AI agent releases: Sep 21 – Sep 27, 2026
426 stable releases from 126 tracked agents and frameworks in the week of Sep 21, 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.
Sep 27, 2026
- Opening a session shows its saved history immediately from cache while the live transcript loads, in one request instead of several.
- Turns start faster: configuration, model catalog and ingress checks run in parallel.
- Sessions retry transient model errors inside the turn and continue through short provider outages instead of stopping.
- Security: the Fortran capital-F cpp step no longer allows an untrusted .F/.F90 source to read arbitrary host files.
- Security: the Aider/Devin monolith --watch snippet no longer interpolates the agent-substituted INPUT_PATH into a shell command — it now reads the trusted graphify-out/.graphify_root written in…
- Security: Terraform secret redaction now also covers a value paired with a secret-named name in name/value pair lists…
- Voice notes play as audio bubbles right in the transcript.
- Snooze a thread until there's new activity or until a time you pick.
- Pin threads. Threads are listed in update order and show when each one last changed, in your time zone.
- 侧边对话:在输入框敲 /btw 问题(或选中文字后点「在侧边聊天中提问」),会在工作区右侧开一个临时对话,继承父会话的上下文和模型但独立运行,追问不会打断主任务。关闭标签或退出应用即清空。
- Agent Teams 启动前人工审核:团队方案就绪后先出「团队执行前确认」卡片,逐个成员改 Agent 预设、服务商、模型、任务归属与前置任务,可批量设置或「跟随主 Agent」,确认后才真正启动成员。启动前的成本由你把关。
- 超时自动回答:「设置 → 通用 → 自动回答问题」开启后,待回答的问题超过设定时长(1/5/10/30 分钟,默认 5)会由该会话的小模型基于对话上下文自行决策;判断不了就继续等,你任何点击或输入都会取消计时。
- Share plugin and motion helpers between the app and the web
- Make the server menu modal, and give Gemini a profile client with no tools
- Share the stagger CSS and the preview message types between the app and the web
- fix: arm emergency shrink on outbound payload size, not raw body
- fix: report the surviving raw-id count in the native-compaction boundary log
- fix: route unparseable compress tool args to the lenient parser instead of degrading to {}
- 侧栏可按账号自定义;角色模板 id 持久化
- 暗色主题下知识库 Markdown 预览不可见
- README 路线图与产品链接更新
- Removed the stream-level Anthropic prompt-cache keep-alive: StreamOptions.anthropicCacheRefresh, StreamOptions.anthropicCacheRefreshRequest, and the zero-output refresh request path.
- Added the openai web-search grounding for OpenAI API models that support Responses web search (gpt-5.5, gpt-5.6-luna, gpt-6-astra, gpt-6-luna)
- Added Model.promptCache, per-retention-tier prompt-cache entry lifetimes in seconds (short / long), declared per provider through the prompt-cache KDL rule (bundled: direct Anthropic, 5 min / 1 h).
- Codex account switching works end to end: a System Account switch now restarts the running app-server daemon even when its control socket is a symlink, and Usage Dashboard opens the current…
- A steadier menu bar on macOS 26.6: no blank slots after quitting, stable status-item identities through recovery and updates (fewer duplicate Bartender entries), and opt-in startup diagnostics for…
- One-click Homebrew updates for cask installs, with Homebrew staying in charge of the installation. Thanks @Yuxin-Qiao!
- npm: bash npm install -g base44@latest
- Homebrew: bash brew install base44/tap/base44
- Version: 0.1.21 NPM Tag: latest
- A connection can be a ChatGPT or Claude plan (subscription) instead of an API key.
- Claude plans use an operator-issued Anthropic OAuth app (EXP_ANTHROPIC_OAUTH_*). Claude Code's client identity is refused.
- Plan usage windows (x-codex-*, anthropic-ratelimit-unified-*) are harvested. A used-up window throttles the rung until its stated reset, so traffic rotates to the next account.
- Works with ChatGPT's newest conversation view. Some accounts got yet another new layout this week, and CoS stopped seeing their chats. They are recorded again.
- Goal works again with the new Temporary Chat design. CoS recognizes ChatGPT's updated temporary-chat button and dialog, which Goal uses to decide the next step.
- New chats and workers start reliably even when ChatGPT swaps out the text box while the model is being selected.
- Subagents keep their search tools in the foreground. A researcher or verifier launched in the foreground failed at once because foreground children never load the parent's extensions, so…
- No alphaXiv account, no alphaXiv search tools. Without an account, every alpha_search, alpha_get_paper, alpha_ask_paper, and alpha_read_code call failed, and research runs retried them.
- The Windows install docs now say Feynman needs Bash from Git for Windows for shell commands.
- QM has a desktop app, with browser sign-in handoff and the download in Settings.
- Admins can see where the money goes: a spend dashboard with time-series breakdowns, reports served from saved daily totals, and per-cron runtime overrides plus separate org defaults for crons and…
- Inbox Loops arrive: flagged loops with durable event ingestion, each with its own icon or uploaded logo.
- Improved scrolling in the full-screen view: bursts of mouse-wheel or trackpad events are now coalesced before redrawing, which cuts CPU use and reduces scroll lag in long conversations and the…
- Improved GEMINI_API_KEY sessions to stop immediately when the Gemini API reports an exhausted daily quota, a project or billing-account spend cap, or depleted prepaid credits, instead of spending…
- Fixed resuming a long conversation in medium or low verbosity taking many seconds and lagging while history replayed: earlier tool groups now appear already finished instead of each re-animating…
- feat: let users choose JSON or SQLite when profile copies diverge
- attest-plan.sh can now attest a root roadmap while a named ticket plan remains active. Use --target root or --target <plan-id>.
- PowerShell named-plan initialization now retries the transient active-pointer pre-check failure during concurrent replacement.
- Closes #294 and #296 through PRs #295 and #297.
- feat(cli): allow skipping startup onboarding questions (a5fcc04d5) — Anwesh
- Improve /resume with a scrollable conversation picker (7ae39a0e4) — Devesh
- Docs: a large Word-parity batch — spelling suggestions and AutoCorrect, Find & Replace with wildcards and regex, a Navigation pane, rich header/footer editing, ruler indent and margin handles, the…
- Sheets: Ctrl+drag on the fill handle fills a series from one number; lowercase formula references shift correctly (@GoodScholar).
- Slides: the audit flags off-slide and stretched elements; long decks no longer keep one canvas per slide (reported
Sep 26, 2026
- Desktop Capture and gateway capture reuse the same native response assembly.
- Preserve provider response evidence, usage, tool content, and exact large integer values for desktop uploads.
- Retain credential redaction and keep desktop-only source annotations out of gateway records.
- feat(ipynb): render notebooks like a notebook, with seamless click-to-edit cells
- feat(ipynb): run notebook cells in a persistent Jupyter kernel
- feat(ipynb): create a .venv when pip is locked out, and show ipykernel setup progress
- Starting a new chat works again. CoS used to open a ChatGPT tab and then never send your message. The first message of a chat started from the app is now sent and confirmed properly.
- Your chats are recorded again. Messages, answers and tool calls show up in the app as they should, including for chats you start directly on chatgpt.com.
- Goal and Loop continue automatically again. CoS reliably notices when ChatGPT has finished an answer, so the next step starts on its own.
- Isolate tests from the shell, mask registered secrets, and trace main-process crashes
- Use typed request timeouts and one Team API codec lookup
- Keep agents working without check-ins, and remove CLAUDE.md imports
- proxy: stop the 0.39.0 TPM limiter from refusing large-context requests forever (7968122)
- 适配 DSH 0.1.7-rc.2 并保留 rc.1 兼容线(#1000,也修了 #992 的 peer 跳过)
- session/title 载荷补齐 messageSeqs/source:fork 与改名不再写出损坏日志(#1010 / #1006)
- inline 静置不再向终端泄 LF,选区复制不再错位(#993);光标已在目标位时不再发空 patch
- Grouped outputs. When the agent shows several results in a row, such as two screenshots, the session renders one tabbed card instead of a card per result.
- Managed model routes. Each Kortix-managed model routes through its own provider route, and the model picker shows the price for each route.
- Wake failures are explained. When a sandbox fails to wake, the session page shows the failure and when it retries.
- fix(cli): publish layout before fetch eval with --layout #868
- fix(cdp): expose viewport metrics before layout publication #870
- fix(screenshot): keep viewport fallback out of DOM geometry #871
- 1278f9dcf4dc6c32c7783e8e2d477686bbe629b9 fix: bind MFA remember to device and block avatar SSRF and forged payment notifies
- 617995563e6fcb835c582d644909ff7fd38ec78b fix: block cross-org API permission fallback on empty owner
- a0701ad84bd50c5e62fd68807b92923d1991ddac fix: harden DCR, session rotation, Kerberos realm and XSS sinks
- Provider resolves instead of aborting. A fresh non-interactive install with no --provider defaults to your Anthropic plan (local memory).
- Saved personal providers are validated, not assumed. A kept gemini / openrouter provider needs a usable key: saved in ~/.claude-mem/settings.json, or exported in the environment.
- Deferred sign-in link. Every non-interactive install that skipped sign-in prints an optional sign-in link as its last line, after the success line, so an agent relaying the output can show it to…
- Benchmark, method and code: evals/context_filter/
- Python 3.12+ is now required.
