning messages
  • fix(web): disable Owner role selection without ownership permission
  • Supervisor OCSF log files are now named openshell-ocsf.YYYY-MM-DD.jsonl instead of .log, so update any collector patterns from openshell-ocsf.*.log to openshell-ocsf.*.jsonl.
  • On Kubernetes, sandbox templates that set pod annotations or labels are now rejected unless an operator allowlists those keys with the new allowed_pod_annotations and allowed_pod_labels settings…
  • The gateway now rejects annotation keys with the reserved internal.openshell.ai/ prefix on CreateSandbox and UpdateConfig.
Ruflo
  • npx ruflo tui: mouse clicking is now on by default — click a page name in the top bar, scroll with the wheel.
  • @claude-flow/cli@3.59.1, claude-flow@3.59.1, ruflo@3.59.1 — latest/alpha/v3alpha dist-tags all verified on the live registry.
  • plugins/ruflo-core: witness manifests support an optional external signing key (RUFLO_WITNESS_SEED/--public-key) for real authenticity.
  • chore(mcp): require graphiti-core 0.31.1 and set MCP version to 1.1.1
  • chore(ci): add the Devin bot to the CLA allowlist
  • docs: state the security scope of the reference MCP and REST servers
OpenFilm
  • #148 0a90640 Thanks @PazChen! - The app's own agent (your own key) keeps a long conversation going when it looks at many frames. Pictures keep their full quality.
  • #146 29346ca Thanks @PazChen! - Sound played faster or slower is heard the moment the speed changes, and sounds natural.
Deep Agents
  • Resume threads in non-interactive mode.
  • Compact large idle sessions before the prompt cache expires and show the prompt-cache expiration time.
  • Support custom Slack OAuth app identity.
  • 💻 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。
  • OS Manager Tool Limits: OSManager takes tool_call_limit (default 8, per coordinator run, including delegations) and member_tool_call_limit (default 24, per specialist invocation).
  • Cartesia Default Model: CartesiaTools now defaults to sonic-3.6. Cartesia sunsets the sonic-2 snapshots, the previous default, on October 20, 2026.
  • Company Routines: Editing a Product's metadata no longer unlinks its Routines. SchedulerTools and Studio can no longer delete, enable, disable or rewrite Company-managed schedules.
  • Fix: Kotlin a typealias declaration is extracted, so a field or parameter typed as the alias resolves to it instead of minting a sourceless cross-file stub
  • Fix: C# a delegate declaration is extracted, so references resolve to it instead of a sourceless stub; built-in Func<>/Action<> stay external
  • Fix: Swift macro declarations are extracted and a freestanding #macro(...) call resolves to the declaration; an unknown or standard-library macro stays unresolved rather than minting a stub
Opik
  • [DND-1779] [CI] Build the sandbox executor image on PRs
  • [NA] [DOCS] docs: one-click install buttons first on the MCP server page
  • [OPIK-8636] [FE][BE] fix: offer missing effort levels and use the new model's default effort on a switch
  • #2560 8c720d3 Thanks @mattzcarey! - Agent Skills read and write a @cloudflare/computer Workspace, for both projection and skills.runner(). See Upgrading to a Computer workspace.
  • #2584 3b3a88e Thanks @ben-reitz! - Browser creates and closes its Browser Run sessions with the binding's typed acquire() and closeSession(), falling back to the HTTP API in fully local dev.
  • #2615 05843e4 Thanks @mattzcarey! - Chat recovery now seals a turn after chatRecovery.maxInterruptedAttempts interrupted attempts (default 5), so a turn that exceeds the Durable Object memory…
Reasonix
  • [Desktop · CLI] 只有思考内容的回复现在会重试一次,仍无答案时给出明确提示 —…
  • [Desktop] 附加超大文件不再让桌面服务退出 — 附加大文件(比如安装包、压缩包)可能让桌面服务退出,因为整份文件会先被读进内存并编码,之后才检查大小。现在文件选择、粘贴和拖入只要文件有路径,就直接按路径附加,不再复制一份;只以数据形式存在于页面里的文件(部分粘贴和拖入)会流式写入磁盘,不再整份放在内存里,每次上传的安全上限默认为 1 GiB,可以在你自己的 config.toml…
  • [Desktop] 手动压缩后可以继续发送,压缩阈值修改后立即生效 — 手动压缩对话之后,下一条消息可能被挡住、无法与模型沟通,现在已修复。在设置里修改压缩阈值会立即作用于正在打开的会话,不需要替换它;修改无关的偏好也不再重启会话的运行环境。已经在运行的子代理保持启动时的阈值,新建的子代理使用新值。会话因其他原因被替换时,你授予的项目之外文件夹的只读访问会原样保留,不会新增任何访问权限。
  • Nothing to do. The browser extension updates itself once no chat is busy. In Firefox the add-on of this release is numbered 2.1.43.1; it is the same build.
