AI agent releases: Aug 10 – Aug 16, 2026
250 stable releases from 95 tracked agents and frameworks in the week of Aug 10, 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.
Aug 16, 2026
- feat(core): add model call timeouts
- feat(sandbox): add run-scoped sandbox working directories
- feat(sandbox): allow Docker sandboxes to disable networking
- CLI: add codexbar cost --group-by session for per-conversation Codex session cost breakdowns in text output. Thanks @Yuxin-Qiao!
- Keychain: keep background browser imports non-interactive, honor the global access gate, retire the obsolete prompt-capable startup migration, and retry unreadable unified migrations without…
- OpenCode Go: label local-only quota windows as estimates in the menu and CLI when Auto has no server-confirmed usage. Thanks @Newarr!
- Fix: graphify install <platform> now advances the .graphify_version stamp only for the platform it actually (re)writes, instead of stamping every installed platform as current.
- Fix: an incremental rebuild no longer collapses the whole graph when the .graphify_root marker records a subfolder while stored source_file paths are relative to the repo root.
- Fix: a Go file that declares both an exported and an unexported symbol differing only by case (e.g.
- Signed Caveman runtime companions for macOS, Linux, and Windows.
- For the installer path, run CodeWhaleSetup.exe; it installs
- Extract the archive for your machine: codewhale-windows-x64.zip or
- Run install.bat (copies to %USERPROFILE%\bin)
- Update Cerebras model IDs and default
- Recover Anthropic tool arguments when the model serializes its call as text
- docs: replace broken profile sync link
- update:Windows 首次升级崩 pnpm 临时目录竞态时自动重试;升级后校验版本,半更新态给出明确修复指引(#225)
- skills:技能命令改走内核 /name 手势路径——/pptx 做演示 的参数不再被吞,与 web 端同一架构;无内核边界的组合保留直接注入兜底
- spinner:工作行显示真实上传 token 数(↑ input+cache)(#230)
- Improve CDP hostname logging
- Upgrade Pillow dependency to 12.3.0
- Document optional macOS remote-debugging approval
- hermes update from an existing install
- Fresh install: curl -fsSL | bash
- Factory: automatic agent runs are now opt-in per Factory (autoRunEnabled defaults off on upgrade); rules proposing runs now park them as proposed decisions until approved.
- CostGuardProcessor renamed to TokenCostControl (deprecated aliases remain for now, but will be removed in a future major version).
- Fixed traces that start under an external parent span not appearing in Mastra Studio. Resumed agent and workflow runs keep their link to the suspended run's trace.
- macOS Apple Silicon: GenOffice-0.7.204-arm64.dmg
- macOS Intel: GenOffice-0.7.204.dmg
- Windows: GenOfficeSetup-v0.7.204.exe
- New platforms — Tencent Yuanbao Bot API; cloud_web self-hosted IM gateway; Google Chat; WPS Agentspace; Tuitui
- New agent — Reasonix HTTP serve API adapter
- Feishu hardening — topic workspace isolation; quoted file on-demand download; first thread mention bootstrap; cross-type image batch flush
- retry helm chart push on non-fast-forward (03adcaf)
- Cursor plugin: 7 auto-capture hooks + 17 skills + MCP, IDE and cursor-agent CLI
- Devin support: connect devin --with-hooks (6 native hooks) and a Devin plugin manifest for /agentmemory:<skill> commands
- DeepSeek Harness connector (connect dsh), with --with-hooks for full auto-capture
- Token usage dashboard. A new Usage panel in Settings summarizes tokens, sessions, projects, runs, and artifacts across Today, This week, Last 30 days, and All time — each period-scoped new metric…
- Composer message queue. Follow-up messages can be staged in a session-scoped queue while a conversation is running — edited, deleted, sent immediately, or reordered with pointer drag and keyboard…
- Branch from completed agent messages. Every completed agent message footer now carries Copy and Branch in new session actions, so continuing from a specific answer creates a new idle session with…
- fix(runtime): support reviewed scan approval defaults
- DeepSeek Harness Long-Term Memory Access: MemOS Local can now be installed to a specified DSH profile via install.sh --agent dsh --profile <name>.
- DSH Installation with Automatic pnpm Completion: When one-click installation detects a missing pnpm in the PATH, it will temporarily prepare and verify pnpm 11.7.0 using existing npm, avoiding…
- More Reliable Native Dependency Installation in DSH Linux: The installer skips unnecessary CUDA downloads of onnxruntime-node, only approves the necessary build scripts, and verifies or fixes…
- Python SDK release: copilotkit 0.1.95
- commit: 6aa907e67a3e9c5f32b59a68c38420ebffc58265
- source tag: release2026-08-16_10-32-40
- regions: cn global
No release notes published.
Aug 15, 2026
- Large-scale migration validation: new release-test CI that seeds a production-shaped two-dataset system and runs the full migration chain to verify fork-splitting and rekey operations.
- Ladybug (graph adapter) stability and speedups: set-based fork re-key, chunked queries, batched rekey/restore, and fixes that make graph migrations more reliable and faster.
- Community plugins and migration docs: new cognee-community skill documentation and a tutorial for migrating from Letta/MemGPT and Zep to Cognee.
- Feature: graphify hook install reads a committed .graphifyrc (viz_node_limit=<int>) and bakes the visualization node limit into the generated git hooks, so a project-wide limit is shared via…
- Fix: graphify install (Claude always-on) now writes the CLAUDE.md registration into $CLAUDE_CONFIG_DIR when that env var relocates the Claude profile, instead of always mutating the default…
- Fix: a JS/TS inline or nested function expression — including a generator function expression (function*(k){…}) — no longer fabricates an INFERRED indirect_call when one of its parameters/locals…
- macOS Apple Silicon: GenOffice-0.7.110-arm64.dmg
- macOS Intel: GenOffice-0.7.110.dmg
- Windows: GenOfficeSetup-v0.7.110.exe
- Run marks navigation rail. Long desktop conversations gain a compact rail of marks — one for each visible human-authored prompt — fixed at the conversation panel midpoint so it stays put while…
- Frame-scoped Host artifact queries. host.artifacts now filters by the exact latest-producer frameId instead of sessionId, returns a fixed camelCase JavaScript projection…
- Bounded notebook run-history payloads. New executable code is bounded at 1 MiB, run output at a shared 2 MiB text budget with per-image and per-run image budgets, all enforced at both the producer…
- Fixed incorrect Windows app icon display
- Fixed Windows shortcut configuration issues
- Updated Microsoft Store packaging manifest and icon assets
- Cursor: on macOS, prefer Cursor.app's read-only local session in Automatic mode, persist validated sessions securely, and surface account mismatches before falling back to browser cookies.