- Default context filter: without a TypeSafe key, the default is now keyword ranking instead of
- fix(search): read Confluence pages through the v2 API the Search grant allows
- fix(tabs): preserve readable labels in crowded tab strips
- fix(search): align landing layout and history rows with Home
- Security: writeAtomic path traversal and host-adapter injection, .env guard anchoring, MCP detection in settings.json, Grade A no longer masks mcpRisk High, repo-dependent toolSafety, reward-hack…
- Correctness: promotion-rule lockouts, sequential evidence wired into run/replay, flywheel graph RangeError, host adapters keep HarnessSpec.autonomous, traceQuality scorer wired, learn dry-run…
- Published: metaharness 0.4.17, @metaharness/darwin 0.10.3, flywheel 0.1.12, weight-eft 0.1.2, and patch releases of nine host adapters.
- feat: chain checkpoint step 2 — shadow recognition (parser, JCS digest, carrier contract)
- fix: compress success receipt gains remaining-range snapshot + stop signal
- fix: downgrade ACP body-artifact detection to advisory
- Feature: five language extractors gained structural depth — OCaml classes now emit their methods (via the method relation) and instance variables
- Fix: a TypeScript "solution" tsconfig.json that only carries references (no paths of its own) now resolves path aliases declared in the referenced project configs, so alias imports in a tsc -b…
- Fix: extraction now skips the process pool up front when the spawn start method can't re-import __main__ (stdin, python -c, REPL, embedded callers), falling back to a correct sequential run…
- hypit packages status <name>@<version> now answers the question a Build asks. It read the machine package home and nothing else, while the loader finds an external package in two places — the…
- FEYNMAN_HOME=~ or ~/path now means your home folder. Unexpanded ~ (common in .env files, Docker ENV, and Windows shells) created a literal ~ folder in the current directory.
- Starting two Feynman sessions at the same moment no longer crashes one of them with "Subagent config changed during settings normalization".
- chore: refine pii review instructions
- bithuman: resolve the SDK on Python 3.10 and 3.14 again
- add pid to log lines from parent process
- fix(presets): install statuses from the organization library; stamp preset skills (follow-up to
- fix: mise-managed CLI PATH resolution + real error text in coordinate_bots dispatch
- feat(ios): audio bubbles for voice notes
- fix: align model field test fixture with provider schema
- docs: add changelog entry for September 24, 2026
- chore: point default models at the refreshed hosted catalog
- Paginated document review. Preview gains paginated Office reading with search controls and paged PowerPoint review, so long documents and decks read comfortably inside the app.
- Enrichr gene-set enrichment. The genes connector adds Enrichr tools: browse enrichment libraries and score your gene set against them.
- Compact library preview. The workspace sidebar gains a compact library preview for quick scans without leaving your flow.
- fix(gateway): make the WebSocket tunnel opt-in
- test(binary-identity): stabilize procfs identity fixtures
- docs(tutorials): run Pi with OpenRouter
- Settings has a new About page. It shows your version and channel, has Check for updates and Restart to update buttons, and lists the release notes for recent versions with links to each GitHub…
- After an update, a one-time What's new dialog now catches you up on recent features. The first one covers SSH remotes, worktrees, pull request status in the composer, and parallel sub-agents.
- On macOS, Help → Export Diagnostics… now opens the diagnostics export directly, so you don't have to find it in Settings
- TTS sentence aggregation follows Settings.language, defaulting to English when unspecified. Language updates apply with TTS settings.
- Added TTSService.pronunciation_transform_ipa(), which builds a text transform from a word-to-IPA mapping so a voice says names and terms it would otherwise guess from spelling.
- Added client-mode reconnect to MOQTransport: when the relay session drops, the transport redials with backoff for as long as MOQParams.connection_timeout allows the client to be missing, keeps its…
- fix(ci-repair): verify a green PR with skipped jobs and keep refusals out of integration failures (e862c72e7) — Yauhen Bichel
- fix(turns): hosted gateway answers carry the whole tool reply, not the 12-line console preview (80822364d) — Yauhen Bichel
- fix(ci-repair): judge skipped checks by their cause and count a green push as success (d47c16542) — Yauhen Bichel
- Merge pull request #145 from palmier-io/fix/agent-model-header
- [fix] Declare the chat model to the Palmier agent proxy
- Merge pull request #144 from palmier-io/fix/playback-tap-stall
- This release is all about making the Windows desktop experience more reliable, more polished, and smoother to operate—especially when you’re offline. 🎉
- Reporting periods are one shared model: pick calendar month-to-date or all available history in menus, Usage & Spend, the CLI, the HTTP cost output, and widgets, with rolling windows and the…
- Portable preferences: export and import UI settings as versioned JSON from Settings or the CLI, keep them in dotfiles, and customize the provider-switcher shortcuts.
- Opt-in Stay Awake keeps the Mac from idle-sleeping while a local agent session is live, and opt-in credential-expiry notifications tell you when a provider needs a fresh login.
Sep 25, 2026
- v0.64.0 → v0.64.3 is all about making OpenHuman feel steadier under real workloads—fewer loops, better recovery, faster turn startup, and clearer billing/usage surfaces. Ship it. 🚀
- Big thanks to everyone who shipped v0.64.0 → v0.64.3: @senamakel, @M3gA-Mind.
- feat(web-inspector): support targeted notifications in What's New (9434edd)
- feat: consume learned skills from multiple containers across SDKs (6252d10)
- feat(web-inspector): track launcher HUD impressions and actions (bbf420b)
- Keeps long and sparse HyperFrames renders local to the requested absolute frame selection. Holds no longer expand once per frame.
- Bounds render working sets without changing requested workers or frame semantics.
- Avoids staging compiler-owned Artifacts that provably do not intersect the render selection. Opaque Browser Programs, fonts and unknown CSS usage remain explicitly conservative.
- add local login for make dev and link the owner at setup (a8efeed)
- agents: add a sales specialist and advice for bootstrapped founders (fd7a8aa)
- briefing: email briefs as a backup and show when one isn't sent (71a75c8)
- Your own keys and ChatGPT everywhere. Models you reach through your own API keys, keys shared with you, project keys, and ChatGPT subscriptions are available in Teams, Slack and the CLI, not only…
- Members bring their own ChatGPT subscription, without needing permission to manage project secrets.
- Saved sessions open on their conversation. Opening an existing session shows skeleton rows shaped like that session while it loads, instead of the full-screen boot screen.
- Match dated gemini-3.8-live ids, reject google_affective_dialog on Gemini 3.1 Flash Live and 3.8 Live at connect, and raise RealtimeError for Gemini Live close codes
- Raise a UserError for a realtime tool_choice that forces a tool call on OpenAI, Azure OpenAI, and xAI
- Add OpenAI GPT-Live support with OpenAILiveModel
- Avatar and pictures on Alibaba Cloud Model Studio answered HTTP 404. Model Studio's OpenAI-compatible host has no images/generations (the public one answers 404 to every model), so "change your…
- The default image model did not match what the page said. The catalogue's first image model (wan2.7-image-pro) was picked while the page recommended qwen-image-3.0-pro.
- Image and video models listed from your key. Settings → Image & video models asks the provider which models this key has — the same /models call the chat models come from — keeps the ones that…
- Fixed follow-up turns failing after background subagents: When using query() with hooks, can_use_tool, or SDK MCP servers, stdin was closed too early if a subagent finished just before the turn's…
- Aligned docstrings with the code and fixed docstring formatting
- Updated bundled Claude CLI to version 2.1.283
- fix(codex): Codex 0.157+ starts in Orca-managed homes instead of failing with SUN_LEN
- Added x-claude-code-prompt-id to the gateway hint headers so LLM gateways can group the requests that serve one user prompt; opt in with CLAUDE_CODE_GATEWAY_HINT_HEADERS=1
- Added availableModelsMatch managed setting: with "exact", an availableModels entry allows only the model version it names, so new releases stay blocked until listed
- Added deniedModels managed setting to block specific models, even when availableModels allows them
- ci(codeql): run nightly scans on main
- feat!(ci): remove daily minor release automatic workflow
- fix(kubernetes): recover compute driver watches
No release notes published.
- feat(cli): add strands update and polish compact, skills, rename, and setup flows (923a5ac2e)
- feat(bidi): support grouped content messages (4342e12b1)
- feat(bidi): add snapshot capture and restore (b8a2e5c00)
- bb desktop 0.44.0
- Add Checkpoints settings for snapshots and automatic session cleanup, including manual cleanup controls and the latest cleanup result.
- Install companion skills when adding MCP servers from the marketplace.
- Publish CycloneDX software bills of materials alongside JetBrains release artifacts.
- compression: describe what CCR compression dropped, not just how much (62d1cc0)
- hooks: add protect_messages, a hard per-message compression veto (d17ac6a)
- opencode: exclude hosts from transport plugin routing (0024b57)
- Skills: Control resource usage and tool persistence dynamically using new ephemeral lifecycles and active skill limits (experimental; enable with ADK_ENABLE_SKILL_LIFECYCLE=1). (43b73e4)
- Database Integrations: Perform high-performance vector and hybrid searches seamlessly within your agent flows with the new MongoDB toolset. (0961ced)
- Evaluation: Optimize agent execution costs and latency with new evaluation metrics for tracking duration, token consumption, and model call counts. (2d428ea)
- Replay sticks to supported actions — clear separation of which actions replay, so runs reauthor on unsupported steps rather than producing confusing mid-run behavior.