  • See what a sub-agent does. Open a sub-agent in the Sub-agents panel and its Activity comes first: its task, every action with its result, and the files it changed.
  • Pick the sidebar layout from a picture. Settings → Appearance → Sidebar layout shows the real Projects and Inbox sidebars side by side. Click one, or use the arrow keys.
  • New enable_auto_approve setting: when the admin-managed config sets it to false, the auto-approve (YOLO) mode is forbidden — the auto-approve agent (and any custom agent with safety = "yolo") is…
  • compaction_max_output_tokens (80000), compaction_thinking (unset), compaction_reasoning_limit_tokens (32000), output_limit_retries (6) and output_token_reserve (4096) settings tune compaction…
  • The app server has experimental session/subagents/wait, session/subagents/interrupt and session/subagents/read methods: a client can wait for the subagents a turn left running, interrupt them, and…
  • Job agent: turn it on in the new Settings → Job agent, give it one search, and once a day it asks your job source for new postings, skips the ones you've saved, dismissed or applied for, and…
  • Score a posting: in Find jobs, Score on the open posting, or the Score pill on an unscored one in the list, rates how well it fits your profile, from 0 to 100, with why and what gives pause, the…
  • Projects: a new Projects section in your profile, between Experience and Education, for side, open-source, course or thesis projects: a name, a link, dates, what you did (one point per line) and…
  • planning_with_files_select selects or clears a conversation's plan, including a plan in another project. Successful named initialization also selects its new plan.
  • Reading a named task_plan.md selects it only for an unbound conversation. Successful writes and patches can switch to one unambiguous plan.
  • Profile-local records survive backend restarts. Hooks, tools, commands, reminders and the completion gate apply request-local selectors without changing the process environment.
  • Automatic conversation translation with Captain or Google Translate. [Enterprise]
  • Conversation monitors, AI automation conditions, and label and priority suggestions. [Enterprise]
  • Refreshed Captain reports, Copilot chat history, and low-credit alerts. [Enterprise]
  • briefing: show the live brief as a Right now card on Home (9121b20)
  • chat: approve held steps with a tap instead of saying go (0bb0d5a)
  • chat: let the Executive change things in the app from chat (f36818b)
  • 修复 v0.5.1 中后续 session commit 生成的记忆无法检索的问题:第一次 session commit 正常,但继续 add message 后再次 commit,会出现任务显示完成、新增记忆却无法通过 find 或 search 检索的情况。升级后,后续 commit 生成的记忆可以正常建立索引并被检索。
  • Fixes a v0.5.1 bug where the first session commit produced searchable memories, but later commits after adding more messages could report completion while newly generated memories were missing…
  • Full Changelog / 完整变更
  • Import .science packages straight from your browser. A download page can now hand a research package to the desktop app in one click — the app receives and validates it automatically, then you…
  • Updates you can follow — and trust, especially on Windows. Downloading, readiness and installation preparation now stay in one dialog with independently scrollable release notes, instead of…
  • Security fixes for MCP OAuth credential handling. Stored OAuth credentials are now bound to the authorization server they were issued for before use, and refreshed third-party security patches are…
  • Fixed OpenCode models failing to load with a missing field npm error.