- Cost history: add a Tokens/Cost switch to daily status-menu charts, defaulting Codex to exact local token totals and marking incomplete local history as refreshing. Thanks @Carl723000!
- Menu bar layout: add a compact run-out forecast token that shows only the predicted duration. Thanks @gnattu!
- supermemory lite — self-hosted is licensed up to 10,000 documents (read the license at git.new/memory); the cap is enforced at the API and shown at boot.
- New local console — the welcome page gets a Memory tab: live counts, recent documents and memories, container tags, and an interactive memory graph.
- ~3s boot — the embedding model loads lazily in the background instead of blocking startup.
- Serve an AG2 Agent over ACP — Any AG2 Agent can now be exposed as an ACP agent, so ACP clients can drive it the same way AG2 drives them.
- Remote ACP agents over HTTP and WebSocket — Drive ACP agents across a network boundary rather than only as local subprocesses, and answer questions those agents ask back mid-task.
- ACP 0.12 — Upgraded to the current protocol version.
- fix(deps): dsh-working-activity ^0.2.4——类型注册根治未知事件连环炸
- Select max reasoning effort for CLI and bulk scans. See the CLI reference.
- Set a cost ceiling with --max-cost; bulk scans apply the limit separately to each repository attempt. See cost tracking and limits.
- feat(cli): support maximum reasoning effort
- feat(providers): add xAI (Grok) to built-in providers
- feat(providers): add glm-5.3 to the Z.AI Coding Plan provider
- feat(pages): add MacOS, Linux, and Windows install channels to hero section
- commit: 9c212821bd5c6c1c410a6b6c8eadee0ea75cb098
- source tag: release2026-08-15_11-41-26
- regions: cn global
- Recommended: roll the schema cursor back to v53 (one dolt sql command; the guide has copy-paste steps; works even for databases created by 1.2.1).
- Need bd this minute: BD_IGNORE_SCHEMA_SKEW=1 bd <command> is a verified-safe stopgap for this schema range.
- Upgrade every machine/clone to v1.2.2 before recovering — a leftover 1.2.1 binary will silently re-migrate the database.
- Allow a UIEventStream to be built without a run_input, and give AGUIEventStream its own thread_id/run_id
- Attribute failed FallbackModel spans to the failing model, not the fallback: wrapper
- Pass openai through Temporal workflow sandboxes
- Added agents.testing, agents.realtime.testing, and agents.voice.testing utilities for deterministic Agent, Sandbox, Realtime, and Voice workflow tests without provider requests.
- Updated OpenAI provider compatibility to openai>=3.0.0,<4, including HTTPX2-aware request, response, transport, and exception handling.
- Hardened RunState interruption snapshots, recursive agent-tool approvals, max-turn finalization, streaming cleanup, and sensitive-error redaction.
- ci(release): remove desktop smoke gate (1b3dbcaaea882eeea90359c1db02e306d2f4f50a)
- bb desktop 0.38.0
No release notes published.
Aug 14, 2026
- Updated bundled Claude CLI to version 2.1.233
- Added GitLab merge request URL support to the --worktree flag and the claude agents view (where MRs display as !N)
- Added an opt-in forward_user_identity apps gateway setting on Anthropic upstreams that sends the signed-in user's identity as headers, so a proxy behind the gateway can attribute spend per user
- Added opt-in memory cgroup support for Bash tool commands on Linux (CLAUDE_CODE_TOOL_MEMORY_LIMIT) so a runaway build can't stall the session
- 0.7.5 changelog
- [fix] Prevent Google sign-in from hanging
- [fix] Prevent muted audio from stalling exports
- docs(cli,agents): no em dashes in user-facing copy; fix stale theme-set restart note
- fix(auth): control arm gets no emails; fail-closed activation variant
- chore(desktop,trpc,db): delete v1 cloud secrets surface and dead v1 cloud routers
- Release v1.68.1
- Codex shell-backed hooks on Linux and macOS now use the shared event adapter. This fixes the invalid JSON failures reported for SessionStart and UserPromptSubmit in issue #220 and covers the same…
- The tracked npm package source and sync manifest now match the 20-script tarball already published for v3.10.0, so the release tag, source tree, and package provenance are aligned.
- Installation docs now distinguish direct npm vendoring from Pi's automatically wired route. The npm listing still includes the Pi extension.
No release notes published.
- OCaml support — .ml/.mli extraction via tree-sitter-ocaml (optional [ocaml] extra).
- Fix: a cross-file INFERRED uses edge now binds to the symbol whose body actually references the imported name (a module-level function is a valid source.
- Fix: a named function declaration nested inside another function — including inside an arrow-defined component (const Panel = () => { function handleClick(){} }) or an arrow callback (useEffect(…
- Own CSS subpixel quantization in Moli
- Fix synthetic italic rendering direction
- docs: improve Japanese README wording
- ci(cargo-deny): add dependency audit with cargo-deny
- feat(sdk/go): complete Go SDK with domain clients, auth, and hardening
- chore(deps): bump jdx/mise-action from 4.2.0 to 4.2.4
- Any web URL is now a valid content source — paste a remote URL as input and kane-cli fetches and ingests it automatically, no manual download step required.
- Log and evidence files now write correctly on all systems — UTF-8 encoding is pinned on every file write, so non-ASCII characters in logs and evidence no longer corrupt or fail silently.
- Clicks and scrolls hit the right element more often — target resolution now looks at parent element context when identifying text nodes, removing a fallback that could silently miss on the first…
- Ingest commands no longer hang on a named pipe. os.walk and glob list a FIFO as an ordinary filename and MemPalace decides what to read from the suffix, so a pipe called notes.md in a mined…
- Project re-mine no longer silently skips a partial or interrupted file. Four related gaps in process_file: (1) multi-batch upserts now stamp every drawer with chunk_total so file_already_mined can…
- Conversation mine completeness matches the project path. Convo drawers now stamp chunk_total; a mid-batch upsert failure deletes that source's partial drawers before re-raising.
- fix: recover background jobs after restart
- fix(multiplexer): wait for child session readiness before attach
- feat(tasks): add read-only result retrieval
- Chinese (Simplified and Traditional) translations. The full interface — settings panels, onboarding, workspace, composer, home, and common components — is now available in zh-Hans and zh-Hant…
- Configurable reviewer model. A new Reviewer model policy under Settings lets the reviewer follow the active model or pin a dedicated provider, model, and reasoning effort.