- kane-cli whoami is shorter — the Identity panel no longer shows an misleading Expires row.
- Config (OSS): ConfigManager no longer injects OpenAI's default baseURL and model into non-OpenAI LLM providers.
- Vector Stores: Turbopuffer filters now apply every operator (eq, ne, gt, gte, lt, lte, in, nin). Only the range operators were read before, so eq, ne, in, and nin were silently dropped.
- Vector Stores: Turbopuffer search() with euclidean_squared now maps the unbounded squared distance to 1 / (1 + distance) instead of 1 - distance, which went negative and inverted ranking for any…
- 0f7d6842f23fdf14f3c88df9cf842a940983a45d feat: support SAML Single Logout protocol
- 3343aa5bd1fcc99d48c66a0665d96d184e9a79f2 fix: authorize session APIs against sessionPkId instead of id
- b40db125bc328dbbeee1183cea1102d916190c34 fix: block cross-org token minting and secret-field query oracles
- fix: scope multiplexer to displayed session directory
- fix(windows): hide child-process console windows on GUI hosts
- fix(v2): emit host-canonical *-resource rules from the permission bridge
- Claude Code no longer shows a notice after every reply in sessions without a plan. The Stop hook runs check-complete --gate, whose no-plan branch printed "No task_plan.md found" on every turn.
- PowerShell pointer replacement cleans up only what it owns. set-active-plan.ps1 names its own backup file for ReplaceFile.
- Cursor hooks match Cursor's current hook schema. Cursor has no userPromptSubmit event, so the plan was never injected.
- 💻 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。
- [OPIK-8454] [FE] chore: frontend container now runs as uid/gid 1001:1001 (was 101) — update pinned runAsUser
- [OPIK-8523] [GHA] fix: replace deprecated ANTHROPIC_API_KEY secret in CI workflows
- [NA] [SDK] fix: validate placeholders inside chat prompt video_url parts
- Security: the /graphify add ... --watch reference no longer passes the raw, agent-substituted INPUT_PATH placeholder unquoted into a shell command (… -m graphify.watch INPUT_PATH), where a scan…
- Fix: incremental updates (update/extract --code-only/watch) now preserve a cross-file imports/calls/uses edge whose target symbol lives in an unchanged file — the symbol-resolution facts pass…
- Fix: C# type references (inherits/implements/parameter/return/base types) no longer resolve to a same-named non-type node — an enum member, property, field, or method sharing a type's name is…
- test: wire-contract anti-regression system — parity fakes, goldens, unified ledger
- release v0.1.153
- feat: toggle all workspace folders from the Conversations header
- feat: scope OpenAI subscription caches by backend
- feat(home): reframe splash copy around day-to-day engineering work
- feat(experiments): streamline side-by-side comparison
- feat(web): Add formatted json to the dataset item
- feat(skills): basic skill management
- More reliable Session recovery — Recover task state, conversation history, elapsed time, and completed output more consistently after interrupted connections or app restarts.
- A clearer task queue — Reorder queued tasks with drag-and-drop or keyboard controls, choose which task starts next, and explicitly stop the current task before starting another.
- Background processes and better previews — Inspect running processes, open their terminal output or local server previews, and stop them from the Session side panel.
No release notes published.
- feat(desktop): themed macOS title bar merged with the ThinkRail topbar
- chore(deps): upgrade pi 0.84.3 → 0.86.1
- Blog: Adds tracking to the links to the main page + minor changes to the contents
- @mastra/playground-ui TaskList API: title, TaskListHeader, and hideWhenEmpty are removed.
- Added per-endpoint discovery features that observability storage can declare from getFeatures(): entity-type-discovery, entity-name-discovery, service-name-discovery, environment-discovery…
- Added planTraceAggregate(). It checks a parsed aggregateTraces() request and returns a TrustedTraceAggregatePlan that any storage backend can execute without re-validating the request.
- fix(browser): composite background agent tabs, actionable refusals, timed pointer paths
- fix(ci): build the ARM64 app image on a 16 vCPU runner
- fix(ci): build the ARM64 app image on a 16 vCPU runner
- The app now starts on machines that set a system HTTP(S) proxy, such as Clash, v2ray, or a corporate proxy.
- On small or scaled displays, the main window now fits on screen. A 1080p laptop at 150% scaling used to open the window larger than the screen, which pushed the settings and account controls out…
- Plugin slash commands like /goal no longer vanish from the slash menu the first time you use them. Plugin commands now load as soon as a workspace opens instead of on your first /.
- Build: 0.12.5 Commit: 0b7013321b Built at: 2026-09-25 06:44:36 UTC
- Improved reasoning effort level for models with different support, selectable with --effort or from the effort gauge in /effort and /model.
- Changed plugins placed directly in ~/.gemini/config/plugins so that a plugin whose MCP server needs configuration variables now starts disabled until you enable it, matching plugins installed…
- Fixed Mermaid diagrams and LaTeX rendering as rows of garbled placeholder characters in WezTerm and the VS Code integrated terminal.
- Public OpenAI-style lm(messages=[...]) calls, including SDK message objects. Use explicit requests as above.
- Custom BaseLM.forward() and aforward() integrations. Implement the engine interface.
- LegacyEngine, AsyncLegacyEngine, and custom complete_legacy() shortcuts. These are transition tools, not permanent escape hatches.
- Full Changelog
- feat(gateway): reload integrations when the organization's secret changes (c0ff91eb0) — Yauhen Bichel
- fix(git): say what the credential lacks when GitHub refuses a push (074df46ae) — Yauhen Bichel
- fix(git): commit as the OpenSRE Agent where the host has no git identity (d34d86580) — Yauhen Bichel
- feat: support per-call machine resource overrides
- feat: Improve org UI
- feat: add custom headers for model providers
Sep 24, 2026
- Full Changelog
- An audit log you can read. Every entry shows a readable title instead of a
- More outcomes are audited: sandbox-provider transitions, App deployments,
- Agents act as their own principal by default. An agent has its own
- MessageEnqueued — ConversationContext.enqueue() now announces a transient event after appending to an agent's inbox, so a live session can react between model responses. The calling API is unchanged.
- MCP errors follow the spec. Prompts with required arguments are validated before rendering, and protocol-level failures return the prescribed JSON-RPC codes instead of a generic error or a tool…
- A shared event descriptor could retarget under concurrency, leaving a tool awaiting a result that never matched. Seen as intermittent worker-thread hangs on Windows CI.
- Feynman now runs on Node.js 26, the current release. The upper version limit is gone: Feynman needs Node.js 22.22.0 or newer, like Pi itself. Tests run on Node 22, 24, and 26.
- The standalone installers bundle Node.js 24.21.0, the latest LTS release.
- When an alphaXiv call fails, the tool error now points the agent to feynman_science_database_search (arXiv or Semantic Scholar) or fetch_content, so a research run continues on another source…
- Avatar page (Settings → Avatar, or tap the face on any header). Muse's layout: the face on its disc with the agent's name and what it is doing right now, the preview cycling through the moods.
- Bring your own image model. Generation goes through the providers you already configured — no new key: any enabled OpenAI-compatible, OpenRouter or xAI instance, with the catalogue's image models…
- The face moves. The avatar cross-fades between moods (260 ms) and animates them: a slow breath at rest (2.6 s, +1.8 %), a quicker breath and a 2.5 dp bob while working, a ±4° tilt while it waits…
- feat(cli): replace setup Q&A with Quickstart, Customize, and Export, and add /tools (9b7657f13)
- feat(py): add a2a_client vended tool (9746df634)
- fix(a2a): extract content from task.status.message when artifacts are absent (7f2b0c5ea)
- fix: cherry-pick subflow retry and connection-reference fixes into 0.91.1 @abuaboud, @hazemadelkhalel, @Louai-Zokerburg, @majewskibartosz
- fix: cherry-pick subflow retry and connection-reference fixes into 0.91.1 @abuaboud, @hazemadelkhalel, @Louai-Zokerburg, @majewskibartosz
- fix(build): unblock the release — MinIO moved off Docker Hub, and drop the source maps the image build deletes @abuaboud
- Preview remote plugin contents before installation.
- Report mis-encoded config.toml files created by older config writers.
- Show how to cancel MCP browser login and handle Escape.
- artifacts: deliver artifacts by message link or email attachment (212cfaf)
- artifacts: publish, reread and revise artifacts in html, docx, xlsx or links (066abd3)
- knowledge: make chat-proposed playbooks drafts a person approves (b775dea)
- Discover alternate Buzz ACP commands (dd1c2afc2a16238656c5ed19fbfc558605248163)
- chore(mesh): upgrade MeshLLM to 0.76.2 (4544ba1713d8d7f5959d7d63886ff3927357e1ac)
- feat(agents): humanize uncurated Databricks model ids with a label grammar (d01e5f82058463709a22e93bb4cd795da5f53e10)
- feat(ai-gateway): batch inference telemetry per project and retry transient failures
- fix(shared): keep metadata keys that collide with Object.prototype
- fix(batch-export): surface the real read failure, not a generic S3 error
- Added a maxProseWidth setting that caps the width of Claude's prose in wide terminals while tables and code blocks keep the full width
- Added a startup notice, and /status and claude doctor entries, listing telemetry variables in a project's settings files that were ignored or that turned telemetry off
- Added the allowClaudeInChromeWithManagedMcp managed setting to let claude --chrome run alongside an exclusive managed-mcp.json; the error shown when Chrome is blocked now names it
- Ten more providers run as bundled plugins (OpenAI, Fireworks, Perplexity, Qoder, Manus, T3 Chat, DeepInfra, ZenMux, Chutes, ai&), and Atlas Cloud, Vercel AI Gateway, DevPass, and llmman join as…
- CLI config writes no longer delete user plugin settings and secrets, and menu bar layouts show balances for every balance provider.