  • d908e4882d9b10e1477b0eccd47a94d63a0e777c MUL-6710: fix(runtimes): revoke access when making runtime private
  • 55e1ebe66698bc4f3f88b53c804a4042002af9b2 MUL-7019 MUL-7963 fix(wecom): fetch the picture a sender quoted, not just its label
  • 92403438ff188bd20722c17889cf063e183ad74b MUL-7460 fix(daemon): stop sending an English sentence for a cancelled task
Dsh Ops
  • 更新锁释放时由持有者删除 .dsh-ops-update-*.lock 边车:先重新打开名字并比较 (device, inode),只有绑定成功的持有者才能发布,并且只在仍持锁且名字仍指向自己的 inode 时 unlink;成功写入、拿锁后被拒的请求与整目录 replace 都不再留下文件,被硬杀留下的边车会被下一次同名写入复用并清除。 @T-Auto
  • 恢复 ops_bash 的后台任务:run_in_background 不再必然报 EXECUTOR_UNAVAILABLE,改为经 ctx.inject(['jobs'], …) 绑定一次宿主 jobs 注册表并传入工具定义,前台超时也能提升为后台。 @T-Auto
  • 面板采集恢复时先计入已有追加日志或新会话头部的字节,再接纳排队事件;恢复已封顶日志不再追加事件或重复的封顶标记,日志初始化失败即停止该会话采集,JSON 解析丢失记录时标记采集降级与用量不完整并保留有效记录。 @ccch1mneyyy
  • 👌 Add a Send Chat Message action to the Blink block. New Blink credentials use App Identities; existing legacy credentials keep working with Get Users and Send Feed Event.
  • 👌 Allow multi-line placeholders on long Text inputs.
  • 👌 Load Custom OAuth profiles from the OIDC UserInfo endpoint when CUSTOM_OAUTH_USE_USERINFO is true.
  • 🐛 No effect cards are drawn on a software adapter, which froze CI's API 30 emulator
  • 🐛 On Windows the lock covers one far byte, so a refused process can read who holds the folder
  • ✨ The Audio page's Concat shelf - CC0 sounds from the assets repository, fetched when asked
  • Full Changelog
  • 跨会话模型分工(#1626,#1628,#1629,#1631):一个会话可以把任务分给若干子会话,每个子会话单独指定供应商和模型;也可以先看一份方案,再决定要不要启动。分工列表移到了「活动」栏,展开就能看到每个会话用哪个模型、跑到哪一步。已启动的会话也能换模型,可以选「本轮结束后切换」或「立即暂停并切换」,切换没生效前会话照常用原来的模型把当前轮跑完;一次最多修改 8 个已启动会话。
  • Token 用量页重做(#1612,#1618,#1627):页面开头直接给出生命周期总量(中文按万/亿、英文按 k/M/B 显示),下面是按模型拆分的 90…
  • 精简 Agent 模式(实验性)(#1613,#1619):桌面「设置 → 通用」可以选择标准或精简,新建会话在发出第一条消息之前也能临时选。精简模式使用更短的系统提示词,只保留读写文件、搜索、终端等核心工具,连接器(MCP)、子 Agent、智能体团队、计划模式等一律不可用,技能仍可按需调用。模式在会话创建后锁定,想换只能新建聊天。
  • Pause all automations: one account switch stops every project's schedules, webhooks, app events and monitors.
  • Trigger run limits: a trigger runs at most 60 times an hour by default, with an optional daily limit, so a runaway trigger stops itself.
  • Drive in the sidebar, with a Drive page, a file viewer, upload, multi-select and file management. Session folders show Drive folders. Repo moves under Customize.
  • Support compact mode for the desktop interface
  • Improve startup retries when the workspace is locked
  • Exclude manually typed closing brackets from block reference search
Pydantic AI
  • Medium — MCP request telemetry: GHSA-xv4x-g8mx-ch56. Provider-native MCP configuration could expose server URLs, authorization tokens and custom header values in OpenTelemetry request traces, even…
  • Added / search to the /diff file list and file view: matches are highlighted as you type, the search prompt shows how many matches the diff and each file contain, Enter jumps to the first match…
  • Improved agy plugin validate to also check a plugin's hooks, sidecars, and component lists: it reports an error when hooks.json, a sidecar.json, or a skills.json, rules.json, agents.json, or…
  • Improved timer labels in /tasks and the active task list to show durations in days, hours, minutes, and seconds, such as 1d 2h timer or 5m timer, instead of raw seconds such as 93784s timer.
  • feat(allowlist): add Shader Languages support
  • feat(language): support for Nushell language
  • feat(rules): add PowerShell review support
  • test(providers): compare guard paths with POSIX separators on Windows
  • feat(arcmira): add YouTube transcript search provider
  • feat(providers): add TwittrAPI provider
  • commit: 8bd60e8549cdcf0ada21e6f73d68dc0d0b71909d
  • source tag: beta2026-10-20_10-16-06
  • regions: cn global
  • sdk: read project artifacts as UTF-8 (87b438a)
  • fix(shell): reopen a closed /choose menu instead of dropping the question (6d317f4d0) — Vaibhav Upreti
  • fix: remediate GitHub code scanning finding #2420 - - Removed the redund (358660f3a) — Vincent Hus
  • fix(scheduler): let a free-form agent loop pass when its task is done (5fcd3c9de) — Vaibhav Upreti
  • Memory as one row per fact — semantic recall across authorized notebooks, plus a nightly tidy.