- Cross-drive folder browsing. Local folder grants now work across all mounted drives — drive letters on Windows, volumes on macOS, and mounts on Linux.
- No changes
- #2922 cd48995 Thanks @wbxl2000! - web: The timestamp under assistant replies now shows the message time instead of the work duration.
- #2922 cd48995 Thanks @wbxl2000! - web: Restyle the slash command and @ file mention menus: matched fragments are bold-highlighted in the slash menu, and long lists in both menus get a scroll fade…
- #2922 cd48995 Thanks @wbxl2000! - web: The background Bash panel now supports filtering by status, and clicking a task shows its command and output on the right.
- Right-click any conversation in the sidebar to mark it unread and come back to it later.
- When you switch an agent's mode mid-conversation and the agent can only apply it on the next turn, the pill now shows the switch as pending instead of instantly claiming it already changed.
- Team MCP capabilities work again for directly-installed CLIs.
- Opt-in transcript telemetry now supports privacy-preserving collection with automatic secret redaction
- Anthropic-compatible provider profiles can now connect to additional compatible services
- Grok Build login can now be completed directly inside the TUI
- Agent Teams workbench: visualize multi-agent teams in the desktop app — members, tasks, a communication feed, and a dependency-lane canvas showing what each member is doing in real time.
- Dynamic Workflow orchestration: the model writes and runs orchestration scripts on the fly, driving subagents concurrently or in pipelines.
- Grok Official now defaults to grok-4.6, with xhigh reasoning effort.
- Fullscreen transcript search — Search and navigate matches in fullscreen mode. See TUI Fullscreen Viewport.
- Configurable default tools — Choose startup built-in tools globally or per project. See Tools.
- Configurable fullscreen exit output — Print the transcript or only a resume hint on exit. See Interactive Mode.
- commit: bca172cc354703d05cf2c8b490fc93d0d3e674d1
- source tag: release2026-08-14_17-46-49
- regions: cn global
- 698013cedb87abc6a18d9595e020d6bda7cd6ff1 MUL-5771 feat(daemon): allow configuring workspaces root
- cc8e020f80a733f8b61f6824ba993136519dd6e7 MUL-5852: add workspace Billing settings UI
- 2d711afdd51e95b8e61c715fd6ce5988417493c8 MUL-5852: 新增 Workspace Billing summary/prices proxy 与 Stripe 回跳落点
- Antigravity: hide untouched model families from compact menu cards and widgets while keeping unknown-usage families visible and preserving every family in provider details and all-untouched states.
- Menu bar layout: show the weekly pace reserve token after 1% of the quota window has elapsed, even when learned history still predicts less than 1% expected usage. Thanks @Yuxin-Qiao!
- Amp: support monthly Megawatt and Gigawatt subscription renewals with separate Other and Orb usage pools, and reset the daily free tier at 8:00 PM New York time. Thanks @3kh0!
- Godot 4.2+
- Active Ziva account
- docs: 0.9.0 changelog + blog posts
- fix(coding-agents): seed the actual harness, not the "opencode" default
- fix(docker): repair PGroonga Compose image
- docs: refresh README and add bilingual guides
- docs: update README and add friend links
- Fixed the full access warning dialog not closing before opening details, improving UI flow
- Tutti mode now requires manual activation instead of activating automatically
- Fixed MSIX Store packaging issue to ensure smooth updates through the Microsoft Store channel
- Release/memos local plugin v2.0.12 beta.1
- ci: add MemOS release publish dry run
- fix(plugin): improve recall relevance and host input handling
- feat(resume): add trusted resume validation and transition lineage
- feat(llm): task-scoped session affinity for prompt caching
- feat(action): render category/severity badge as a shields.io image
- GHSA-q2xc-rrxj-58x9: the local dev web chat UI (Agent.to_web(), clai web) didn't validate the Host header, so DNS rebinding from a website you visit could reach it and run the served agent with…
- Added support for GEMINI_API_KEY, so the CLI can run against the Gemini API directly without signing in.
- Improved /codesearch resilience by falling back to a local search when the ripgrep binary cannot be executed, such as when an endpoint security agent blocks it.
- Improved artifact list scrolling by caching each file's preview line count, which previously re-read every listed file from disk on every keypress.
- #619 250296c Thanks @DecentralizedJM! - feat: add built-in custom-mcp connector for arbitrary mcp sources
- #603 20f88c9 Thanks @akyourowngames! - fix: gate connector tools to personal/local-wiki runs
- #266 3dcb382 Thanks @ousamabenyounes! - feat: allow openai-compatible provider to opt into responses api
- agent-framework-core, agent-framework-mistral: Add a Mistral chat client with native chat, streaming, tools, structured output, and embeddings support
- agent-framework-core: Add experimental AGENT-HOOKS-0.1 enforcement middleware behind the opt-in agent-hooks extra
- agent-framework-openai: Add request preparation and response parsing hooks to OpenAIChatCompletionClient
- Introduce execution context management with UUID support
- Record what kind of exception ended a flow
- Record when a trace batch is shared with AMP
Aug 13, 2026
- Updated bundled Claude CLI to version 2.1.232
- The runtime container image now installs curl, so in-container HTTP
- Reborn PR test planning and the root-partition runner now share one test
- Windows first-start filesystem publication now uses native atomic rename
- Subagent forking is now on by default: a subagent_type: "fork" subagent inherits the full conversation and prompt cache, and non-teammate agent spawns in interactive sessions now run in the…
- Type @ in the prompt to mention another Claude session by name; Claude then uses SendMessage to reach that session directly
- SendMessage now delivers to a bare name that exactly matches one live session, instead of asking to confirm with a ref first
- Models declare their own capabilities: an agent pairs an output schema with tools when the model actually supports it, instead of ADK inferring support from the model id. (2aff82c, dc5dbfa)
- Tools can return media: a tool's function response carries images and other media straight back to the model, on Gemini and on Anthropic, LiteLLM, Apigee and OCI alike. (df9d6de, 703cf43)
- Conversation history keeps what Gemini needs: thought signatures, server-side tool call parts, and every result of a parallel function call survive the round trip. (d4ed347, e908137, 93dff41)
- feat: increases the username character limit to 64
- Fixes typos and mistakes in Catalan translation
- Update default fallback models for OpenAI/Anthropic
- Stamped plugins update in place through ncl groups create --template <ref>. When a group already carries the template's plugin, the same command becomes an in-place update instead of minting a…
- [BREAKING] Agent templates are now Agent Plugins 1.0.0 directories. plugin.json replaces context/instructions.md as the required file; MCP servers move to a spec-shaped mcp.json.