- Lower background cost: Codex and Claude history caches stop rewriting unchanged files, Cursor backs off geo-blocked requests, and stalled menu catch-up passes no longer loop.
- New Google Drive OAuth connector plus Gmail and shared-drive sync — easier, safer import from Google workspaces.
- Streaming /visualize/json responses in chunks so large graph views load without exhausting memory.
- GLiNER installer overhauled: installs off the event loop, auto-installs CPU Torch on first use, and reports install telemetry and errors.
- Version 0.2.4
- 🖥️ Speed keeps only Rate, the pitch switch moves to Audio, and Reverse goes with its plumbing
- 🖥️ The launch screen is a grid of projects, and the form is a sheet its first card opens
- fix(index): openai-compat and minimax already replay natively
- feat(agents): choose a model when creating a specialist bot
- Record surface-pin provenance; move auto-pinned threads when Works on changes
- Automatic recovery: /continue has been removed. fx now retries and continues on its own.
- Default model: fx now uses Grok 4.7 as its default model. Fast mode remains opt-in.
- Custom connections: fx now supports named OpenAI Chat Completions connections for local servers and other gateways.
- New provider: ai&, an OpenAI-compatible endpoint serving open-weight models from Japan.
- Refreshed the model catalog to 6,386 models across 209 providers. The resolved default model changes for 19 providers that do not pin one, 11 of them to Claude Opus 5.5…
- Raised the minimum js-yaml version to 4.3.2 to pick up a security fix in the parser used to read rule and skill frontmatter.
- [OPIK-8531] [CI] fix: replace unpullable quay.io/minio/mc with the MinIO image's bundled mc
- [issue-8410] [SDK] fix: render the judge reason of Hallucination and SycEval as text
- 三合一会话管理界面:/resume、/agentview、/bg 合并为一套界面,会话浏览、后台任务与状态一览在同一处完成。(
- IDE 选区通道:编辑器选区的实时徽标与提交自动附加,配套 VS Code companion 扩展。(
- 安全模式 dsh-tui safe:只读控制面、救援 profile 与插件清单/修复指引;补充非交互救援提示与两个被剥离环境变量的锁定。(
- Release Notes
- Breaking Changes
- @PrathamS1 made their first contribution in
- [Remote] Device Connections: pair a compatible mobile client to import providers and use desktop Agents over an encrypted LAN connection, with reconnect recovery, execution failures, approvals and…
- [Document] Support self-hosted MinerU 4 document parsing while retaining compatibility with legacy Open MinerU servers.
- [Screenshot] Use system OCR on macOS and Windows to select and copy screenshot text without downloading a local model.
- Code-signed Windows installers. Stable Windows packages are now signed, so the SmartScreen "unrecognized app" warning no longer appears on first launch.
- STRING protein-interaction enrichment. The protein-annotation connector gains a STRING PPI enrichment analysis that scores the functional associations among your gene list.
- Persistent live-source previews. Browser previews of live sources now use a persistent session, so sign-ins and state survive app restarts.
- npm: bash npm install -g base44@latest
- Homebrew: bash brew install base44/tap/base44
- Version: 0.1.20 NPM Tag: latest
- Added structured Claude Code launch arguments for session configuration and plugins
- Reduced background Git polling for repositories the watcher cannot observe
- Reduced Add Project directory search time on large home directories
- Serialized OutputAdapter and ConditionalRouter components containing Jinja custom_filters must now be loaded with Pipeline.load(..., unsafe=True)…
- The tool_result_offloaded meta key that the hook sets on an offloaded message now always holds a list of store references rather than a single reference string, since a result can span several…
- The + operator for combining Toolsets has been removed. Pass Toolsets as a list wherever tools are accepted instead: Agent(..., tools=[toolset_a, toolset_b]).
- Trace ids are unconditional — every Gateway HTTP response carries an
- /mnt/skills is reserved for managed enabled-only skill projections.
- sandbox.replicas is enforced for E2B as a capacity limit; the default
- openai is pinned below 3. Fresh installs from PyPI resolved
- pip install --upgrade everos brings openai back to 2.x. If a fresh
- hermes update (git installs), or re-run the installer one-liner.
- Docker / Hermes Cloud: images build from this tag (nousresearch/hermes-agent:v2026.9.24).
- 9df0cae78a86f83b14d8f28ce4e4e11c0d9c8482 MUL-7429: complete issues when every linked PR merges
- e909e9c89d524cd54c1d5f4fa5963882093efdc9 MUL-7587 fix(daemon): show on-behalf-of identity in agent context
- d5a25d743862d28d6aa6bcc60bc1f3db0a96097a MUL-7608 feat(mobile): add Simplified Chinese localization
- Native window chrome: the dashboard and editor draw their own title bar on Windows, with the window controls aligned to the sidebar and the project search living in the title bar of the projects…
- CLI on Windows: the app installs a diffusion command shim on first launch and rewrites it on every update, so the CLI keeps pointing at the current build.
- Tools on Windows: fonts, screenshot (including from a minimized window), capture, and export all run on Windows.
- tradingagents.dataflows.interface is tradingagents.dataflows.router, and dataflows.symbol_utils is dataflows.symbols.
- Vendor modules live under tradingagents.dataflows.vendors: yahoo…
- tradingagents.agents.utils is gone: the agent tools are in agents.tools, and agent_utils, agent_states, rating and structured are agents.context, agents.state, agents.rating and agents.structured.
- #4013 929403b Thanks @7Sageer! - Roll back some of the overly defensive changes in 2.1.0
- #4015 c7dd841 Thanks @sailist! - Revert filesystem watchers for config and workspace files to on by default. Set [watch] enabled to false or KIMI_CODE_WATCH=0 to keep them off.
- fix(fetch): preserve HTTP header bytes across the transport #672
- fix(fetch): include required headers in Blob URL responses #690
- fix(xhr): normalize script-visible response headers #688
- @mastra/playground-ui: removed @mastra/playground-ui/lib/springs; FluidHoverHighlight API changed to accept only hover and className.
- @mastra/playground-ui: PageLayout/shell APIs reworked (removed PageLayoutRoot, MainContentLayout, MainContentContent, and multiple AppShell header-related props/contexts).
- Trace grouping deprecations (Core/Server/Client): grouped results in queryTraces() / trace-query group are deprecated; migrate to thread query (queryTraceThreads() / queryThreads contract).
- Added medium verbosity mode to /config, between high and low, which groups related tool calls and thoughts into concise summaries (such as Explored N files) while keeping commands and responses…
- Improved Mermaid diagram and LaTeX rendering in the artifact viewer on Kitty-compatible terminals: artifacts are pre-rendered in the background so diagrams appear as soon as the viewer opens…
- Improved the artifact viewer footer by combining the separate top and bottom hints into one g/G top/bottom entry, labeling m with the view it switches to (diagram, LaTeX, ASCII, or raw), and…
- New whole-object local cache: immutable data objects on S3 are kept on the local node, so repeated reads after a restore — and re-importing the same snapshot — hit the local cache instead of going…
- Reduced read amplification during restore: scattered small reads are coalesced into whole-object GETs, and the concurrent fetch budget is capped.
- With the default MinIO setup, cross-node restore / create-from-snapshot reaches running in under 1 second.
- fix: break logger EPIPE feedback loop — async stderr error re-enters uncaughtException
- fix: lane-aware instance discovery — attach across all live registry entries
- fix: identity-based launcher proxy reuse — code fingerprint + lane
- 💻 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。
- commit: e16d7d962aa05f4bcfc82205948eb6e864bcdcb1
- source tag: beta2026-09-24_12-10-16
- regions: cn global
- 🤖 Devin runs as a native harness. omni devin now mirrors the real Devin TUI into the web UI, similar to the Claude Code and Codex native harnesses! (#7595, #7501, @dhruv0811)
- 🗂️ Curated model catalogs everywhere. Pi, custom ACP agents, and managed Kubernetes / Agent Sandbox sessions can show configured model shortlists — with per-harness gateways and optional live…
- 📎 A richer composer. Preview attached images before sending, attach more types of files to Claude Code and Codex sessions, and manage queued messages and active sub-agent work directly from the…
- fix(sidebar): stop a workspace with a structured chat reading as asleep
- Fix stale listings when search query changes
- Virtualize artifacts list with reusable component
- Add GitHubCopilotOAuthFlow for device authorization
- Add BoolCriteria to say what a bool field's yes and no mean, and let a True/False Enum do the same
- Use what the user wrote about None as the "none of these" option in TypeSafeModel
- feat(hosted-gateway): let a queued prompt wait for a turn slot (9a83cf6dc) — Yauhen Bichel
- fix(hosted-gateway): say what to fix before relaying the gateway's question (91666526c) — Yauhen Bichel
- feat(skills): let demo C delegate to the hosted gateway, label unchanged (2717f1dc8) — Yauhen Bichel
- Added daemon incident notices to the agents view: recent worker crashes, command-timeout bursts, and update restarts surface as a dismissible status line pointing at prime-agent incident.