  • Goals that handle getting stuck — a governor steps back, then pauses; people can resume or cancel.
  • Reply to Slack conversations from the web UI, with unread dots for new agent replies.
  • e4e0cd1: Fix activation trace-content markers to respect the resolved session policy, provider capture settings, and live audience ceiling.
  • 54d63fd: A session keeps accepting messages during a deployment or compaction handoff, so the first message after a handoff reaches the model sooner and messages sent mid-handoff no longer retry…
  • fc4dbd0: Cancelling a turn while the model is responding no longer marks its agent.step span as an error, so user-initiated stops stop showing up as failures in error tracking.

Oct 19, 2026

Omnara
  • ghcr.io/omnara-ai/omnara-web@sha256:23b3a97b325ec31b1fd79a5fa7e63a897782902a033946754c2d62fa9d8d747a
  • ghcr.io/omnara-ai/omnara-api@sha256:ad63277220f21ef6f6e1948251caf78782c30dfe85b8fd4d276c66e6d6c64f39
  • ghcr.io/omnara-ai/omnara-worker@sha256:c4b3f68e73b5f264edbcdf5b557e983ec676ca61dce25a5dd9ffc2942e86674e
Glimmervoid
  • Finished sessions are green again: a session that finished a turn you have not looked at shows as Complete with a green check and a green row highlight, instead of looking the same as an Idle…
  • The browser tab icon only marks what needs you now: it shows a question mark when a session needs input and a red X when one failed, and otherwise the plain Glimmervoid mark.
  • The needs-you readout stays visible when all is clear: the box at the top of the rail keeps its full size, outline and bright ALL CLEAR text when nothing needs you, and shows the Cmd+J hint…
Ruflo
  • agentdb_graph-query/agentdb_graph-pathfinder resolved the memory DB path ad hoc instead of through the shared resolveDbPath() priority order, so an explicit CLAUDE_FLOW_DB_PATH binding could…
  • backupMemoryDb() fell back to a raw file copy after any native online-backup failure — including ordinary SQLITE_BUSY/disk errors on an unencrypted source — and reported backedUp: true regardless.
  • JsonMemoryBackend.isExpired() used a falsy check on ttlSeconds, so an explicit ttl: 0 never expired (cli-core 3.7.5, #4295, #4296).
  • #15235 c61a974 - Let the browser_open permission prompt remember your choice. The prompt now offers the current origin and all origins as always-allow rules, and Auto-Approve settings list…
  • #14866 f01b273 - Let a primary agent choose which agent runs an Agent Manager session it starts, so a session fanned out through Agent Manager can run a custom agent…
  • #15119 cbdb045 - Redesign the Agent Manager New Worktree dialog. Pick the base branch next to the project, set an optional branch name next to the worktree name, and choose versions next to the…
  • fix(web): restore overlays after preload recovery
  • style(web): remove border from question card
  • Fix universal native dependency packaging for desktop v0.3.1
  • macOS (-sea assets): unsigned (Apple secrets not configured)
  • Linux (-sea assets): SHA256SUMS-sea published with the release
  • Forget sessions gone for a day, with the mail they never read
  • fix: preserve Windows UNC home pipe names
  • defang: a digit, @ or $ counts as its letter or as nothing
  • feat(library): How Do You Connect a Google Drive MCP Server to an AI Agent in Sim?
  • feat(library): LLMs vs AI Agents: What Is the Difference?
  • feat(greenhouse): connect Harvest v3 with a customer-owned API credential
OpenMausBot
  • Cloud memory roughly halved under load. In our test of a Chief handing work to four teammates who all browse, peak memory on a 4 GB Cloud fell from about 3 GB to 1.6 GB.
  • Bots on OpenCode share one engine process instead of starting one each.
  • OpenCode reaches MausBot's own tools directly, without a helper process for each tool in each chat.