- Plugin MCP servers may declare a working directory. cwd in mcp.json (fixed forms ./p, ${PLUGIN_ROOT}[/p], ${PLUGIN_DATA}[/p]) now launches the server in that directory instead of being skipped…
- Existing installs: hermes update
- Fresh install: see the installer one-liner in the README
- chore(desktop): bump version to 1.20.2
- fix(deps): sync bun.lock with the nanoid 6.0.0 bump from
- fix(desktop): ship chat-runtime migrations + Claude SDK binary path to host-service
- Removed paseo chat and paseo loop
- Added host-wide agent profiles for reusable provider, model, mode, thinking, and feature settings
- Added workspace file search to Cmd/Ctrl+P
No release notes published.
- feat(deep-scan): add configurable graceful time limits
- fix(deep-scan): settle completed workers without cancellation races
- fix(scan): emit POSIX separators in Windows inventory
- Terminal picker: offer Ghostty 1.3+ as a default terminal for Open Terminal and provider login commands, falling back to Terminal.app when unavailable. Thanks @darwinz!
- Currency: Czech koruna (CZK) in the preferred-currency picker. Thanks @vAhyThe!
- Codex: confirm a legitimate server-side early weekly reset when the account has zero available reset credits, instead of preserving a stale weekly percentage forever.
- support LDAP search by uidNumber and gidNumber (7d2ad5f)
- fix(gator): separate review budget from approval gate
- perf(supervisor-network): avoid reparsing native policy input
- fix(dev): separate Podman Machine loopback listeners
- Fixed an issue where desktop windows might not close properly, improving overall window management stability
- Fixed idle Agent ACP processes becoming unrecoverable, reducing hangs and unresponsive behavior
- Internal beta validation release with Agent daemon stability improvements and desktop window management fixes
- JS/TS for...of / for...in loop bindings are shadowed, so they no longer fabricate indirect_call edges (#2685, @ousamabenyounes) — completes the loop/closure/catch shadow family (#2568/#2569/#2517).
- Python relative subpackage imports (from ..pkg.sub import x) resolve to the package __init__ (#2688, @ousamabenyounes).
- A non-regular file (FIFO/device) in the tree no longer hangs extraction; non-regular files are skipped (#2463, @itskaism).
- Forking — branching a conversation from any earlier message and keeping both lineages — now works in Aion CLI conversations, not just claude and codex.
- "Check for update" could offer an older version as if it were an update (e.g. 2.1.54 → 2.1.53) when the release channel was rolled back. It now only offers strictly newer versions.
- Aion CLI gains session forking with turn-anchored boundaries — the backend half of the feature above.
- fix(provider-registry): resolve prefixed size-colliding model ids to their own override
- perf(main-runtime): reduce cold idle memory with lazy loading
- fix(pins): synchronize pin changes across windows
- macOS Apple Silicon: GenOffice-0.6.389-arm64.dmg
- macOS Intel: GenOffice-0.6.389.dmg
- Windows: GenOfficeSetup-v0.6.389.exe
- Open apps in movable, resizable windows with a launcher, taskbar, notifications, and saved layouts
- Installed and marketplace apps now share one catalog across the App Center and QwenPaw OS Shell
- Browse, preview, edit, compare, upload, and download workspace files without leaving Chat
- StudioRunnerTools: New identity-aware dispatch toolkit (agno.tools.studio_runner.StudioRunnerTools) that splits execution out of StudioTools.
- list_components name filter: Added a name filter to list_components.
- A2A: Fixed the A2A stream client dropping Task-level metadata by breaking on status-update.
- fix(xlsx): SUBTOTAL ignores AutoFilter-hidden rows (1-11) and all hidden rows (101-111)
- fix(xlsx): comparison treats an empty cell as Excel's empty value, not text
- fix(xlsx): arithmetic coerces date/time-formatted text to its serial like Excel
- Global proxy settings. A new Settings → Network → Proxy setting lets researchers choose System (OS/PAC resolution), Manual (fixed proxy for Electron and child processes), or Direct (no proxy) modes.
- Prominent update reminders. Available app updates are now surfaced as compact actions on both the Home project grid and the workspace session sidebar, with two lines of context revealed on hover…
- camelCase delegation APIs. The public Host JavaScript delegation interface is renamed from snake_case to camelCase — stopChild, sendFrameMessage, messageReceipt, resolveMessage, and submitOutput —…
- 254a02a79532fd554ca520b36737bc8fa2f74f98 MUL-5799: feat(daemon): allow relocating per-task temp dirs via MULTICA_AGENT_TEMP_BASE
- 587fbc5559adf950dd593169a271f9f9948d302f MUL-5852: 接入 Workspace Stripe 订阅代理
- 8060aa7277d9c98bdc06c36176f6408bf37e6935 MUL-5947: feat(dingtalk): route groups to different agents
- Fixed the aspect ratio selector in the prompt box not updating its icon and label after choosing a new ratio in image mode. The selected ratio was already applied correctly to generations.
- Full Changelog
- For the installer path, run CodeWhaleSetup.exe; it installs
- Extract the archive for your machine: codewhale-windows-x64.zip or
- Run install.bat (copies to %USERPROFILE%\bin)
- Private-network sessions fail closed at startup. Explicit share_net: false sessions now require execd to bind-pin the workload NetNS/UserNS and return retryable 503 on failure.
- Background runs in isolated sessions. run accepts background: true and returns a JSON handle; new runs/{runId} status and .../logs endpoints support polling. In-flight runs block idle GC.
- Read-only PTY viewers. Opt-in ?mode=viewer lets multiple clients watch a running PTY session alongside the exclusive holder, with replay + live output and backpressure isolation.
- #2830 ec84a6f Thanks @liruifengv! - Add an experimental fullscreen TUI mode. Set the KIMI_CODE_TUI_FULL_SCREEN=1 environment variable to enable it.
- #2700 c9bfe8b Thanks @7Sageer! - Add a configurable model pool for spawned subagents behind the secondary-model experiment (KIMI_CODE_EXPERIMENTAL_SECONDARY_MODEL=1, or the master flag): with the…
- #2830 ec84a6f Thanks @liruifengv! - Render LaTeX math formulas ($…$ / $$…$$) in messages as Unicode formulas.