- The Anthropic subscription-auth warning now names the risk: subscription requests identify as Claude Code, which may violate Anthropic's terms and can get the account restricted or banned.
- Branch summaries now run through the auxiliaryModel setting, not just refinement and compaction passes.
Sep 23, 2026
- Changelog for v0.60.0-preview.0
- chore(release): bump version to 0.61.0-nightly.20260908.gc647533d6
- Changelog for v0.59.0
- Research runs are now visible in telemetry, without content. The research extension inside Pi sends feynman_session_started, feynman_workflow_started and feynman_workflow_completed…
- First-run notice. The first interactive run prints a one-time notice on stderr (never in scripts or CI) saying telemetry is on, what it collects, and how to turn it off with FEYNMAN_TELEMETRY=off.
- Telemetry stays on by default and follows the existing opt-outs: FEYNMAN_TELEMETRY=off or DO_NOT_TRACK=1 disables the CLI and the Pi runtime events alike.
- ui: add an assistant-led workflow wizard and compact the jobs catalog (c52b96c)
- ui: validate check-in cadences and clear emptied leave dates (5940557)
- Release notes — formatted for people, with expandable sections.
- Changelog — plain Markdown for AI agents and tools.
- npm package
- fix(codex): keep custom MCP servers apart from same-named entries in config.toml
- fix(composer): keep keyboard focus in the draft after adding an image
- fix(chat-mcp): start custom MCP servers with the widened PATH
- feat(mcp-ts)!: swap to @modelcontextprotocol/client 2.0 (ab8ff46f1)
- feat(bidi): require explicit model IDs for providers (8732c7c90)
- feat(agent): add Agent.shutdown() for scope-based cleanup (fec042766)
- The planning-with-files npm package description now discloses selected project planning context and confirms there is no network upload path.
- The full npm README links to v3.20.7 repository files. The contributor grid uses @Fat-Jan's current profile and stable identicons linked to archived credits for two unavailable accounts.
- @kuei51307-hub for the phase-status locking fix from PR #282 carried forward from v3.20.6.
- Updated bundled Claude CLI to version 2.1.281
- Pinned default model for e2e tests to claude-opus-5 to work around CI failures with the CLI's new default model
- Add session notes, archived history, and context-window resets, with /notes and /compact
- Show queued messages under the working indicator until Dexter picks them up
- Default new sessions to GPT 6 Astra and upgrade saved older model IDs
- Fix: the PYTHONHASHSEED determinism pin (0.9.66) now re-execs via python -m graphify instead of replaying argv[0], fixing a Windows regression where update/extract/cluster-only/label failed to…
- Feature: PHP closures are now extracted — an anonymous function(){…}, an arrow fn()=>…, or a closure passed as an argument now produces a node and its inner calls are captured, in all positions…
- Fix: absolute Python package imports (import pkg.sub, from pkg.sub import x) now resolve to the local package/module node within the scan root, reusing the canonical resolver…
- Added Claude apps gateway support for newer Claude Desktop keys in desktop policy blocks, including blockReadsOutsideWorkingDirectories and disableBypassPermissionsMode
- Added assume_role on Claude apps gateway Bedrock upstreams: the gateway calls Bedrock as an IAM role it assumes through STS, in another AWS account if needed, optionally one session per developer
- Added guardrail: {id, version} on Claude apps gateway Bedrock upstreams to apply an Amazon Bedrock guardrail to every request sent through them (set it on all Bedrock upstreams or none)
- Search: The search_memory tool description no longer asks the agent to search proactively before answering anything that may depend on earlier context.
- Automatic recall: Injected memories are introduced as "Mem0 found these relevant memories from earlier work:".
- Non-TTY output is now a versioned NDJSON stream — every run and testmd execution emits structured events with a version stamp at the front, so tooling can depend on the shape of what it reads.
- Events are also written to events.ndjson — the stream is teed to a file alongside the terminal, so you get a durable record even when the terminal output is dropped.
- A live run is findable from disk — an active-run pointer lets a separate process or file reader attach to a run that's already in progress.
- Redesigned file browser navigation — configurable path bar with breadcrumb and text modes, folder autocomplete with keyboard navigation, and fully clickable file rows without a loading overlay
- Consolidated create-new dropdown replaces separate New File and New Folder buttons
- Inline rename and new-folder creation in the shared footer input row, with modal rename preserved for external callers
- feat(mcp): support OAuth on workflow MCP servers
- fix(copilot): clarify knowledge connector authentication
- @waleedlatif1
- 5b456e10a1c9f5a41a9ecdb58c1ab0445e866c88 fix: make account item names selectable in organization edit page
- 9db04afb31c44d246f6d8820bc6fc8a10f63bd09 fix: restore web-old behavior in the console's editable sub-tables
- 2ad0722d9a4c696f0becaeaf333ab5ff2605d0a1 fix: send Vary: Origin from the CORS filter
- feat(desktop): universal macOS DMG with per-arch bundled runtimes
- feat: Support Light+ and Solarized Light theme
- feat: tag Agent Canvas telemetry by deployment kind (OSS-13530)
- [BREAKING] Custom source that composes agent instructions must move to the new module.
- [BREAKING] Forks that merge upstream by hand must install their credential gateway before restarting.
- Claude groups with no model set move to Opus 5.5, which changes cost and usage-limit burn. API-key installs move from Opus 5, and Pro and Team Standard installs from Sonnet.
- fix: guard the CLI entry against host-process imports (opencode plugin crash)
- fix: watchdog tracks the real claude host, not the transient sh hook wrapper
- fix: shared stable-port proxy outlives its first owner (watcher set + grace window + gated route)
- Fold and unfold large pasted text blocks directly from the prompt input and transcript.
- Add a Marketplace page to JetBrains settings for browsing, installing, and removing marketplace agents, MCP servers, and skills without leaving the IDE.
- Install Marketplace plugins from npm packages or from a git repository (for example git:github.com/owner/repo@v1.2.3#subdir).
- Lazy MCP caches now warm without blocking the event loop.
- Token and cost breakdowns now refresh while the view is open.
- c526854 release(deepagents-code): 0.1.75
- fix(flows): stop template imports from failing when the v23 upgrade has no stored flow- #15736 #15736 > 📌 Upgrading a self-hosted instance? Review the Breaking Changes page for any actions…
- Added SuperGrok sign-in so SuperGrok subscribers can use Grok models in the Agent Panel. (#63248; thanks thibaudgg)
- Added the agent.prevent_idle_sleep setting, enabled by default, to prevent idle system sleep while agent threads are running. (#53130; thanks cppcoffee)
- Added support for DeepSeek Flash 4.1. (#64014; thanks cppcoffee)
- add gpt-6-sol, gpt-6-luna, and claude-opus-5-5 to the cost tables and the playground (59d09f4)
- js: add dataset split write helpers (aab229f)
- js: add secrets management helper (c282563)
- Teams 主持人改写派工,成员回复同步 IM
- 企查查、OpenAlex 连接器
- 专家编辑、聊天欢迎标题与专家切换预填
- [OPIK-8468] [QA] test: cover the failed playground run error state (from
- [OPIK-7791] [QA] Proposed e2e specs from the 2.2.71 → 2.2.72 release exploration: dataset-item field mappings
- [OPIK-7791] [QA] Proposed e2e specs from the 2.2.73 → 2.2.74 release exploration
- Rust CLI: support for configuring Vibe Code remote projects
- Clean up managed worktrees after archiving local sessions
- Archive local sessions so they can be hidden from active session lists
- Live voice conversations in the desktop app
- Decisions provider crate with OpenRouter and Jev implementations
- Z.AI Coding Plan provider with streaming tool calls
- #893 53fba6b Thanks @tuandinh0801! - feat: add Oh My Pi (omp) coding-agent integration
- #903 4e84ff0 Thanks @eugeneliu-86! - feat: run repository page workers in parallel
- #905 0c0b35d Thanks @colifran! - feat: implement openwiki search, read, and link to support queryable wikis over mcp and wiki linking for multi-wiki reads
- fix(scim): scope Users/{id} reads and mutations to the caller's org (v3 backport of
- Knowledge Retrieval Pipeline: Knowledge.search() now runs a knowledge level retrieval pipeline that can widen the candidate pool before a reranker reorders it, so rerankers no longer need to be…
- MMR Reranker: Added MMRReranker (Maximal Marginal Relevance) as a retrieval strategy that balances relevance against diversity so near-duplicate chunks do not crowd out the result set.
- Page Source Relocation: Added Knowledge.inspect_page_source / ainspect_page_source and guarded migrate_page_source / amigrate_page_source to move an indexed documentation source to a new hostname.
- Agents act as their own principal. An agent session now acts as the agent, on behalf of the person who started it, and every audit event records both.
