AI agent releases: Jul 20 – Jul 26, 2026
199 stable releases from 82 tracked agents and frameworks in the week of Jul 20, 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.
Jul 26, 2026
- 0.6.15 changlog
- Additional layouts
- [feat] Support Seed Audio generation
- Tool discovery works again for users whose config was frozen while discovery shipped opt-in, so discover_tools is registered instead of staying silently unavailable
- The desktop preview shows which provider and model is answering, and greets an empty session with an animated halftone donut and rotating composer hints
- Discovery gained a financial-data category for market, pricing, and company-data tools
- Resolve registry skills through the runtime's CrewAI+ client
- Recover from the GPT-5.6 tools + reasoning_effort 400
- Make tool calling work on the Responses API path
- 20e493e569c922d1253bdeff068c5e56c94957fb chore(release): bump version to 1.1.2
- e331e1c72a4ce88f2d5a2530a2db0b79a2126621 chore(release): refresh MCP lock for v1.1.1
- feat(models): add newly released Claude Opus 5 model support
- feat(mosaico): make docker/mosaico a self-contained deployment bundle
- fix(mosaico): report tool failures as FAILED, not empty success
- fix: serialize nested executor requirements
- fix: accept null path args in skill script/reference tools
- feat: add TrustedRouter as an OpenAILike model class
- Agent-requested skill imports. Ask the agent to install an attached .zip or .skill package and it routes through the app-owned preview and confirmation dialog (Select all / Invert) before anything…
- Preview before you import. Installed, GitHub, ZIP/.skill, and bare-Markdown skill candidates can all be inspected (parsed SKILL.md body, metadata, file/reference names) before selection.
- Sharper, zoomable PDFs. Pages rasterize at their on-screen resolution (no more soft A4/Letter pages) and gain a width-based zoom (0.5–3) with overlay controls and Ctrl/Cmd+wheel.
- feat(opencode): add native OpenCode integration
- feat(allowlist): add Julia (.jl) support
- feat(rules): add macro correctness checks to Rust review
- Godot 4.2+
- Active Ziva account
- Attaches to a reusable tab (about:blank, New Tab, or an existing chrome://inspect tab) instead of always opening a new one, so startup is faster and you get fewer prompts.
- Only closes leftover chrome://inspect tabs the harness itself opened — your tabs are left alone.
- Opens chrome://inspect/#remote-debugging at most once every 3 minutes, and guards against stale DevToolsActivePort files.
- Improved permission prompts for compound shell commands so the full command is shown when any part of it needs approval.
- Fixed disabled plugins still running their hooks and contributing other customizations, which could keep a broken hook active and break file-editing tools even after the plugin was turned off.
- Fixed MCP OAuth against providers that do not strictly follow the spec (such as Salesforce and Atlassian) by relaxing issuer validation and including the refresh_token grant.
Jul 25, 2026
- fix(docx): keep read-only commands byte-preserving
- Claude 5 models now use the correct context windows.
- cookbook: number the filesystem cookbook as 13_filesystem
- fix: FileSystem tools carry no instructions; the developer composes them
- cookbook: add examples - three agents you run and serve
- feat(config): update default model to OpenAI's GPT-5.6
- Add OpenAI's GPT-5.6 model support
- Add support for Gemini 3.6 models
- Tutti Mode now supports finer-grained task planning and execution feedback for complex workflows.
- File Manager adds a preview panel action row for more intuitive file previewing and operations.
- Mobile conversation rail UI aligned for a consistent experience across devices.
- Switchable app icon. Pick between the Light (dotted mark) and Dark (original) built-in icons under Settings → General; the choice applies to the window and dock live and on every launch.
- Claude subscription sign-in, shared and isolated. A first-class Claude subscription provider with a shared browser login (credentials in your ~/.claude) and an isolated setup-token mode…
- Tighter session permissions. Open Science now owns the Once / This conversation scopes end-to-end, with conversation grants that are memory-only, visible, revocable, and stable across backend…
- Style mobile pairing QR codes (50655ac09)
- Refine community management flows (384c72dee)
- docs: replace VPN-vendor references with generic wording (bcca885ba)
- Claude Opus 5 — Available on Anthropic and Amazon Bedrock with adaptive thinking (including xhigh), inference profiles, and prompt caching. See Providers.
- Anthropic gateway bearer auth — ANTHROPIC_AUTH_TOKEN authenticates against Anthropic-compatible gateways that require Authorization: Bearer, including compaction and branch summaries.
- Faster, more resilient model catalogs — pi.dev catalogs revalidate with If-None-Match so unchanged providers answer with an empty 304, and llama.cpp models stay listed across restarts. See llama.cpp.
- Vector Stores: Apply every operator in a Cassandra compound field filter (e.g. { age: { gte: 10, lte: 20 } }) instead of stopping after the first, so the remaining bounds are no longer silently…
- Vector Stores: Stop the Chroma where-clause translator from dropping filter conditions.
- Vector Stores: Skip "*" wildcard filter values in Milvus instead of matching them literally, so a filter like { user_id: "*" } no longer returns zero memories
- fix(core): reset user away timer on final STT transcript AGT-3149
- Fix stripping of Gemma reasoning markers in streamed output
- feat: Spatius avatar plugin
- One command to the workbench - nanobot webui prepares the local channel, starts the gateway, opens the browser, and leads first-time users into guided setup.