- Support FastMCP 4 and MCP SDK v2 in MCPToolset alongside FastMCP 3
- Add Azure AI Voice Live via the azure_voice_live setting
- Fix silent truncation of gzip response bodies
- commit: 240247e52e938710f409c921a7a393097dcef52b
- source tag: release2026-08-13_11-29-55
- regions: cn global
- fix: emit SSE ping before message_start on Bedrock streaming path (issue
- fix(litellm): add async_post_call_success_hook to HeadroomCallback
- deps: bump brace-expansion from 5.0.7 to 5.0.9 in /docs
- A change with a diff is an expandable row that opens the hunks inline.
- A paths-only change renders as a plain row labelled diff unavailable
- A started edit stays hidden until the tool result confirms it, so a denied
- feat(conversations): add client-side conversation archive
- feat(chat): add inline markdown artifact previews
- feat: add ready-for-dev issue readiness gate with type-specific criteria
- Select the Kimi system prompt correctly for official Moonshot and Kimi providers
- Fix xhigh reasoning effort for xai models
Aug 12, 2026
- perf(desktop): persist channel snapshot hash (c86443c5997c96c42829ce200e73e6e6efe52d96)
- fix(agent): raise output limit and allow 3 recoveries (72d56e7bd3a94fa3ee20b5a50bd1b868a9329d9c)
- fix(desktop): defer foreground resume work (59f613c404958d8ac99525b4aaaf26843257de31)
- ConversationResetMessage for conversation reset events: The SDK now surfaces the CLI's conversation_reset frame as a ConversationResetMessage dataclass, letting applications detect when a /clear…
- Message origin on UserMessage and ResultMessage: New origin: MessageOrigin | None field on UserMessage and ResultMessage surfaces why a turn was initiated — distinguishing application-submitted…
- resume_session_at / resume_drops_turn options for truncating resume: ClaudeAgentOptions now supports resume_session_at (fork a session at an earlier transcript entry) and resume_drops_turn…
- feat(model-routing): add ModelRouter and accept it via Agent(model=) (26770e17c)
- feat(catalog): add strands-github-storage integration (7f19592f9)
- feat(storage-py): add top level storage (0b296c37d)
- Made session compaction keep complete recent turns and produce clearer summaries for smaller models.
- Added MERGE Gateway reasoning variants so those model options work correctly. (@MatthewFeroz)
- Capped automatic session retries and added jitter to reduce repeated retry storms.
- 804324fb80a1133f678b0d8c3d74f32c16988c12: feat(session): print session id on exit (@BrunoKrugel)
- b82d7d05d8883e3f05c9cb86a6f47bfa3b2ec59b: feat(session): restore last used provider/model from session (@meowgorithm)
- 90b48cda0991c37d3202327e20680f262c798c24: feat(ux): add bash syntax highlighting to bash tool (@andreynering)
- Remote sessions now preserve active skills and recover cleanly after clearing session state
- OpenAI-compatible and OpenRouter streams handle transient failures and fallback tool calls more reliably
- ACP sessions tolerate host-provided MCP servers and report turn token usage
- fix(cli): stop optional peer deps blocking a piece publish @OdaiAhmed99, @AhmadTash
- fix(deps): bump nanoid, nodemailer, undici to clear Dependabot alerts @KhaledR57
- refactor(brain): flatten the wiki into brain/ and give decisions an index @abuaboud
- CLI: load bundled provider-plugin resources when codexbar is installed as a symlink to the packaged helper or standalone executable. Thanks @djbclark for the report!
- Menu bar layout: restore dragging placed chips to reorder or remove them while preserving click and keyboard selection. Thanks @ikkira!
- Plan history: avoid atomically replacing byte-identical provider history files, reducing unnecessary disk writes. Thanks @guocity for the report!
- Unrolled agent loop
- Cache-safe request assembly with append-only turn context and declared cache semantics
- Streaming shell output while commands run
- Added support for dragging files from the Project Panel to external macOS apps. (#58161; thanks notJoon)
- Added the multiple_selections key context for assigning different keybindings when multiple selections are active. (#60018; thanks AidanV)
- Agent: Fixed the first line of terminal output sometimes being hidden in the Agent Panel. (#61939; thanks timvermeulen)
- Stored workflows renamed to dynamic workflows: addStoredWorkflow(s) → addDynamicWorkflow(s), StoredWorkflow* types → DynamicWorkflow*, and workflow.origin now reports 'dynamic'.
- Channel agent replies now post as markdown by default; set textFormat: 'plain' on channel adapter configs (e.g. Slack) to preserve literal plain-text posting.
- Add resolveSession and onStaleToolApproval to agent controller channels.
- feat(build): add glibc-static supervisor libc variant
- fix(sbom): detect sha256 hashes in expression-form licenses in needs_fix
- feat(gateway,cli): windows compilation support
- macOS Apple Silicon: GenOffice-0.6.279-arm64.dmg
- macOS Intel: GenOffice-0.6.279.dmg
- Windows: GenOfficeSetup-v0.6.279.exe
- Discord joins Slack as a built-in channel: enter a Bot Token in Settings → Channels, test and enable, then approve pairing requests and manage authorized users.
- When you select a link in chat or preview, you can now choose to open it in the built-in browser or your system browser.
- Mermaid diagrams in the preview panel can be zoomed, panned, fullscreened, copied and downloaded, instead of being a fixed static image.
- New Agent Editor UI with Easy and Advanced modes for editing agent identity, prompts, model presets, tool policies, skill visibility, and avatars — all without destructive overwrites or model calls.
- Project-scoped profiles: switch between Global and project scope when managing agents; sparse reads, writes, resets, and policy configs stay bound to the selected layer.
- Profile lifecycle management: per-scope availability, duplication, active-profile status, inline project selection, scoped restore for inherited profiles, and synchronized profile switching across…
- Adds a persisted screen selector so recordings can target a specific display.
- Resolves displays through stable identity when Electron reassigns source IDs, shows monitor model labels, and falls back safely when a selected display is disconnected.
- Keeps the recorder controls at a stable width while their height is fitted automatically.
- Fix: graphify update refuses to overwrite the graph with a shrunken one when the shrink was caused by an extractor failure this run, instead of silently replacing good data
- Fix: a JS/TS catch binding passed as a call argument no longer fabricates an indirect_call edge to an unrelated same-named callable (thanks @imagineers-tyler).
- Fix: Cargo.toml is recognized as a package manifest
- Hermes 集成稳定性增强:优化 provider 启动等待、链接管理和桥接版本读取,提升本地插件与 Hermes 的兼容性。
- Production subagent delegation. The agent can now delegate structured work to a sub-agent within the same project.