- Microsoft Teams is complete. Agent questions, images, and the remaining Teams flows now work like Slack.
- Saved history works while the sandbox is off. Session transcripts and their attachments load on web, SDK, and CLI without waking the sandbox.
- #3990 32000d0 Thanks @Grapedge! - In fullscreen, click a folded block to open or close it.
- #4004 e17ff67 Thanks @Grapedge! - Add a TUI mode setting to /settings with an experimental fullscreen layout; switching modes takes effect after restarting Kimi Code.
- #3964 6451f1e Thanks @7Sageer! - Block file tools from accessing files outside the working directory through symlinks.
- npm: bash npm install -g base44@latest
- Homebrew: bash brew install base44/tap/base44
- Version: 0.1.18 NPM Tag: latest
- PyPI: bash pip install crawl4ai==0.9.4
- Docker: bash docker pull unclecode/crawl4ai:0.9.4 docker pull unclecode/crawl4ai:latest
- Note: Docker images are being built and will be available shortly. Check the Docker Release workflow for build status.
- Claude Opus 5.5 at medium effort is now the default Anthropic model, and anthropic-api: model specs use the native Anthropic runtime
- The browser tool detects your default browser and supports Chrome, Edge, Brave, Chromium, and Safari in addition to Firefox
- Reset exhausted OpenAI usage with banked resets through a confirmed command, with a hint shown when your account has one available
- af62622e50e5d0cb5d3dfc8d0701311e14c21dca MUL-7288: fix(server): wake guest squad leaders on worker replies
- b7cdd76ad8129807608fa69bf395ed9baa0c1412 MUL-7349: feat(server): add the duplicate-mark column and keep it correct (1/2)
- 1879bc35e3441e602132ee37e9463043d6bad986 MUL-7351: fix(daemon): fence and retry task start acknowledgements (LANX-3)
- fix(oauth): return sign-in through the hosted callback page in hypit auth login now works when the CLI runs on a remote host, in a container, or under a remote agent.
- fix: ignore album artwork in media probes in An MP3 with cover art is read as audio, so cutting it to WAV works. Fixes #346.
- fix: include bundled packages in vocabulary listings in hypit vocabulary lists the packages bundled with the Distribution, including in an empty project. Fixes #344.
- Godot 4.2+
- Active Ziva account
- Agent Controller stream message events changed: one full message_start, then ID-addressed message_update deltas; message_end now contains only the message ID (store by ID and apply updates).
- @mastra/mcp@2.0.0 is a major rewrite: MCP 2026-07-28 only (legacy handshake removed), suspend/resume replaces elicitation APIs, and several server/client transport and protocol surfaces were…
- Make the workspace grep text-extension whitelist extensible. Added .sas, .log, and .jsonl to the built-in text extensions and MIME type map so they are searchable by default.
- fix(geometry): unify cross-frame coordinates and screenshot conversion | 统一跨 frame 坐标与截图转换
- refactor(observation): decouple DOM processing from capture orchestration | 解耦 DOM 处理与采集流程
- refactor(observation): unify cross-frame capture coordination and data ownership | 统一跨 frame 采集协调与数据归属
- fix(skill): replace broken skill-tools filtering with glob prefix matching
- fix: remove skill search reminders and related infrastructure
- fix(gpt): align Codex prompt and shell tool exposure
- Three new providers: track Charm Hyper credits, GitKraken AI weekly credits, and Bifrost gateway budgets and rate limits, bringing the catalog to 80 providers.
- More flexible accounts: save labeled Kimi web sessions and Doubao Ark API keys, and use per-account OpenCode Go API keys without changing cookie preferences.
- Zed spending: opt in to browser billing for token spend and remaining budget while keeping the editor login as the default.
- Smart Literature collections. A collection can now screen its references against a description with explicit inclusion and exclusion criteria, sorting them into Included, Needs review, Excluded…
- Research packages embed RO-Crate metadata. Newly exported .science packages describe their final snapshot with RO-Crate 1.1 metadata, reusing selected immutable payloads and rebuilding references…
- Three new connectors. Zenodo public record discovery searches records and retrieves metadata and file inventories without authentication.
- feat(agent-launch): deliver a launch prompt to a terminal agent
- feat(session-search): sort search results by newest, and break relevance ties by recency
- Virtualize artifacts list with reusable component
- MemoryOS 2.0.34 release.
- <!-- doc-agent-local-plugin-release-intent…
- Report full-output scan failures as errors
- Review original observation records through a host
- Add GitHub Actions workflow for Bedrock scans
- commit: 1375c3cd96a23cf11306760821e63bf92ca29939
- source tag: beta2026-09-23_11-35-08
- regions: cn global
- Added @<subagent> <message> prompt syntax to send a message directly to a subagent conversation, with autocomplete listing running and completed subagents
- Added Vim numeric count multipliers so counts apply to operators, motions, and actions in Normal and Visual modes, including 3dw, 2d3w, 3dd, 3x, 3rX, 3p, 3u, [count]G/gg/$, and counted text…
- Improved GEMINI_API_KEY sessions to reduce behavior discrepancies with the non-API-key sign-in path
- Support genai-prices 0.1.7 and new OpenAI audio models
- Add OpenAI gpt-6-sol and gpt-6-luna model support
- Add Claude Opus 5.5 (claude-opus-5-5) support
- test(desktop): stabilize unread and audio release smoke fixtures (fe56d311a7156802408afb04023c3f75d4015007)
- fix(desktop): remember Inbox unread-only choice (312cf674fad5563c24a66ec5350d306d192ed64a)
- fix(desktop): refresh channels after access-revoked closure (a024b83d5448598928d512364fea099ca96b373c)
No release notes published.
- Build: 0.12.4 Commit: 5cfeff4780 Built at: 2026-09-23 01:49:56 UTC
Sep 22, 2026
- Use Claude Opus 5.5 directly through Anthropic, with explicit API-key and Claude-login routes in the model picker
- API-key and OAuth model catalogs refresh independently, so one account or authentication method cannot hide models available on the other
- Update Claude OAuth client compatibility to Claude Code 2.1.280 for the latest model release
- Bump version to 1.0.54 (2352506b5)
- Preserve onboarding example attribution (#SKY-16818) (0e6274ff8)
- Stop syncing dev_scripts to the open source repo and inline the invisible hCaptcha test fixture (b6749ea20)
- fix(cli): align @strands-agents/sdk range with harness (fixes interventions instanceof crash) (d13c21486)
- fix(cli): improve onboarding and terminal responsiveness (2db407431)
- fix(bidi): remove retired Nova Sonic 1 from integration tests (d29c36e1f)
- feat(jev): add TypeSafe evaluation models to Agent
- feat(models): add GPT-6 Sol and Luna
- feat(providers): add Claude Opus 5.5
- feat(onboarding): copy a hosted prompt link instead of the CLI command (refs PE-217) (6f8d0ae)
- fix(inspector): keep conversation results and failures accurate (9b3ccbb)
- Feature: five new language extractors — COBOL (.cbl/.cob/.cobol/.cpy; programs, paragraphs, PERFORM/CALL/COPY, pure-regex, no new dependency)
- Feature: Go interface method requirements now resolve to the interface, and their parameter/return types emit references edges
- Feature: a Rust self.method() call resolves across files for simple generic impls (impl<T> Foo<T>)
- feat(assistant): slash command menu and 100-step budget
- feat(agui): let callers opt out of graph state on the stream
- fix(frontend): keep assistant reopenable mid-run
- feat: add configurable event webhooks
- fix: use NULL for cleared MCP initialization errors
- refactor: store internal config IDs and expose saved compiled configs
- Composio connectors now load all their tools, not just the first 20. Google Calendar, for example, showed only 20 of its 47 tools, and the Installed view wrongly said "20/20." The full list is now…
- You can now connect to a Mac as an SSH remote from a Mac. Picking a Mac host used to fail with "Remote target darwin/arm64 is unsupported in SSH" even though Test passed on the same profile.
- Session errors now stay in the transcript when you leave a session or open it in another client.
- chore(desktop): bump version to 1.30.1
- feat(marketing): add App Store icon to footer
- docs(readme): add Superset for iPhone download and requirements
- Latest frontier models — Use Claude Opus 5.5, GPT-6 Sol, and GPT-6 Luna through supported providers, including GitHub Copilot. See Choose a Model.
- Grok 4.7 by default for xAI — New xAI sessions now default to Grok 4.7. See Provider Authentication.
- Added inherited Claude Opus 5.5, GPT-6 Sol, and GPT-6 Luna support for GitHub Copilot.