- A stronger browser workbench - Topic-style chats, composer model switching, quoted follow-ups, document attachments, native folder selection, responsive settings, and clearer fallback state.
- Agent coordination without core sprawl - Inline subagents can consult inside the active task, authorized work can continue through verification, and typed runtime context keeps provider-owned…
- OpenAI max-effort reasoning turns no longer time out mid-thought: requests ask for reasoning summaries and the idle budget scales with reasoning effort
- The /model picker shows pretty model names and matches queries like "opus 4.8" against them
- New and unrecognized models no longer silently fall back to a 200K context window
- Updated bundled Claude CLI to version 2.1.220
- Add AdvisorTool support for Anthropic and OpenRouter
- Add 'us'/'eu' multi-region locations to GoogleCloudProvider location type
- Add external_web_access option to WebSearchTool for OpenAI Responses
Jul 24, 2026
- fix(discovery): spawn PowerShell install commands natively on Windows (f3981dbfe)
- fix(desktop): use augmented PATH for model discovery subprocess (3bd3a014c)
- Improve huddle audio failure handling (fb4a801ad)
- Improve Claude adaptive thinking handling across more response shapes.
- Avoid OpenAI Responses phase handling that could break some conversations.
- Preserve grep symlink paths in search results. (@remixz)
- For the installer path, run CodeWhaleSetup.exe; it installs codewhale.exe, codew.exe, and codewhale-tui.exe under %LOCALAPPDATA%\Programs\CodeWhale\bin and adds that directory to the current-user…
- Extract the archive for your machine: codewhale-windows-x64.zip or
- Run install.bat (copies to %USERPROFILE%\bin)
- feat(python): add per-call MCP tool cancellation (67693b72)
- feat(site): add Storage category to community catalog (7db6bc0c)
- feat: add storage docs page (972f3727)
- The one where Voice Mode goes live, Vellum becomes its own web search provider, the memory graph and bookmarks graduate to everyone, and the billing page gets a full restructure.
- Voice Mode, live. The most delightful, real agentic AI voice mode you'll ever use! You can now have real-time voice conversations with your assistant.
- Vellum is now a web search provider. Web search used to be a "mode" your assistant toggled on and off. It's now a proper provider, sitting alongside Google and Tavily in the search provider catalog.
- feat(mobile): diff review surface — inline comments, native review toolbar, rebuilt diff renderer
- fix(desktop): remove misleading auth_completed event + label pageview pathnames
- chore(desktop): bump version to 1.16.1
- Upload local media, replay anywhere — local files are uploaded to the cloud media store so that replays work on any machine without manual file provisioning.
- Media config flows through the full run — media configuration is now passed into the runner and threaded through the session manifest, so replay picks up the right assets automatically.
- kane-cli now ships its own Node runtime — the runtime is bundled per platform (kane-cli-node-<platform>) so kane-cli works without requiring a separately installed Node version.
- Fix detection of Anthropic preview tool-use blocks.
- Preserve strict tool schema property names.
- Dispatch execution_end hook on failed crew and flow executions.
- 28a4203bc2a5d3051ea22c63a2ac89b569294763 MUL-5261: revert domain-specific prompt additions
- ecce589867b949a5a1751a69685a0a61b76e606f MUL-5265: GitHub API-snapshot PR cards — CI status + mergeability
- 3c6bebaff04ff4fc8b57d6a2ab179b2af6b1c065 MUL-5274: allow explicit persistent service handoffs
- 6f8a7fe8b2cd619e82683a70de252238f333052f: feat(client): reconnect the event stream after it drops (@meowgorithm)
- d291d9f2c189d1d8c1c4a18d4b73650fc4d20ac9: feat(cmd): add --channels opt-in flag for MCP channel servers (@joestump)
- 37fc995daca4486501acb6d03445a4bd6969d7f3: feat(mcp): OAuth 2.1 authorization for HTTP MCP servers (@taciturnaxolotl)
- Session recovery silently found nothing for any project path containing a dot, a space, or any other non-alphanumeric character (closes #209, reported
- Claude Code names ~/.claude/projects/ entries by folding every character outside [A-Za-z0-9-] to -.
- One of the three sits on a live install route. marketplace.json declares "source": "./", so the plugin root is the repository root and the SKILL.md restore block resolves…
- Claude Opus 5 is available
- Fixed premature stdin closure when background tasks are in flight: query() no longer closes stdin on the first result frame when background tasks (e.g.
- Updated bundled Claude CLI to version 2.1.219
- feat: chunk batch reflection scoring
- Revert "feat: chunk batch reflection scoring"
- release: merge dev-v2.0.25 into main
- Live context-usage indicator. See how much of the model's context window a session is using, right in the workspace.
- Native Office document previews. Preview DOCX, XLSX, and PPTX files inline, rendered in isolated out-of-process frames so a document can't destabilize the app.
- Zoom & pan in the image viewer. High-resolution scientific figures open with Ctrl/Cmd+wheel / pinch zoom (1×–8×), drag-to-pan, and double-click to reset — with reduced-motion respected.