  • Added a transcript line when a plugin that runs first skips one of your mod's hooks, naming the event and the plugin
  • Added decider: "person" to a mod's tool.check answer: only the permission dialog or a host's callback can approve the ask, and it is refused with the mod's reason where nobody can be asked
  • Added e.isReadOnly to a mod's tool.check event, set when the tool call is read-only
OpenFilm
  • #110 7d35283 Thanks @PazChen! - Desktop: agents work for people who installed only the maker's app: Claude Code and OpenCode from inside their apps, and an Install button that runs the official…
  • #98 f771d05 Thanks @cristinamadureira353-svg! - Studio: an export can be named where it is asked for — a name field beside the folder it goes to, the project's name in it to begin with — instead…
  • #106 3582b88 Thanks @taavie-z! - Keep Final Cut Pro subtitles white with a black outline and a soft shadow.
Hive
  • MCP servers agents run from the terminal. An MCP server's tools load into every agent's tool
  • hive mcp add takes --env, --cwd, and the server's command after --, so a command
  • extra_body for worker_llm and each queen's own model. Only the main model sent
AI SDK
  • af9972b: fix(ai): stop emitting buffered stream parts after a streamText transform calls stopStream
  • Add --disable-client-telemetry flag to turn off client telemetry uploads from the CLI
  • Add --trust-folder for one-off prompt-mode sessions started with -p or piped stdin
  • Install verified Skills and plugins from GitHub install links via copilot plugin install "ghapp://air/install?..."
  • AgentTool gained the optional approvalCacheTargets hook, which the always-ask-cached approval mode uses to check and record per-file grants
  • Tools with lenientArgValidation now receive unknown root arguments that their closed schema would otherwise drop, so they can reject them
  • Added validateToolArguments(tool, call, { keepUnknownRootKeys: true }) to keep unknown root keys that a closed schema would otherwise drop during argument repair
  • Disk and network safety: Claude and Vertex cost history now writes only the project that changed, cutting cache reads and writes by about 99% for large transcript histories.
  • Claude: Desktop Cowork sessions now count in token and cost history; codexbar usage follows the active claude-swap account like the app does; Linux Auto tries OAuth before starting the CLI.
  • Accounts and menu bar: Overview can show one row per account for claude-swap, managed Codex and token accounts; saved token accounts can be edited and reordered in place.
  • Base resume language: a new setting in Settings → Documents picks the language of your base resume's headings and dates. It's English unless you change it.
  • Headings and labels in the document's language: a resume written in German had German content under English headings and labels: PROFILE, EXPERIENCE, "Present", "Native".
  • Calendar: the week and the month now fill the window instead of letting the page scroll past them, like the board.
  • feat(trace): Move messages tab and reuse session timeline
  • feat(api-keys): enable project API key permissions by default
  • feat: rename Langfuse Assistant to Halo
OpenHands
  • chore: bump agent-server to 1.56.0
  • chore: remove merged PR artifacts
  • phoenix-sqlean: bump bundled sqlean to 0.28.5 (5e0262e)
  • 6d4325dbe34babe55c1bbe3a28ce3e3639ab45c9 feat: accept client ID metadata document URLs as client_id for MCP clients
  • 6bb563d30802f3be0f73a66c8a9f337cc7248422 fix: default the OIDC user ID mapping to sub and fall back to sub when the mapped claim is missing
  • c5029ae761c742cd49582de78b459c74b353c6d1 fix: expire tokens and clear the forced password change when update-user changes the password
  • The app-server reports each model's compaction threshold, so a client can show how full a session's context is before it compacts.
  • Desktop can import projects and sessions from another coding agent during onboarding.
  • Rust CLI: /leanstall and /unleanstall install and uninstall the Lean agent.
  • fix(terminal): copy from agents, stop claiming TERM_PROGRAM=kitty, keep keys typed while attaching
  • chore(mobile): publish OTA updates to production on every merge to main
  • feat(marketing): list recent changelog entries in llms.txt
Openbot
  • Show the settings search results on the left
  • Remove most exactOptionalPropertyTypes type errors
  • Move the mobile chat to Legend List for smoother scrolling
  • Docs: drag a table row's bottom border to set its height and the corner handle to resize the whole table like Word; a table selected via the move handle highlights every cell (reported
  • Sheets: embed a picture in a cell as an Excel rich value (@tokg-venv, requested
  • Slides: Slide Show > Record narrates the show slide by slide (@MidsummerYU); every provider builds pages locally.