- bb desktop 0.43.4
- Update the model picker to GPT 6 Astra, Sol, and Luna, with Astra as the OpenAI default
- Update Claude, Gemini, Grok, Kimi, and DeepSeek to Sonnet 5, Opus 5.5, Fable 5.1, Gemini 3.8 Flash, Grok 4.7, Kimi K3, and DeepSeek V4.1 Flash
- [x] LLMman - OSS is now available as a local LLM provider, replacing Docker Model Runner
- [x] Google Vertex AI is now available as an LLM provider
- [x] New generate-image agent skill for creating images from chat
- Added Opus 5.5 to the Claude catalog as its default model, with a 1M context window and Fast Mode, on Claude Code 2.1.280 and newer
- Fixed the sidebar keeping only recently changed conversations after the app reconnects to a daemon
- Fixed Import session offering only the newest 100 Codex conversations
- [OPIK-8274] [BE] refactor: route experiment item aggregates through JsonEachRowBulkInsert
- [NA] [SDK] feat: sign in to Claude Code after opik mcp configure
- [OPIK-7381] [BE] fix: keep placeholder-project aggregates in the experiment groups query
- 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
- Added Claude Opus 5.5 (claude-opus-5-5), now the default Opus model — 1M context, $4/$20 per Mtok with $0.20/Mtok cache reads
- Added mouse support to more lists in fullscreen mode: the wheel scrolls the /skills list, and a skill's state options in /plugin can be clicked
- Added CLAUDE_CODE_MAX_MCP_DESCRIPTION_LENGTH to change the 2,048-character cap on MCP tool descriptions and server instructions for every MCP server in the session
- Let owners and admins open every session. An owner can turn on an account
- See who owns each session. The Sessions page shows each session's owner and
- Sending from the project home sends the message instead of freezing the input,
- feat: test remote MCP servers on cloud backends via the app server
- feat: open Git Sync to org admins on cloud backends
- feat(i18n): translate recently added Turkish UI strings
- fix(websearch): use session model for Xiaomi sidecar requests
- fix(session): avoid _- in session IDs
- test: drop the resetDatabase fixture
- Search: the home screen searches the names, folders and full text of your Word, Excel, PowerPoint, PDF, Markdown and HTML files from a local index, CJK included.
- Slides: charts closer to PowerPoint — pie of pie, logarithmic and date axes, per-point data labels, axis titles, legend order, chart fonts; 3D bevel lighting.
- Docs: large documents open and type faster; EMF/EMF+ pictures render with correct text placement, pen widths and pattern brushes; the cursor lands in the document when it opens.
- Updated dependencies [aab229f]
- @arizeai/phoenix-client@7.14.0
- fix: runtime respawn for native attach mode when the shared proxy dies
- fix: carry the image term in the anonymous prefix-affinity nudge bound
- test: pin the overflowArmTokens RELEASE direction + restart round-trip
- feat(slack): download private files into transit storage
- fix(gmail): classify quota failures as rate limited
- feat(gmail): stream attachment downloads to transit files
- Rate-limit banners: unified rate-limit data now feeds subscription banners; usage beyond the limit is silently rolled into additional usage with clear notification.
- Official plugin marketplace mirror: the plugin marketplace is now served from an official GCS mirror, making plugin installs more stable and reliable.
- Smoother long-session interactions: estimated scroll offsets and zero render-time filesystem syscalls reduce a cold first frame from ~1.6s (1000 lines) to ~160ms.
- fix(session): preserve findings across file renames
- fix(config): preserve unknown JSON fields during config updates
- fix(review): keep git stderr out of --commit background
- Fix stale listings when search query changes
- More menu-bar control: pin Cursor's Grok Bot usage and recover icons stranded beyond an attached display.
- Kimi regions: choose China or International for usage, browser sessions, and dashboard links.
- More reliable quotas: preserve OpenRouter balances under load and show each Antigravity quota pool once on Linux.
- Allow /teleport with Vibe and workspace API keys, including free accounts, without an internal feature gate.
- Explain which API keys to use when teleport rejects a Codestral key.
- Warn when managed configuration contains removed Vibe Code settings.
- npm: bash npm install -g base44@latest
- Homebrew: bash brew install base44/tap/base44
- Version: 0.1.16 NPM Tag: latest
- Adopt OpenCode v1.18.14 through v1.18.15 improvements, including message ordering fixes, compaction serialization, locale coverage, and TUI enhancements.
- Adopt OpenCode v1.18.16 through v1.18.18 improvements, including conversation-aware compaction, expanded reasoning effort support, provider compatibility fixes, and retry handling.
- #14360 675ed4b - Retry transient connection drops when approving permissions so auto-approve and manual approvals no longer leave agents waiting.
- fix(auth): evict the API key cache after deleting the row (v3 backport of
- Added cache bust timer to the footer.
- Added a durable Debug Console cost breakdown.
- Added support for copying the highlighted thread ID with Ctrl+C.
- Improved context compaction to spread its user-request budget across all captured prompts, so a long initial instruction is no longer truncated to a small fixed slice when the other prompts in the…
- Improved context compaction to size summary and truncation budgets from the model's full context window instead of the compaction trigger threshold, so compacted summaries and background-task…
- Fixed PDF and audio tool outputs and attachments failing with unsupported mime type errors on custom models configured in settings.json.
- Name a None output route None rather than NoneType
- Raise when UserPromptPart.content is not a str or a sequence, instead of silently sending a dict's keys
- Ask three more field types TypeSafeModel can already answer
- 引用之前的对话:输入 @ 可以搜索并引用历史会话,例如「参考 @登录方案 的结论,补充退出登录流程」。引用不会把整段历史复制进上下文,Claude 按需读取被引用会话的内容;点击消息里的引用可直接打开那段对话,原会话不会被唤醒。
- 跨会话聊天与控制:可以让 Claude 新建独立会话并行推进任务,并在会话之间收发消息、等待状态变化、汇报结果(ListSessions / ReadSession / CreateSession / SendSessionMessage / WaitSessions)。
- 新会话会出现在会话列表里,可以单独打开继续聊;其他会话发来的内容以带来源标注的普通消息显示,工具卡片上可以一键打开相关会话。
- feat(analytics): record shell render at the login screen (430387420) — Vincent Hus
- Fix weekday cron schedules running Tuesday–Saturday (0e8d6afba) — Devesh
- Verify commit events through real ingestion and ClickHouse (e17b6dda5) — Vincent Hus
- fix(query): a > combinator no longer re-applies the parent's predicates to the child
- @argszero made their first contribution in
Sep 21, 2026
- beacon: schema v2 — routing/compression training signal, gzip transport (67eb910)
- cache: add the headroom-cache-ttl offline TTL estimator the docs reference (52c7aa2)
- dashboard: Cost Saved headline card; label per-row savings as message-only (17b0412)
No release notes published.
- security: harden unauthenticated settings writes and telemetry error scrubbing
- feat(defaults): opus 5 and high thinking by default (c3fc5ff81)
- feat(site): launch the redesigned Strands website (1fd2ad684)
- fix(cli): depend on the published harness in exported projects (8ecf1ebc1)
- Your assistant can now create documents straight from the conversation or the Library, Telegram plays nicely with groups, and computer-use screenshots stick around in chat.
- Create documents without leaving the chat. Your assistant can now mint brand-new documents, and you can start one from the conversation menu or straight from the Library.
- TinyFish search and fetch are built in. Two new providers bring TinyFish search and page fetching to your assistant's toolkit, one more way to reach the live web.
- Fixed Bedrock image attachments so they are only hoisted for Claude, Nova, and Llama 4 models.
- Fixed Together AI streaming usage reporting.
- docs: add DeepSeek V4.1 Flash to Zen
- Canonical omnarad 0.1.26 binaries built from ae7d22d5dfa3bb5158f58e36e3d80d54da6c0295. macOS binaries require macOS 13 or later.
- <!-- Release notes generated using configuration in .github/release.yml at omnarad-v0.1.26 -->
- Full Changelog
- The chat composer's send button no longer starts out disabled on desktop. The task page disables the composer while it checks whether the task is paused, and in React Query v5 that pending state…
- Duplicate document inserts wrapped by the ORM are classified as retryable conflicts again, instead of surfacing as opaque database errors; unrelated database failures keep their original identity.
- No migrations, no configuration changes, no API changes. Upgrade in place.
- fix(mcp): decode resource template URI variables
- feat(mcp): let a served agent own its HTTP transport settings
- feat(tealtiger): add rate_limit governance policy
- Three new providers: track Helmcode quotas and prepaid balances, v0 billing and rate limits, and TypeSafe spending and credits.
- Two providers in one icon: stack your chosen providers in a single menu-bar icon, with automatic or explicit row choices.
- Clearer Usage & Spend: browse sources and models by provider, use simpler chart controls, and follow Antigravity's local cost estimates.
- fix: cherry-pick subflow retry and connection-reference fixes into 0.91.1 @abuaboud, @hazemadelkhalel, @Louai-Zokerburg, @majewskibartosz
- fix: cherry-pick subflow retry and connection-reference fixes into 0.91.1 @abuaboud, @hazemadelkhalel, @Louai-Zokerburg, @majewskibartosz
- fix(build): unblock the release — MinIO moved off Docker Hub, and drop the source maps the image build deletes @abuaboud
- OpenCode session replay tolerates null or array part rows, non-string tool names, and non-string text instead of aborting the session.
- Shell and PowerShell initialization now report failed attestation as NOT attested, with a reason and recovery command, instead of claiming success.
- The analytics task-plan template includes a non-empty ## Next Step section across its shipped copies.
- Canonical session context and extension boundaries — Edit model context without rewriting history and add actionable lifecycle hooks. See ContextEditEntry and extension events.
- Full-transcript context extensions — Use context_with_system for per-request system-message transformations. See context_with_system.
- Per-model image input limits — Configure cache-safe image resizing per model for attachments, read, and tool-result images. See Image Input Limits.
- hermes update (git installs), or re-run the installer one-liner.