- feat(llm): add iFlytek Spark as a built-in provider
- feat(allowlist): add GraphQL (.graphql/.gql) support
- feat(viewer): fix top breadcrumb to get more comfortable experience
- fix(examples): add missing workspace fields to governance interceptor
- fix(auth): report gateway authentication status
- refactor(supervisor): pass agent proposal state explicitly
- Fix Burrito release tag verification
- Add generic tracker interface with Linear adapter
- Add GitHub Issues tracker adapter
- The desktop app now sends a system notification when an agent finishes its turn while the window is unfocused — no more staring at the window waiting, or missing that a long task completed while…
- Common UI bindings added: Cmd/Ctrl+B toggles the sidebar, Cmd/Ctrl+L toggles the workspace panel, Ctrl+Tab / Ctrl+Shift+Tab cycles between open conversations, and Cmd/Ctrl+T starts a new…
- Codex slash commands are restored on the direct-CLI session path, and dead resume anchors now recover instead of leaving the session stuck.
- FileSystem (durable agent filesystem): A net-new state primitive — agents get a private, persistent filesystem with pluggable DB/local backends and fail-closed per-user namespace isolation.
- Cookbooks: Rewrote the AgentOS cookbook (284 → 132 files, 24 lessons) and overhauled the agent quickstart for Gemini 3.6 Flash.
- cookbook: overhaul agent quickstart for Gemini 3.6 Flash
- fix(runtime): include execution status in action metadata
- feat: add Oracle Cloud provider
- feat(latchshot): add Latchshot provider
- /api/v2 API prefix — every business endpoint (memory/*, ome/*, knowledge/*) is now served under /api/v2, aligning the open-source API with the EverOS Cloud contract.
- Native OpenTelemetry tracing — memory operations (add/flush, memcell boundary, episode extraction, search, and OME reflection) export to any OTLP backend (e.g.
- Retrying pool acquire policies — Go sandbox pools now support AcquirePolicyRetryNextIdle and AcquirePolicyRetryNextIdleThenCreate, allowing Acquire to skip stale idle candidates before failing or…
- Sandbox create metrics — Sandbox.Create now reports fire-and-forget sandbox.create latency events to the lifecycle server.
- Client IP header — the SDK now best-effort detects the host intranet IPv4 and sends it as OPEN-SANDBOX-CLIENT-IP on outbound requests. User-provided headers are never overwritten.
- emitter: off() with a compiled regex should only remove the matching listener
- emitter: match compiled regex anywhere in the event path
- cache: clear cache entry on ChatModelToolCallError
- Go SDK PTY: Improved interactive terminal capabilities.
- Python SDK X-API-Key coverage: Carry X-API-Key on all request paths.
- Go SDK Connect protocol frames and dataplane fixes: process.Process/Start, exit-0 end event parsing, encode deduplication, and dataplane defect fixes.
- PawApp SDK & Kanban App: A new mini-app platform that lets plugins build rich interactive UIs on top of QwenPaw. Ships with a built-in Kanban task board app for project management
- User-editable Agent Mode: Create and edit custom agent loop modes directly from the console — define iteration limits, stop conditions, and approval gates without touching code
- Oh-My-Paw Plugin: Five ready-to-use Agent Loops (UltraQA, Ralph, Ultrawork, Autopilot, Team) for multi-agent collaboration and long-running tasks
- Constrained tool sampling — Tools can prefer or require strict JSON Schema sampling or use OpenAI Lark/regex grammars, with model capability metadata preventing unsupported requests.
- OpenRouter and Kimi Code sign-in — Use /login to authorize OpenRouter or a Kimi Code subscription without manually configuring API keys. See OpenRouter.
- Session-aware, streaming bash integrations — Bash tools receive current session/model metadata, while direct RPC bash commands stream correlated output.
- #2065 527d485 Thanks @7Sageer! - Add global default MCP server timeouts in config.toml and env vars.
- #2104 66f611a Thanks @RealKai42! - Fix loss of thinking content with OpenAI-compatible endpoints that return reasoning under a different field name (e.g. newer vLLM).
- #2089 ca38b7e Thanks @liruifengv! - Remove the toolbar tip that suggested trying the "superpowers" plugin.
- Redesigned TUI header: left-aligned, shows git branch beside the working directory, vertical auth list, and capped MCP/skills lines
- Much faster fuzzy filtering in pickers, with instant /model open for large remote catalogs
- New versioned harness API socket lets external apps drive jcode sessions
- Custom Agents (Markdown Format). Added support for defining custom agents using Markdown files (agent.md) with YAML frontmatter and H1-delimited system prompts.
- Added an optional index argument to /copy so /copy <n> copies the n-th most recent response to the clipboard, while /copy and /copy 1 still copy the latest.
- Improved /codesearch to render results progressively as they stream in, showing a live count while loading and letting you cancel an in-flight search with Esc instead of blocking until the whole…
- Make RequestUsage and RunUsage support arbitrary fields for upcoming genai-prices
- Cache per-message OTel serialization to avoid O(n²) instrumentation cost
- Fix ContentFilterError for real Model Armor response blocks (finishReason: MODEL_ARMOR) and record moderation wire-truth cassettes
Jul 23, 2026
- fix(desktop): suppress Windows console flashes and reject WSL bash alias (5afa16157)
- fix(desktop): fix Windows PATH clobber and .cmd shim EINVAL (cca16635d)
- Gate default relay auto-connect behind release flag (e67303f60)
- agent-framework-github-copilot: Forward input attachments (images, documents, and other inline binary content) to GitHub Copilot as inline blobs
- agent-framework-github-copilot: Promote the package from release candidate to stable
- Full Changelog
- Arbitrary command execution in goose CLI via goose review (via git core.fsmonitor config) - GHSA-r5pp-p5r8-466r (credit to @0xmagic0)
- Support for latest Gemini models #10630
- Pass working_dir to the Stop hook context #10296
- TwelveLabsTools: Added support for Marengo video embeddings. See docs.