Opik
  • [issue-4521] [SDK] Fix @track not detecting async functions wrapped by multiple decorators
  • [#9002] [SDK] fix: log stream usage and output on ResponseFailedEvent
  • [issue-8010] [FE] [BE] feat: view an old dataset or test suite version without restoring it
  • e3d1497: Add github.attribution(ctx) to eve/extensions/git and the deprecated code({ github }) config.
  • 5c53f1f: An approval answer that is neither Approve nor Cancel now gets feedback instead of being dropped, and a repeated answer to an approval that was already decided is discarded before the…
  • 4bc7106: Stop reading the parent session's event stream once the remote subagent stream route finds the child's agent.started binding.
Agno
  • Eval Case Run Arguments: Case takes session_id, user_id, session_state, dependencies, metadata and knowledge_filters, forwarded to the agent or team run, and input accepts a Message, so a case can…
  • OS Manager Docs Access: OS Manager queries the Agno documentation directly for framework questions instead of delegating to Platform Manager, and keeps delegated tasks scoped to the question.
  • Result Offloading Search and Readback: search_result returns separate passages from one long line, such as single-line JSON, instead of only the first match, and read_result takes max_chars to…
OpenViking
  • Gateway 可以直接运行在 OpenViking Server 内。 开启 gateway.enabled 后,模型请求和管理接口统一位于 /gateway;独立部署仍受支持。Studio 新增快速配置、从 Server 导入模型设置和真实模型生成测试;上游支持精确 endpoint、自定义认证头及 URL 中的 key 占位符,覆盖 Azure…
  • 隐私配置支持字段级引用和原子更新。 ov privacy reference 默认返回绑定 owner 的引用,resolve 才显式读取值;新增字段元数据、字段/目标删除、原子 PATCH,以及对已存内容的扫描和修复。扫描默认 dry-run,返回引用而不返回命中原值;多个字段登记了同一个值时报告歧义。修复单文件 no-split 导入的隐私处理(#6161、#6260)。
  • 批量写入支持更多处理选项。 batch_write 新增整批 processing_mode,每个文件可带 tags 和 ACL;增加内容版本冲突检查及保留冲突文件的选项。变化文件统一维护语义与向量,vectors_only 不刷新父目录语义。正文已写入后,后台刷新入队失败会如实报错,正文不会回滚(#5897、#5750)。
  • The golden image of the inputs a61faff1b2d9, built by the golden image workflow from ef69f5e19c1b46e41e743234fa5019e39442adcb.
  • Added Gemini 4 Argon to the Google AI models
Andee
  • A swipe moves as far as the finger did, and stops there. It used to lift mid-motion, so the list
  • The tap right after typing into a reply box now has to be aimed. On an app that hides its tree,
  • submit_input stops claiming success on a box that cannot be submitted that way. WeChat's
  • Copy workspace paths and open installed terminal apps
  • Full-screen, editable artifact viewer with selection feedback
  • GitHub organization members see the App repositories they can access
  • fix: escape pck-banner session ids before innerHTML (#2645, found in #3781 release review)
  • gate: give claude's small-window preflight marathon a 600s turn budget
  • fix: pin snapshot-render D-lane test to unified shadow mode
  • A damaged auth.json says so. When ~/.feynman/agent/auth.json isn't valid JSON, for example after a hand edit, every command used to print only Expected property name or '}' in JSON.
  • feynman --update points to feynman update, and feynman model setup points to feynman model login.
  • [app] Fix APP_CLIENT_ENTRY_MODE redirects not working in dev mode when APP_PUBLIC_PATH is not /.
  • [evaluators] Fixed a math.js expression sandbox escape in workflow calculation nodes by upgrading mathjs.
  • [server] Fixed the plugin dependency compatibility check for Axios 1.20.x.
  • npm: bash npm install -g base44@latest
  • Homebrew: bash brew install base44/tap/base44
  • Version: 0.1.33 NPM Tag: latest
Qwen Code
  • fix(managed-agent): admit team_plan_approval into the journal-only interrupted-turn settle
  • test(managed-agent): add the FG8e/FG8f monitor_run and background-shell fault gates
  • feat(extensions): add bounded generated failure explorer views (PR4)
n8n
  • core: Reuse connections in the log streaming webhook destination (b79afe2)
  • core: Turn off expression spans by default (6965193)
  • Added background workspaces, hidden from the sidebar unless Show background is on
  • Added a Send feedback button to the diff pane, so review comments reach an agent without opening its chat
  • Added a fullscreen composer on phones
  • Added support for clients sending files and images to the bot via the new RTVI send-file message, incorporating them into the LLM context.