- Immutable resource identities. Connectors now have an immutable lowercase invocation name separate from an editable display name, so renaming a connector regenerates its generated skill prose…
- camelCase Host JavaScript APIs. The public Host JavaScript surface — host.agents, host.compute, and related methods and caller-supplied option keys — now uses standard camelCase.
- Test against emulators and simulators — device setup, persistent runner sessions, and evidence capture (screenshots, logs) all work end-to-end for mobile targets, mirroring what web automation…
- The AI understands mobile screens — perception, accessibility tree inspection, vision coordinates, and code generation are all platform-aware.
- Replay works on mobile too — recorded .testmd flows can be replayed against an emulator or simulator; the frontmatter now carries target vocabulary so the file knows what platform it belongs to.
- 90eadd940a47b55d654a1496b04520ef69a1188c MUL-5905: feat(wecom): send the files an agent produced into the chat
- 0f8c300dacfa548176ab4624e259865ea4ac7534 MUL-5915: fix(slack): neutralize links in member issue titles
- 8d292ef7ceaf3461d85a4ba28e99f96a80c8c848 MUL-5963: feat(server): let web chat read back history via multica chat history
- chore: land PRs #139/#148/#149 with original authorship
- Revert Windows private-file ACL restriction (broke main)
- docs: add CONTRIBUTORS.md
- For the installer path, run CodeWhaleSetup.exe; it installs
- Extract the archive for your machine: codewhale-windows-x64.zip or
- Run install.bat (copies to %USERPROFILE%\bin)
- Isolated context: the intermediate work stays out of the main conversation, saving tokens and keeping the model focused.
- Parallel execution: several sub agents can run at once, and the main Agent gathers their results when they finish. Both the Web console and the desktop client show the full run.
- Customizable: two built-in types, general-purpose and explore, plus your own — drop a .md file under the workspace subagents/ directory to define a type with its own system prompt and tool set.
- kernel artifacts build
- commit: ac727519a27cd38c14e827c4eef3ffa6708a43f7
- source tag: release2026-08-12_14-39-17
- regions: cn global
- feat(pages): redesign hero with two-column layout and three install channels
- feat(providers): add siliconflow (GLOBAL) to built-in providers
- feat(providers): add siliconflow-cn as a built-in model provider
- fix(core): preserve target invalidation through memoization
- fix(lancedb): apply column schema evolution under full reprocess
- chore(deps): upgrade dependencies, clearing security alerts
- 0.7.4 changelog
- [agent] Add in-app skill management
- [agent] Refresh system prompt for current editing capabilities
No release notes published.
- #2816 ad12ad8 Thanks @liruifengv! - Show the live work progress of background subagents in the /tasks panel.
- #2840 68ce3c7 Thanks @wbxl2000! - web: Image and video tool results now open in a fullscreen preview on click, with zoom support for images.
- #2731 437a1b8 Thanks @pvzheroes125! - Detect MCP servers that require OAuth without needing auth: "oauth" in the config.
- GHSA-h4xc-3qfq-jf93 (high) — the development web chat UI's (Agent.to_web(), clai web) chat endpoint didn't check the request's content type, so a plain cross-origin request from a website open in…
- GHSA-3gh4-cghq-f8v4 (low) — retry-prompt content wasn't redacted by InstrumentationSettings(include_content=False) when the retry wasn't tied to a tool call.
- Backport #7357 to v1
- feat(cli): inspect saved scan and worker logs
- feat(sdk): attribute Responses traffic to SDK and CLI
- feat(scan): show current findings for each repository
- bb desktop 0.37.0
- core-py: declare the engine version the bindings build against (92cdf43)
- Report flow outcome, duration, and human-in-the-loop signals.
- Emit FlowStartedEvent when a boundary hook aborts the flow.
- Scope span export to our own tracer provider.
Aug 11, 2026
- fix(files): honor AWS fallbacks in Write File
- fix(api): retry flow updates on sqlite locks
- fix(frontend): close protocol gaps in markdown sanitize schema (GHSA-7rw4-v4gc-r864 audit)
- 🔌 Claude sessions routed through the Databricks AI gateway now connect cleanly — Omnigent sends the x-databricks-use-coding-agent-mode header the gateway requires
- Updated bundled Claude CLI to version 2.1.228
- Channel status reports transport and provider legs separately (@copilotkit/runtime) — status().overall === "online" proved only that the runtime reached the Gateway with a valid project key, so a…
- The in-memory agent runner is bounded — no more OOM — InMemoryAgentRunner (the default runner when the runtime has no Intelligence backend) kept every thread, every run, and a per-run…
- AGUISendStateDelta initializes arrays before /- append — a state delta appending to an array that no prior state event had initialized produced a patch that failed event compaction with…
- Changes since langchain==1.3.14
- release(langchain): 1.3.15 feat(langchain): expose trace_policy on AgentMiddleware chore(langchain): fix type errors in tests chore: bump h2 from 4.3.0 to 4.4.1 in /libs/langchain_v1…
- Vector Stores: Add an Oracle AI Vector Search vector store (oracledb) to the OSS SDK, with pooled connections, HNSW/IVF indexes, an optional indexAccuracy target, JSON payload filtering, and six…
- Core: Percent-escape %, &, and = in user_id, agent_id, and run_id when building the session scope key for the recent-conversation buffer, so the key stays unambiguous for ids containing those…
- Vector Stores: Close and clear the Oracle pool that Mem0 created when initialize() fails, and reset the cached init promise so the next call retries instead of replaying the rejection forever.
- Ingest a Confluence page directly by URL — ingest <confluence-url> pulls page content and images into kane-cli.
- Jira ingests now carry comments — comments are fetched with full cursor pagination and appended to the canonical envelope.
- Better error surfaces on the shared transport — when the transport hits a 429, it retries with a bounded backoff. Wire errors map to clean codes and include the Integrations setup URL when relevant.
- Caveman 2 — the engine ships. v1 shrank the model's mouth (output). v2 adds the ears: a real compression engine that squeezes what the model reads — tool outputs, logs, code, JSON — before it ever…
- The skill is untouched and still MIT. Nothing about your /caveman install changed.
- Real run: a 40-record JSON went 16,098 → 1,091 tokens (93% smaller), byte-safe, fully recoverable via CCR. Savings are inferred — measured on your machine, not projected, not "verified."