- Slack: respond_to_other_agents flag for peer-agent communication.
- Learning Stores: extraction_tool_call_limit to prevent infinite loops.
- docs: update OpenWiki
- feat: generate mermaid diagrams in the generated wiki
- docs: update OpenWiki
- Consolidated goal tool — Replaced the separate create, get, update, and storage modules with a single multi-action goal tool while preserving the slash command, API, goal strip, and active-goal…
- Resumable, secret-safe chat compaction — Compaction now preserves authorization, evidence, pending work, loaded skills, and secret references in a fixed resumable-state summary.
- Smarter runtime secret redaction — Masking of usr/.env values is now limited to API keys and login/password credentials, preventing ordinary settings from corrupting chat text.
- Added axe-core accessibility audits with agent-browser a11y [url], WCAG tag filtering, selector scoping, iframe-aware text and JSON results, an embedded offline and CSP-safe audit engine, and a…
- feat(code_ast): add tree-sitter Dart grammar for syntax-aware splitting
- Target-state inspection improvements in cocoindex show
- Render tracked target-state paths readably
- Redesigned app icon. A new dotted-mark app icon across macOS, Windows, and Linux builds and the in-app branding.
- Session pinning. Pin key sessions to keep them at the top of the sidebar.
- Resilient downloads. Model-runtime and asset downloads now resume from a byte range after an interruption, show transfer speed, and retry automatically on transient failures.
- fix(docker): remove broken npm self-upgrade
- fix(security): protect Docling Serve outbound requests
- fix(auth): verify current password on password changes
- #1922 cb4c1c7 Thanks @cjol! - Support both AI SDK v6 and v7.
- Uses stepCountIs, system, experimental_telemetry, onStepFinish, and
- The tool-execution-finished event is normalized across majors: v6's
- fix(e2e): detect gateway workload for health port-forward
- feat(tui): add config key support to provider create/update forms
- fix(vm): reduce registry rootfs staging pressure
- Support structured output
- Support environment information injection
- Support OpenSandbox, Daytona, K8s, Bubblewrap
- Claude and Codex now run over a new direct-CLI session path that talks to each CLI's native protocol instead of going through a third-party ACP shim.
- The chat stream now surfaces backend advisories — like a rejected model/mode switch or a Codex deprecation notice — as inline tips, without confusing the running/timer state.
- Entering a team session no longer wakes up every member behind a full-screen overlay. Only the leader warms up first.
- Full Simplified-Chinese localization of the v2 shell: landing, sign-in, marketplace, community, onboarding
- EN / 中文 switcher in the nav rail, native-speaker reviewed, with CI guards for string + plural completeness
- Commonly HQ — everyone joins the shared hall on sign-up
- e0e9383efeca0398ae6720356f4a3b66ab94df94 Improve UI animation transitions (MUL-5172)
- f2ef7ab0d1842967a79c924f04d9393e2ecf86c9 MUL-5127: show issue agent activity in Inbox
- 09dce598dfef5f24b35c11222b593c3937842f62 MUL-5155: KAP-1051: diagnose Codex thread/start timeouts fail-closed
- Voice call dashboard and WhatsApp calling fixes.
- Captain generation path, overview, and multimodal session fixes.
- WhatsApp coexistence, BSUID, phone lookup, and reply-window fixes.
- 检索标签改为严格 AND 语义:find() 和 search() 的多个 tags 现在要求结果包含全部标签,REST API、Python SDK、CLI、本地客户端和 LangChain 集成保持一致。
- cuVS 支持可选的请求 micro-batching:兼容的 exact brute-force 查询可合并为一次 GPU matrix search;同时降低恢复和过滤搜索的主机开销,并避免加载未请求的向量字段。
- 飞书/Lark URL 可经 UnderstandingAPI 导入:支持异步导入、最终资源 URI 回填、产物图片重写和安全的临时凭证传递。
- Langfuse OpenTelemetry integration example for tracing add, flush and extraction, search, and reflection operations.
- Process a queued modified event as a deletion when the source file disappears, preventing stale indexed rows and permanently failed queue items.
- Limit synthetic Langfuse child spans to responses with stage details so live-server traces use real telemetry.
- fix(agents): count billed calls that fail to parse against cost_limit
- Add mistral_prompt_cache_key setting and pass parallel_tool_calls to the Mistral SDK
- Add ToolFailed for model-visible failures without retries
- Add optional run_id= to agent runs, durable wrappers, and UI adapters
No release notes published.
- Improved branch filtering with all, local, and remote options in the branch picker. (#60857; thanks chirivelli)
- Settings UI: Improved support for window widths below 900 px. (#60786; thanks feitreim)
- Improved semantic token highlighting to distinguish function parameters from local variables. (#60949; thanks lingyaochu)
Jul 22, 2026
- Python: Add MCPStreamableHTTPTool security guidance for custom http client
- Harden workflow credential selection
- Python: preserve Gemini 3 thought_signature across function-call replays
- Fixed find role to match implicit ARIA roles and browser-computed accessible names through the accessibility tree, so semantic elements like <h2> (heading) and <ul> (list) resolve, with…
- Fixed element-not-found errors to preserve the locator detail (selector, role, name, or index) instead of flattening every miss into one generic message, and aligned the advertised find actions…
- @cooleryu
- fix(desktop): strip GIF metadata extensions before upload (47d7eb698)
- feat(desktop): gate sign-out behind key backup + typed confirmation (fd967c6ea)
- feat(desktop+acp): spawn a harness per (agent, community) pair at GUI startup — warm sockets, lazy LLM pool (61cc738ee)
- Run-bound agent secret access — Agents can now fetch the secrets they've been granted on demand through a run-bound API, instead of relying only on ambient environment injection.