  • New frames: FileRawFrame, InputFileRawFrame, and UserFileRawFrame, carrying a file payload as inline base64 bytes or as a URL.
  • New LLMContext methods: create_file_message(), create_file_url_message(), and add_file_frame_message(), producing universal file_base64 (base64 data URL) and file_url content items translated per…
  • Native chat (experimental): Use Antigravity in chat, choose which agents open as chats, and install a missing agent from the chat itself.
  • Accounts & agents: After switching Claude accounts, restart running sessions on the new account.
  • Phone: Edit queued messages in place, follow tool runs with desktop-style cards, and see your conversation immediately when resuming from history.
  • Extract the entire archive into a writable directory. Windows: Start GenericAgent.cmd; Linux: ./Start GenericAgent.sh.
  • Windows requires Windows 10/11 x64 and Edge WebView2.
  • Linux requires x86_64, glibc >= 2.36 and a graphical desktop. Tested in Debian 12 with Xvfb. If FUSE is unavailable, use APPIMAGE_EXTRACT_AND_RUN=1 ./Start GenericAgent.sh.
  • Unified SDK: vectorless RAG. Local or cloud. Index or chat.
  • Faster: trees are created from layout stats.
  • Cheaper than vector index: LLMs only write node summaries.
  • Nothing to do. The browser extension updates itself once no chat is busy.
  • Bring back your earlier messages. In an empty message box, press Up to bring back the last message you sent in this chat, and again for the one before.
  • Answers read calmly. While ChatGPT writes, the answer shows whole paragraphs and fades each new one in, instead of jumping and reflowing with every word.
  • Trigger run history. Every trigger keeps a list of its runs with status, error and end time. You can test-fire a trigger with a payload, replay a past run, and missed cron slots are bounded.
  • More credit tiers and larger top-ups. The Free plan now includes 500 credits and the managed models.
  • DeepSeek V4.1 Flash now serves the 1017 snapshot on Morph (US region, zero data retention).
  • 原生文件工具面从四项扩到十一项:新增 ops_write_file、ops_edit_file、ops_diff、ops_stat、ops_copy、ops_move 与 ops_remove,统一使用单一 path 和 SHA-256 revision/expectedRevision 的 CAS 契约,写文件与编辑支持 dryRun 预览。 @Thatgfsj
  • 新增安全与原子性层:替换回执基于实际观察到的字节,合作写者共享持久路径锁,Windows 保留 owner/group/有序 DACL 与受支持属性,Darwin 非空扩展 ACL 在 staging 前以 METADATA_UNSUPPORTED fail-closed,并用…
  • 新增默认关闭的 dsh-ops/hide-official 组件:只接管 shell/file/ptc 三个面,隐藏有经审查 OPS 等价物的官方工具(read、glob、grep),其余官方工具保持可见并汇总为一条告警。 @Thatgfsj
  • sdk: read project artifacts as UTF-8 (87b438a)
Reasonix
  • CLI archives built from the Studio release studio-v2.45.0.
  • [Voice] Add Kokoro Chinese-English TTS with 103 voices for Windows x64, Linux and macOS, and Qwen3-TTS for Apple Silicon. Download models from Settings > Local Models before first use.
  • [Voice] Show live audio spectra in speech playback and dictation controls. Restore the microphone after empty dictation.
  • [Voice] Add local FunASR speech transcription on Windows, Linux and macOS.
  • Orbit: orbiting about a picked point keeps that point under the cursor in perspective views too; before, it drifted off and the model swung about the camera.
  • feat(scheduler): readable loop reports on every channel, and email delivery (8eba8c930) — Vaibhav Upreti
  • fix(billing): name the acting member on hosted credit spend (b3b132deb) — vaibhav upreti
  • feat(security-fix): push a fix straight to the default branch when asked (6dddbff57) — Vaibhav Upreti
  • Automatically retry turns on the next pool account when one hits a usage limit, including for Desktop and SDK clients
  • Hibernate idle sessions to free memory, with measured RAM status and transparent wake on the next interaction
  • Reload and resume large sessions much faster by showing the local transcript immediately and sending only new history

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.