- Local folder access with @path mentions. Researchers can now type @path/to/folder in the composer to grant the agent read access to an entire local directory, without attaching each file…
- Connector runtime status and recovery. Connector sign-in state, connection health, and recovery flows are now surfaced in the workspace so a broken or expired connector session is immediately…
- Bounded REPL inference. The agent can now evaluate short JavaScript expressions in a bounded runtime to test hypotheses, transform data, or verify logic inline — without spinning up a full…
- npm: bash npm install -g base44@latest
- Homebrew: bash brew install base44/tap/base44
- Version: 0.1.9 NPM Tag: latest
- fix(sandbox): acknowledge unchanged policy revisions
- fix(gator): allow same-sha state nudges
- feat(proxy): bind static credentials to provider endpoints
- disable persisted-entity actions in user, application, invitation and payment add pages (7325999)
- Fix: the 0.9.37 partial-parse warning no longer fires on valid TypeScript/TSX
- Fix: file_hash()'s stat fastpath no longer serves a stale digest when a file is rewritten to the same size within one mtime tick
- Fix: stored-path absoluteness is detected cross-platform, so a POSIX-absolute source_file from a Linux/CI-built graph no longer leaks into node ids on Windows
- release(langgraph): 1.2.11
- feat(langgraph): expose trace_policy on add_node
- chore(deps): bump the minor-and-patch group across 1 directory with 7 updates
- Inline ghost-text completion for mid-prompt skills
- /clear and /new accept an optional first prompt and show the resume id
- Compacted conversation history is preserved
- Agent logstream coordination (RFC 003). Append-only event layer for multi-agent work: durable task packets, wait/ack handoffs, patch and file artifacts, and live tailing over the MCP HTTP hub…
- Logstream multi-master sync foundation (RFC 004 step 0). Estate-level logstream replication hooks so coordinated agents can share the coordination layer across palace replicas — the storage step…
- OpenAI-compatible embeddings. Opt-in embedding_model: "openai-compat" talks to any /v1/embeddings endpoint (LM Studio, llama.cpp, vLLM, Ollama's OpenAI shim, or a self-hosted server) over stdlib…
- Failure counters for DNS resolution and nftables updates. New egress.dns.query.failed_total{reason} (no_upstreams / upstream_error / empty_response / rcode) and…
- mitmdump output now lands in the egress log sink. mitmproxy and credential-proxy logs printed via ctx.log are piped line-by-line into the egress zap logger, so they reach the log file and rotation…
- mitmproxy 11.0.2 restores HTTP/2 throughput. The upstream HTTP/2 flow-control fix (mitmproxy/mitmproxy#7317) removes severe throttling of large HTTP/2 responses: a 216 MiB response went from 0.61…
- Desktop sessions now include a restored Super-key overview with clickable session navigation
- The desktop overview shortcut works safely across compositors
- Transient stream_read_error failures are now recognized as retryable transport errors
- Public preview.
- Full Changelog
- 00e12d34030156b0927a520a3df4253d196129f6 MUL-5906 feat(wecom): read the photos, files and videos people send the bot
- f26b97e395f5337766f30144c252504a5dfd015a MUL-5932: fix(daemon): give Hermes agents memory that survives the task
- eb3cd4fe9e720c5efbf863bbb39e63647ef6a8c1 MUL-5951: fix(issues): reject a cross-workspace project on issue update
- CLI app launch fix: the dapi bin wrapper exports ELECTRON_RUN_AS_NODE to run the CLI through the app's Electron binary.
- Security Vulnerability: Fixed flaws in permission verification to block a high-risk privilege escalation vulnerability where anonymous application session tokens could be exploited to obtain the…
- Knowledge Base: Fixed garbled paragraph content when uploading PDF files where the title text is identical to the body content.
- feat(viewer): add review comment tag filters
- feat(llm): add dynamic cache breakpoint on last message for Anthropic
- fix(codeql): Workflow does not contain permissions
- test(ego-browser): migrate uncovered regression cases into real-browser e2e
- Clarify task space tab closure
- Sync: collapse duplicate usage rows for the same provider account across Macs, with the current Mac's configured result taking precedence over synced copies.
- Menu cards: show session quota as exhausted until the final reset of every binding weekly/monthly plan lane, while preserving session-specific details and independent purchased credits.
- Menu cards: keep primary usage values and long localized reset timestamps readable by stacking the row only when both cannot fit. Thanks @Yuxin-Qiao!
- #616 7531d61 Thanks @colifran! - fix: pin patched js-yaml and undici via pnpm overrides
- #513 adc03d6 Thanks @colifran! - chore: reorganize repo code to make into domain specific directories and improve test coverage to prevent regressions
- #610 c74ae1e Thanks @Tomaskobel! - fix: preserve exec bit on dist/cli.js after build
- commit: 06e6197164997d7425e6ad3daabc20d3f0451fda
- source tag: release2026-08-11_11-00-36
- regions: cn global
- The default model is now gpt-5.6-luna; explicit models, run-level overrides, and OPENAI_DEFAULT_MODEL continue to take precedence.
- Local MCP connections support MCP Python SDK v1 and v2 across stdio, SSE, and Streamable HTTP transports.
- Applications using custom MCP HTTP authentication or client factories must use the HTTP types owned by the installed MCP major version, or pin mcp<2.
- Restore tool spans for failed argument validation
- Fix XaiStreamedResponse finish_reason mapping for streaming responses
- Point offline web UI hosting at the self-contained chat UI build
- core: keep RLAC cycle-detection state per analyzer invocation (d754985)
- Added a heading outline to the artifact viewer, opened with t, so a long markdown document's structure is visible at a glance and you can jump straight to a section instead of scrolling.