- Local agents run on Windows — The embedded ACPX engine no longer wraps local agent commands in a generated Bash script, so Claude, Codex, Gemini, and custom ACP adapters now spawn natively on…
- User-named PAPERCLIP_* bindings now reach your agents — The heartbeat previously stripped every PAPERCLIP_-prefixed env binding before resolution, so a secret you deliberately named something like…
- NodeChatStream: one agent's live chat, streaming as it works.
- FleetTable* and *NodeStageStrip: a dashboard for fan-out runs
- RunEventLog: now structured, severity-tinted rows, coalesced
- terminal_reason on ResultMessage: ResultMessage.terminal_reason now surfaces why the query loop ended ("completed", "max_turns", "aborted_streaming", "aborted_tools", etc.).
- Typed model_usage on ResultMessage: ResultMessage.model_usage is now typed as dict[str, ModelUsage] instead of dict[str, Any], with a new ModelUsage TypedDict that mirrors the TypeScript SDK's shape.
- Updated bundled Claude CLI to version 2.1.218
- Terminal spawns inside tmux now open in split panes instead of new windows
- Expanded Jcode Bedrock subscription model catalog
- Todo planning now measures how well objectives and feedback loops match your request
- Vector Stores: Fix reset() silently leaving stale vectors behind on local (on-disk) Qdrant when the old collection directory could not be removed, for example an open file handle on Windows or NFS
- Core: Stop update() metadata from overwriting or injecting user_id, agent_id, run_id, or actor_id.
- Vector Stores: Scope Pinecone delete_col()/reset() to the configured namespace instead of deleting the whole index, so resetting a namespaced Pinecone store no longer wipes out the other…
- 互动桌面宠物:新增 Electron 原生透明悬浮窗口,内置搭搭、弧弧、补补、回回 4 个角色;宠物会根据工作中、等待处理、失败和空闲状态切换动作,并可展开活跃任务、点击返回对应会话。
- 本地自定义宠物:支持导入透明 PNG / WebP 制作轻动画宠物,也支持导入 1536×2288 的 v2 逐帧动画图集;可调整尺寸、动画开关和任务面板显示,整个导入过程在本地完成,不调用当前聊天模型。
- 完整的桌面 Agent 管理:设置页可以浏览内置、用户、项目、插件和策略等来源的 Agent 及覆盖关系,并创建、编辑或删除用户级与项目级 Agent。
No release notes published.
- sandbox: bump default inner-sandbox image to 1.1.0
- chore: release v1.3.1
- docs(agents): keep project skills synchronized
- fix(proxy): include OPA deny reason in CONNECT 403 response
- fix(cli): respect CARGO_TARGET_DIR in openshell wrapper script
- This patch release tightens up Composio task execution and makes tool-calls more resilient, so your agents can get to work faster and recover more gracefully when model output gets messy. 🚀
- Thanks to everyone who contributed to v0.63.1: @M3gA-Mind and @github-actions[bot].
- #1860 f5b1dd8 Thanks @mattzcarey! - Group SDK-managed initialization, startup, chat interactions, turns, and durable submissions into semantic phases.
- #1860 f5b1dd8 Thanks @mattzcarey! - Add agents/observability/ai with wrapAISDK for AI SDK v6 and createAISDKTelemetry for AI SDK v7.
- #1959 a3cbed1 Thanks @mattzcarey! - Cache MCP JSON Schema conversion for the current catalog on each live connection, and let Think agents skip direct MCP AI-tool exposure when those tools are…
- fix: keep running task tool_result parts byte-stable to protect prompt cache
- fix: send ACP client version during initialization
- fix(tasks): prevent duplicate active task launches
- Remote SSH compute hosts. Register a remote machine over SSH as a notebook execution target and route runs to it alongside local kernels — the first step of Phase 4 remote compute.
- Headless task SDK and CLI. A zero-dependency Node.js package (open-science) exposes a task-automation SDK and headless CLI that share the same daemon, projects, sessions, credentials, and…
- Desktop task notifications. Long-running notebook runs and agent tasks now fire OS-level desktop notifications on completion or failure, so the window doesn't need to stay in the foreground.
No release notes published.
- Look-ahead-bias fix across all five portfolio optimizers
- Security hardening — all 10 findings from the 2026-07-10 external audit closed: AST-hardened backtest sandbox, short-lived SSE auth tickets, hardened Compose, /correlation auth + rate limiting…
- 80× signal-alignment vectorization
- feat(pages): add social preview meta tags and og-image
- fix(llmloop): drain per-file comment work without racing pool submissions
- fix(diff): review merge commits against their first parent
- docs: changelog and blog post for v0.8.4
- fix(control-plane): show all mental models instead of capping at 100
- fix(codex): implement strict_schema via forced tool call + repair invalid \escape
- d43e500ff62f15c3e8ef76f3deb1c1d6f0a9db52 MUL-5100: Move issue table grouping to the server
- d804dedcc64c8a171164fb9af385a07c06562752 MUL-5137: fix(agent): parse Grok ACP token usage from session/prompt _meta
- 3171e6607fef5f5765fd4d00c8e9f8c7bdb4b815 MUL-5141: fix(agent): inject --yolo in Qwen headless runs so shell/edit/write tools are available
- #1992 a8f1ca3 Thanks @RealKai42! - Support selecting a thinking effort level from ACP clients: the thinking picker now lists the current model's declared levels…
- #1735 ce0e3ce Thanks @7Sageer! - Let custom agent files restrict which sub-agent types they may delegate to (v2 engine only).