- Docker / Hermes Cloud: images build from this tag (nousresearch/hermes-agent:v2026.9.21).
- feat: encode v4 native blocks and event bytes in Rust
- feat(automations): filter prompt events by tags
- refactor(normalized-io): enhance additional input handling
- Edit a person's account role from a group's member list. You no longer
- A new API key or ChatGPT account is private by default. The Add dialog
- One place to pick provider keys for a session, and the choice saves.
- Install with
- npm install -g codeburn codeburn menubar
- That command drops the app into ~/Applications (replacing the copy you already have), records the persistent codeburn CLI path used by the menubar, verifies the downloaded checksum, and launches it.
- Slides: keyboard shortcuts, mouse modifiers, context menus and insert sizes now match PowerPoint; new Edit Points, Section/Summary Zoom, Save as Picture, media animations in the show.
- Docs: long documents open in phases and stay editable while loading, typing and page switching several times faster; separate East Asian and Latin fonts (@GoodScholar).
- Sheets: status bar statistics on streamed workbooks; Format Cells dialog fits long labels.
- feat(status): acknowledge a pull request the moment automatic commands start
- feat(gitlab): content-based auto-resolve for fixed inline threads
- feat(tickets): make description issue matching configurable
- fix: resolve a ton of security alerts
- fix: resolve a ton of security alerts
- fix: resolve a ton of security alerts
- 56abdfa7f1fa2cbe1e073f553bc10b06598046f9: themes: prefix diff theme keys with diff_ (@meowgorithm)
- a88b5735ff056b52eee8d30121b85c8a8ca79601: themes: small adjustments to the key names in the theme API (@meowgorithm)
- Add aos console on Unix: a keyboard/mouse Command Center for permission
- Add a developer-preview macOS AOS menu-bar app with permission dialogs,
- Package Darwin share/AOS Command Center.app from AOS_COMMAND_CENTER_APP
- docs(skill): sync the docs skill's openapi.json to v0.10.0
- feat(api): attachment filenames for store-owned banks
- fix(attachments): never reclaim attachments for a store-owned bank
- fix(dropbox): escape non-ASCII in Dropbox-API-Arg so a CJK path can be downloaded
- fix(slack): recognize rotating access tokens
- release(langgraph): 1.2.12
- chore(deps): bump soupsieve from 2.8.4 to 2.9 in /libs/langgraph
- feat(langgraph): add response_schema to interrupt()
- Dedicated kernel assets release (BM + PVM guest + PVM host).
- Memory Recovery and Capture Boundaries: Fixed dirty-data recovery, orphan trace writes, and LLM empty-response retry stability issues.
- Memory Evolution and Hermes Installation Improvement: Improved the reliability of memory evolution and Hermes installation, enhancing overall performance.
- Reload skills when skills_metadata is reset to None.
- Reject parallel edits to the same file to prevent conflicting changes.
- 0114a3f release(deepagents): 0.7.16
- Doctor plugin registry: preserve the complete bundled plugin inventory when Doctor creates or repairs registry state, so default Browser, Canvas, pairing, file-transfer, phone-control, Talk voice…
- Security and credential safety: harden command parsing, browser origin checks, plugin Git installs, diagnostics, service credentials, and webhook logging across Gateway and official plugins.
- Message and session integrity: preserve queued, imported, streamed, and tool-result messages across retries, hooks, recovery, and channel lifecycle transitions.
- commit: 5b10e9babcdee311d6bf0d258a1475170b9920a1
- source tag: beta2026-09-21_19-34-08
- regions: cn global
- feat(rules): add F# review support
- fix(config): exclude dependency and build-output directories by default
- fix(viewer): keep vertical mouse-wheel working over scroll regions
- #14054 3b6746c Thanks @vkeerthivikram! - Reintroduce automatic session cleanup in Settings > Checkpoints.
- #14301 f03ebe3 - Create new Git projects and clone repositories directly from Agent Manager. Choose where projects are stored and attach them without changing the VS Code workspace.
- #14282 d0475a6 - Label prompts that Kilo sends on your behalf, such as worktree updates from base, expanded slash commands, and editor or terminal code actions, with a "Sent by Kilo" header.
- 2bd4f1757106042b625b4789b60cd5a5528067db MUL-7058 feat(cli): support self-hosted Gitea release sources
- 41db4282951c24fb9bf7c11bb0ca371c362a46f0 MUL-7324 feat(wecom): answer in the message the question opened
- 2561b492c35aae6e4567c25836bab049400d08d4 MUL-7324 refactor(wecom): trim the stream bubble test suite and its test-only hooks
- Windows: .exe installer or portable. macOS: .dmg (Apple Silicon aarch64 or Intel x86_64). Linux: .AppImage (universal) or .deb (Debian/Ubuntu).
- Debian 12+ ships without FUSE 2 by default, which AppImage requires. If ./omniget.AppImage fails with a libfuse error, run
- bash sudo apt install libfuse2
- Persistent PDF annotations and document notebooks. Notes and annotations now stay with the file version they belong to: text styles, area marks, page and document notes, comments, colors, global…
- Complete RO-Crate artifact export. A verified artifact version can now be packaged as a complete RO-Crate 1.1 archive together with its exact inputs — declared sizes and checksums are verified…
- NCBI BLAST sequence search. New asynchronous tools submit a nucleotide or protein query to NCBI BLAST, track the job until it completes, and retrieve the report in the format of your choice —…
- [NA] [SDK] [DOCS] Update automatically OpenAPI spec and Fern code
- [NA] [BE] Update model prices file
- [Model] Add oMLX as a built-in local provider, with its base URL preset and no API key required.
- [Model] Add Moonshot AI global endpoint for users of the international Moonshot console.
- [Settings] Renamed Recycle Bin to Archive and promoted it to a standalone settings page with improved layout and navigation.
- Add an optional Review SKILL.md button beside the skill install/export actions. Open a read-only modal to inspect the complete file, including frontmatter and substituted values, before placing it.
- Keep the existing direct Add to Scout and export flow unchanged for users who do not want to review the file.
- Install or export the exact previewed text without another model turn. Closing and reopening the modal reuses the prepared file; editing the plan invalidates it.
- fix(endpoint-kit): share the poll error decision across asynchronous Providers in Progress polling now keeps a job pending on HTTP 429 as well as 5xx and transport errors, waiting the Retry-After…
- split-output sizes that round down to zero bytes (such as 0.5b) are now rejected instead of looping forever.
- Output and skill paths are classified by real directory boundaries, so a sibling directory with a matching prefix is no longer treated as inside the base.
- Remote skill names are derived from the owner and repository of GitHub URLs.
- First unified umbrella release: every artifact ships at 1.1.0 from tag release-1.1.0, with a Sigstore-signed BOM as the digest authority
- First release of the Fast Sandbox integration: a Firecracker microVM runtime behind the FastPath fleets backend, with pause/resume, ingress FastPath routing, and fsb template management in all…
- Snapshot persistence gains an opt-in PostgreSQL store with multi-process HA and a SQLite migration command, and sandboxes gain preStart/periodic lifecycle hooks
- refactor(marketing): share the store badges with the mobile launch post
- fix(host-service): treat a gh that cannot start as no token
- fix(desktop): markdown preview links follow the link click policy
- The PageIndex SDK, local or cloud — vectorless, reasoning-based RAG, end to end.
- Much faster indexing — the PageIndex Flash engine gets the tree from layout stats: no LLM involved for the structure generation itself, LLMs only write the node summaries, and tree expansion…
- Flash engine: the local default (mode="standard" keeps the classic LLM pipeline).
- wren: reach cube orderBy from the CLI and the MCP query_cube tool (be1f9b5)
- bigquery: support Application Default Credentials (19495ee)
- wren: bigquery and duckdb LIMIT wrap breaks on SQL ending in a trailing comment (0be34c0)
- bash curl -fsSL | bash -s -- --release --version 0.1.2026.9.21
- Start OpenSRE
- bash opensre
No releases match.
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.
- Nov 16, 2026431 releases · 120 agents · this week
- Nov 9, 2026717 releases · 152 agents
- Nov 2, 2026632 releases · 157 agents
- Oct 26, 2026628 releases · 145 agents
- Oct 19, 2026614 releases · 144 agents
- Oct 12, 2026545 releases · 150 agents
- Oct 5, 2026495 releases · 127 agents
- Sep 28, 2026429 releases · 121 agents
- Sep 21, 2026426 releases · 126 agents
- Sep 14, 2026299 releases · 107 agents
- Sep 7, 2026245 releases · 101 agents
- Aug 31, 2026268 releases · 99 agents
- Aug 24, 2026286 releases · 108 agents
- Aug 17, 2026234 releases · 95 agents
- Aug 10, 2026250 releases · 95 agents
- Aug 3, 2026230 releases · 89 agents
- Jul 27, 2026201 releases · 79 agents
- Jul 20, 2026199 releases · 82 agents
- Jul 13, 2026222 releases · 83 agents
- Jul 6, 2026198 releases · 81 agents
- Jun 29, 2026137 releases · 71 agents
- Jun 22, 2026157 releases · 72 agents
- Jun 15, 2026154 releases · 81 agents
- Jun 8, 2026158 releases · 72 agents
- Jun 1, 2026130 releases · 59 agents
- May 25, 2026123 releases · 67 agents