- Added non-interactive answers for more read-only slash commands in print mode, so -p "/permissions", /hooks, /help, /changelog and /config each emit one tab-separated record per line — or a…
- Added machine-readable output to the models and agents subcommands through an --output-format flag accepting json and stream-json, and moved their error messages and progress spinner off stdout so…
- Added persistent session-to-tab binding for shared Chrome sessions. Named sessions connected through --cdp or --auto-connect now remember their CDP target across commands and daemon restarts, and…
- Added a Remote Agent Browser provider guide covering Vercel authentication, disposable and stable sessions, custom VCR images, snapshots, and live-view URLs
- Fixed parallel sessions sharing one Chrome hijacking each other's tabs. Event-discovered targets no longer steal a pinned session's active tab, re-attach restores the persisted target instead of…
- fix: run post-scan prompts after failed scans
- fix(deps): patch high-severity js-yaml advisory
- fix: remove authentication output and instruction limits
Aug 10, 2026
- Channels recover from error-only websocket failures (@copilotkit/channels-intelligence) — Node 22's built-in WebSocket can report a failed HTTP upgrade (including a transient 502) as an error…
- ~56 MB of unreferenced git-LFS assets removed — four files under assets/ that nothing in the tree referenced (demo.gif, animated-banner.gif, travel-planner-gif.gif, proejct-perplexity-clone.png)…
- Runtime's Channels dependencies resolve to 0.8.1 — the published @copilotkit/runtime picks up @copilotkit/channels-core@^0.8.1 / @copilotkit/channels-intelligence@0.8.1 from the concurrent…
- Updated bundled Claude CLI to version 2.1.227
- Polish desktop onboarding flow (3f2f32641f4093d087fd9506bfac1fa0329e8b2e)
- fix(desktop): quiesce renderer polling while hidden (07a3c768d619db31fee3f0590f9433cdd1213e8f)
- fix(channels): restore member invitations to private channels (2777189d960fa5b1d863166f36d6e37ff8ce0819)
- CodeBurn Desktop 0.9.20 — see CHANGELOG 0.9.20 for the full release notes. Highlights: panel fetches drop from seconds to milliseconds via the resident serve process, Spend punchcard, Pull…
- 0.7.3 changelog
- [ui] Make timeline track header resizable
- [ui] Show not-charged notice for refunded failures
- Fix: affected now traverses a dynamic import('…') made inside a function or at module scope
- Fix: a Python member call on an untyped receiver (x.get(...)) no longer binds by name alone to a same-named module-level function, fabricating a false high-confidence calls edge and a god node
- Fix: fuzzy dedup no longer over-merges two distinct entities in the same file whose long labels differ by a content word
- Darwin Shield: a defensive zero-day harness, Darwin Mode applied to security response itself (ADR-155…167).
- Self-learning WASM-memory cost-optimal escalator: tiered compute escalation (E2/E3/E4 cores) with peer-reviewed mitigations (ADR-169).
- Non-stationary CFR/CFR+ poker solver: Darwin evolving against a genuinely adversarial, shifting-strategy opponent via Chebyshev schedules — a harder generalization test than static benchmarks.
- chore(desktop): bump version to 1.20.1
- fix(host-service,desktop): broken import flow, stop v1 importer from duplicating already-imported projects
- fix(desktop): route project/session selection through single draft-store writers
- fix(llm): omit parallel_tool_calls on tool-less requests
- fix(container): reclaim abandoned browser sessions
- fix(container): keep the browser env alive where image ENV is dropped
No release notes published.
- restrict cross-org webhooks and check org on add APIs (7f622b1)
- fix(e2e): separate Podman Machine loopback listeners
- fix(policy): prevent implicit authorization inheritance
- fix(gateway-interceptors): apply tls-native-roots for HTTPS interceptor endpoints
- ROB-856 Add Atlassian Rovo MCP integration documentation
- ROB-855 pulling skills with github app credentials
- Weekly Benchmark Results 2026-08-05_12-14
- commit: 30564c785d72549e44e9d07344bae97f07ce0c51
- source tag: release2026-08-10_21-47-16
- regions: cn global
- Slack moves out of "coming soon": configure a Bot Token and App-Level Token in Settings → Channels, test the connection, pick a model, and manage which Slack users are authorized to talk to your…
- AionUi used to ship its own pinned copies of claude and codex, so the same prompt could behave differently in AionUi than in your terminal — invisibly.
- The panel is restructured into collapsible sections with drag-to-resize dividers and double-click-to-reset, closely following the VS Code sidebar.
- macOS Apple Silicon: GenOffice-0.6.101-arm64.dmg
- macOS Intel: GenOffice-0.6.101.dmg
- Windows: GenOfficeSetup-v0.6.101.exe
- run_dcf — FCFF bridge, WACC build, mid-year discounting, net-debt bridge, WACC×g sensitivity grid. Dual terminal value: each method is cross-checked against the other's implied multiple and implied g.
- run_comps* — EV bridge, LTM + calendar-year calendarisation, multiple matrix. A peer with a non-positive denominator is excluded and reported*, never averaged in as a negative multiple.
- threestatement — linked projection with a hard balance assertion, an explicit revolver plug, and an iterated interest↔debt circularity that must converge or raise.
- Persistent side conversations. A lightweight side conversation can now be opened alongside a session's main thread — same project, same agent, independent history — so a researcher can ask a quick…
- Conversational skill creator. The agent can now draft, refine, and install a new skill from a natural-language description inside the session.
- Unified plan-approval composer. Plan approvals, permission requests, and ask-user questions are now consolidated into a single composer lane with consistent layout, status chrome, and keyboard…
- b15b902d3bc26e10ff9ada9218b3143aa6ba7d08 MUL-5842: fix(ui): drop the write-only sidebar_state cookie and align the settings nav trigger
- cd9f4147a57dc469328ad7bbd06f1781e0dc8314 MUL-5853 feat(agent): support Kimi thinking levels
- ad7e621139b56a1b2df6831fcd7de716c1177cbe MUL-5860: unify internal link click semantics across all surfaces
- feat(llm): support custom retry status codes via retry_codes config
- feat(llmloop): backfill comment thinking from turn output
- feat(pages): show live npm downloads in highlights stats
- None noted in this changelog.
- Added transient to non-state agent signals. Set transient: true when a processor should deliver a reminder to the current model call without retaining it in conversation history.
- Added tool definitions to MODEL_GENERATION span attributes. The tools made available to the model (name, description, and JSON-schema parameters) are now captured once per generation as the tools…
- feat(composer): add smooth wheel control for reasoning effort
- fix(sidebar): add sidebar management menu entry
- fix(agent-channels): strip citation markers from IM replies
- The desktop app now includes a project file explorer and an in-app help overlay
- Desktop sessions open as spatial panels with familiar new-tab shortcuts and animated navigation
- The desktop window can hot-reload app code without being recreated
- Ignore unknown top-level config fields instead of failing config parsing
- Register projects opened from Home so they are available to the rest of the app
- Open the project menu with a right-click in Home
- DeepInfra: show billing-cycle spend against a positive spending limit in the automatic menu-bar icon. Thanks @selfagency for the report!
- z.ai: restore pace for verified 5-hour, weekly, and MCP-monthly usage windows without treating rolling 30-day limits as calendar months. Thanks @kiranmagic7!
- Codex: publish refreshed core quota immediately while optional Credits and OpenAI Web enrichment continues, without unfreezing cards whose layout still needs reconciliation. Thanks @Yuxin-Qiao!
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