- #1735 ce0e3ce Thanks @7Sageer! - Support custom agents defined as Markdown files with frontmatter, usable as the main agent or a sub-agent (v2 engine only).
- Add bsk CLI auto update support
- Fix pnpm lint formatting
- Add PR CI workflow
- Optimized agent reasoning to mitigate redundant thinking loops and duplicate tool invocations.
- sdk/go/v0.0.1
- feat(openai): support explicit prompt caching for gpt-5.6
- Inline text-like files in MistralModel prompts
- Support overriding the tool-retry budget at run/iter/override time
- fix(xlsx): round-trip in-cell images through dump
- @Hanxcellent made their first contribution in
- 0.6.14 changelog
- [agent] Expose clip fades to Agent
- [fix] Convert HEIC generation references to JPEG
Jul 21, 2026
- feat(evals): group trajectories by run and cross-link Braintrust experiments
- chore(docs): Add docs ownership and review rules
- feat(core): support OpenAI Chat Completions via openaiEndpointFormat
- Updated bundled Claude CLI to version 2.1.217
- Update custom gemini llm connection logic to be used for all 3_x models, not just 3.1 (4ef72fe)
- This release closes #208 and the one gap v3.8.0 dogfooding surfaced.
- Pi extension: plan resolution no longer depends on the live shell cwd (closes #208, reported
- Every Pi injection now states which plan it resolved (plan: <id> or plan: root). A stale .planning directory shadows a root task_plan.md by documented precedence.
- You no longer have to click "Allow in Chrome" for every task! The harness now opens one connection and keeps it, so you accept once per Chrome session and never see the popup again until Chrome…
- The one where Skills and Plugins become one page, your assistant gets a voice you can actually pick, and the billing page grows up.
- My Superpowers. Skills and Plugins used to live on separate pages with separate vibes.
- Pick your assistant's voice. Managed TTS now has a voice picker. Browse available voices, preview each one before choosing, and see which provider it comes from.
- Keep avatar preview visible during upload (f609fcee0)
- fix(desktop): keep machine onboarding for unrecognized identities after reset (fd55ab662)
- fix(managed-agents): stale harness pin shadows runtime edits; add Restart quick action (ec0e450f1)
- Added MOQTransport, a Media over QUIC (MoQ) transport that gives bots a bidirectional, low-latency audio + RTVI channel over QUIC instead of WebRTC or WebSockets.
- The bot runs as its own MoQ server (serve=True) and accepts the browser's direct connection, removing the need for a separate moq-relay process in local dev.
- Audio rides a single Opus track; RTVI messages (including the transcript) ride a compressed, ordered JSON stream track, so MoQ is on par with the Daily and WebSocket transports for RTVI support.
- agent-framework-azure-cosmos-memory: Add an alpha Azure Cosmos DB semantic-memory context provider with fact extraction, user profiles, samples, and integration coverage
- agent-framework-azurefunctions, agent-framework-core, agent-framework-durabletask: Add HITL response-URL addressing for requests raised from inside workflows
- agent-framework-core: Add cross-session origin attribution to context-injected messages
- bb desktop 0.33.0
- Installed and linked plugins, including their enabled state, are now global to the current user instead of isolated by Herdr session.
- Added a live-agent CLI facade with named start, atomic prompt, logical send-keys, and server-owned wait workflows.
- Added transient declarative Agent view queries through agent.view.set/clear; filtered and sorted views now define sidebar, mobile, mouse, and agent-keybind navigation order.
No release notes published.
- Thank you @yh928 for improving tool-call robustness by recovering pipe-garbled <tool_call> tags from small models.
- feat: Add gemini 3.6 flash and 3.5 flash lite
- release: 0.2.2
- Verifiable release source archives — GitHub releases now include deterministic, checksummed source archives with instructions for rebuilding standalone binaries.
- Resilient compaction and branch summaries — Transient provider failures now follow the configured retry policy, with retry lifecycle events available to interactive, JSON, RPC, and SDK consumers.
- Added deterministic, checksummed source archives to GitHub releases with documented standalone binary rebuild instructions
- Added Connector Gateway UI and MCP wiring.
- Added typed workflow event infrastructure for app-level workflow events.
- Improved the in-session preview panel experience.
- AionUi has entered a strategic partnership with Kimi (Moonshot AI). To thank our community, Kimi is offering free "Allegretto" subscription plans ($39/mo · ¥199/mo value) to AionUi contributors.
- Campaign runs July 21 – September 21, 2026. Two ways to claim: veteran contributors with 10+ merged PRs in the AionUi repo, or 3 newly merged PRs during the campaign via the claim-an-issue process…
- Campaign details: · Claim thread
- fix(runtime): stage symlink-safe copies for LocalDir uploads
- fix(runtime): retry transient sandbox startup failures
- feat(runtime): resolve sandbox ports over a shared Docker network
- Codex signs in with a subscription, not just an API key. Add a Codex provider backed by the locally installed Codex CLI and reuse an existing ChatGPT/Codex subscription login — either your…
- Two more model vendors. SenseNova (SenseTime, China region — Messages + Chat Completions endpoints, sensenova-6.7-flash-lite multimodal) joins the built-in vendor list, and the GLM Coding Plan…
- A navigable Files tab. The project file library gets indexed pagination and session grouping, so large projects list and scroll their uploads and generated artifacts without loading everything at…
- timeout: per-command timeout in ms (default 60000)
- onError: "fail" (default, aborts the pack) or "skip" (warn and keep the original content)
- Language hints are now resolved for extensionless filenames in subdirectories (for example docs/Makefile), instead of falling back to no hint.
- /new command as an alias for /clear
- ACP review support: approve or revert agent changes from ACP clients, with live refresh
- Full migration to ConfigOrchestrator for configuration handling
- ci(e2e): reuse prebuilt CLI and gateway artifacts
- docs: fix broken links and small inconsistencies
- fix(gateway): honor tty flag for interactive exec
- revert(agent-profiles): restore default tools in cloud launches
- b03e118aa711303cbdf9b1cfbbb7859bfb7966a0 docs(changelog): add 0.4.7 release entry (MUL-5111)
- 645e8a298343aa81e4e49e9728c2784a59a9335d feat(issues): 确认指派弹窗去掉打开时的 run 预览等待 (MUL-5010)
- 5a11232c474bc13ceb156e8c2e006d5f5288fd1f feat(rich-content): unify Chat and Issue/Comment on one RichContent renderer (MUL-4922)
- 📄 Richer file viewing: inline PDF preview with review comments, find-in-file across the editor plus markdown and notebook previews, a diff-viewer "Wrap lines" toggle, and a file panel that stays…
- 🎨 Custom themes: a guided theme editor with shared accent/tint/contrast and sidebar translucency controls, a new Nord palette, a randomize dice, and a file editor that follows your Omnigent theme
- 📥 Bring your existing work in: import Claude Code and Codex chats with omni import (up to 50 at once), automatic first-turn session titles
- fix(tui): hide council and dynamic councillors from sidebar
- docs(readme): add issue #823 models to per-agent recommended ranges
- refactor(config): remove legacy tmux and council.master config keys
- feat(examples): add Gerrit CI integration example (Jenkins + Gerrit Trigger)
- feat(pages): serve landing site at custom-domain root with clean URLs
- feat: add pot code review rules
- Always fetch MCP instructions via a durable step under DBOSDurability
- Temporarily withdraw durable wrapper-agent deprecations
- Added a /effort command to view and change the current model's reasoning effort, with a left/right timeline-gauge picker and a direct /effort <level> form so you can trade latency for depth on the…
- Added an --effort flag to select a model's reasoning-effort variant when launching the CLI.
- Added stable, user-facing model slugs that appear in the /model picker and are accepted by --model, so you can pin a specific model reliably across sessions.
- athena: push LIMIT into SQL and strip trailing semicolon on wrap (79af489)
- canner: strip trailing semicolon on unlimited query path (c7e7462)
- clickhouse: catch SqlglotError in type string parsing (e99d5ee)
Jul 20, 2026
- Refused batch script CLI spawning on Windows: Blocked spawning .bat/.cmd CLI scripts (including npm's claude.cmd shim) on Windows to prevent command injection via cmd.exe metacharacter re-parsing…
- Windows cmd.exe metacharacter rejection: resume and session_id values containing cmd.exe metacharacters (& | < > ^ % ! ") or newlines now raise ValueError on Windows, preventing injection even if…
- Dash-prefixed extra_args value binding: extra_args now uses --flag=value form when the value starts with -, preventing a dash-leading value from being misinterpreted as a separate CLI flag
- LanceDB FTS with_position crashes optimize(), bloating the index
- FTS now defaults to with_position=False — lossless for EverOS, since
- A marker-guarded startup migration rebuilds pre-fix indexes and reclaims
- plan previews without editing — running with --plan records the change and stages the work; nothing in the suite is touched until you explicitly apply it.
- Resuming a pending plan works — the same reconcile command picks up where it left off, supersedes plans that have since moved, and never silently does nothing.
- Pass in your own file — reconcile --from <file> --source-id <id> accepts an explicit source instead of guessing slugs, with fail-fast validation before any work begins.
- Every surface now shows the same numbers. CLI, TUI, menubar, desktop app, and web dashboard totals all come from one durable aggregation path and match exactly, including history whose session…
- Never lose history again. The daily cache carries forward every (day, provider) slice a re-parse can no longer derive, and adopts days from older cache generations instead of wiping them on schema…
- True Lifetime period on the CLI, dashboard, desktop app, and menubar. The desktop tab formerly labeled "All time" showed a 6-month window.
- 0.6.13 changelog
- Slip Edit in UI
- [feat] Add AI transition and ripple delete to timeline gap menu
- fix: add Gemini API key to credential redaction and add missing unicode regex flags
- docs: update OpenWiki
- fix: tolerate optional OKF index metadata
- lsp_rename - Rename a symbol and all its references in the project
- lsp_replace_symbol - Replace, insert before/after, or delete an entire function/method/class by name
- When focused on the chat view: <kbd>l</kbd> or <kbd>right arrow</kbd> focuses on the sidebar
- Hermes got dramatically faster — first token in a fraction of the time — Cold-start "Initializing agent..." used to eat ~4.3 seconds before your first turn even reached the model.
- The desktop app speed wave — 20+ targeted perf PRs — Long replies used to cost 14× more CPU in the markdown splitter than they do now; giant diffs froze the review pane until we virtualized it.
- Manage your Nous plan from the terminal — /subscription and /topup — Changing your subscription used to mean a trip to the billing website.
- fix(desktop): composite harness loader opacity (77900e69)
- fix(desktop): modal sign-in + branded communities onboarding page (8a65d81f)
- test(desktop): model membership relay in invite QR spec (9ab52d4f)
- Skill Studio & skill organization — A three-pane skill IDE with sandboxed test runs, so you can author, edit, and try out skills without leaving Paperclip.
- Attention queue & Decisions — A new attention queue and Decisions surface bring everything that needs your input into one place, with faster scrolling and readable, mobile-friendly decision rows.
- Better search — Search gains filters, sorting, and operators, with command-palette parity so you can find issues and entities the same way everywhere, plus a new bulk extract endpoint.
- Authenticate skill registry downloads
- Update snapshot and changelog for v1.15.4
- Restores the portable Linux x86_64 download by including the complete runtime needed to build bundled OpenSSL
- Windows x86_64 and ARM64 downloads publish reliably from the release pipeline
- Release publication now remains gated until every Linux, macOS, Windows, and FreeBSD artifact is available
- bank/banks, tags, recall ("auto", a fixed query, or false),
- Recall is scoped per bank, so distinct banks keep distinct recall scopes
- Every injected block and remembered tag is capped by an explicit bound, so
- A real multi-kernel notebook runtime. The one-shot Python executor is replaced by a persistent runtime registry: warm Python, R, and REPL control-plane kernels routed through a single per-session…
- A third agent backend: Codex. Pick Codex in onboarding or Settings alongside Claude Code and OpenCode.
- StepFun, multimodal image input, and clearer provider labels. Add StepFun as an official vendor (Global + China regions, Messages / Chat Completions / Responses endpoints, step-3.7-flash multimodal).
- agno.scorer — Turn a run into a number
- CodeScorer: wraps any callable (bool | float | Score; typed-field comparison under output_schema recommended)
- JudgeScorer: LLM judge with the model always an explicit choice and numeric verdicts normalized to exact endpoints ((score - 1) / 9)
- Use adaptive thinking controls for Kimi models on Anthropic-compatible providers, with summarized reasoning output by default. (@chouqin)
- Reduce OpenAI provider header timeouts during slow connection setup.
- Respect provider-defined reasoning options instead of falling back to the wrong reasoning controls.
- #934 c5b6103 Thanks @tt-a1i! - Allow ACP sessions to start with configured non-OAuth model credentials instead of requiring terminal login.
- #1967 ad8cc85 Thanks @sailist! - Run web servers foreground-only end to end: the /web slash command now always starts a new server, and the kimi web kill / kimi web ps subcommands are removed —…
- #1948 f6f4192 Thanks @sailist! - Fix running subagents not observing permission mode switches made after they started.
- Unicity AOS Community Edition 2026.1.3.
- This product release bundles Astrid Runtime v0.10.4. The exact runtime release identity, WIT commit, and compatibility pins are published in runtime-compatibility.toml and in every archive's…
- The release also contains the 19 signed, installable capsule artifacts selected by Community Edition. Published astrid-capsule-* identities remain unchanged.
- The welcome page's default prompt suggestions are now task-oriented — "explain this project", "find duplicate files in this folder", "build me a personal homepage" — so the first click shows real…
- Prompt suggestions got a cleaner look: plain left-aligned text with a subtle hover arrow, replacing the old full-width bordered cards.
- The "More" assistants dropdown on the home page is now a wide multi-column panel that opens on hover and adapts its column count to the window size — much faster to scan when you have many…
- 9e2cfe098c28c65cbe8193d9b6146bc2aad215e8 MUL-5038: fix project resource preparation decoding
- Skills as first-class citizens — SkillToolset discovers skills through progressive disclosure, so the model sees only names and one-line descriptions until it loads one, keeping context lean.
- Dynamic tool selection at runtime — pass tools=... to run / run_async so one reusable Agent serves different teams, tenants, and tasks.
- Native async tools — @tool routes async def callables to a Tool's new async_function.
- Credential Vault requires dns+nft enforcement — DNS-only egress can no longer carry credential-bound traffic. Migrate to [egress] mode = "dns+nft".
- X-Forwarded-Proto no longer trusted by default — the server proxy strips client-supplied forwarded headers and rebuilds them from the connection.
- OPENSANDBOX_EGRESS_MITMPROXY_SCRIPT removed from the env allowlist — the variable let a sandbox point the egress sidecar's mitmdump -s at any file under the shared /opt/opensandbox emptyDir…
- Browser tool support: the client bundles browser capabilities and prefers reusing the system's installed Chrome / Edge, with optimized browser startup and access performance.
- UI polish: refined visuals and interactions for the chat view, message bubbles, tool-call steps, and channel pages.
- Windows code signing: Windows installers are now code-signed, reducing security warnings during installation.
- feat: add po code review rules
- feat: add install.ps1 for Windows one-line install
- feat(pages): expand Integrations sidebar group by default
- core-py: sync Rust crate versions in Cargo.lock (7f352e0)
- deps: patch idna, cryptography, and pytest security advisories (8c72492)
- finish wren-engine → wrenai rename cleanup (0733d37)
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