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

AI agent releases: Sep 7 – Sep 13, 2026

245 stable releases from 101 tracked agents and frameworks in the week of Sep 7, 2026, newest first, with the headline changes from each set of release notes. Our lists keep each project's latest releases, so a very active project's older releases can be missing.

Sep 13, 2026

  • New: genoffice command line + agent skill. Every installer ships a CLI that runs GenOffice's engines headless, so coding agents (Claude Code, Codex, OpenCode, …) can build and edit real Office…
  • Create, convert and read pptx / xlsx / docx / pdf / md / html / csv; render any file to one PNG per page so the agent can check its work.
  • Structured edits in place for Docs, Sheets and Slides: text, images, charts, tables, comments, header/footer, filters and rules.
  • An expired Cline sign-in now produces one actionable error instead of two dead ends.
  • The Account page now offers a sign-in prompt when your Cline refresh token is rejected, instead of an error card whose Retry failed the same way.
  • Signing out of Cline and Cline Pass now sticks. Sign-out removes the provider entry, but the runtime re-imports missing providers from the classic extension's stored credentials on every command…
  • fix(browser): load newly opened background tabs
  • Unify tab surface selection across workspace activation
  • fix(native-chat): keep worktree active during chat creation
  • 聊天支持 @ 提及工作区文件
  • STT 探测改为真实识别调用,与 TTS 对称
  • 语音探测失败直接展示错误,不再返回 500;录音上传前转码为 WAV
  • chore(desktop): bump version to 1.28.0
  • fix(deps): patch Oneleet dependency security alerts
  • fix(agent-setup): notify when Codex waits for user input
  • Cleaner setup, compact profiles and better sidebar controls.
  • More reliable Goal/Loop recovery, queued messages and conversation history.
  • Updated runtime, MCP and plugin dependencies.
  • Discover and install skills from a verified marketplace. Browse signed marketplace catalogs through the official CDN — category, author, score, and provenance detail on every card — install in one…
  • Take a whole session with you as a .science package. Export a session with its conversation branches, artifact and upload versions, Notebook records, verification receipts, and environment locks.
  • Extract figures and tables from PDFs locally. An optional figures-and-tables workflow reconstructs figure crops, captions, cross-page content, and merged headers in an isolated process, and…
  • Saved named plans can now be listed with sh scripts/set-active-plan.sh --list or PowerShell scripts/set-active-plan.ps1 -List.
  • Counts handle mixed status formats, duplicate markers, and the templates shipped in all six languages. Code examples and unrelated sections do not count as phases.
  • The helpers and usage instructions ship across the canonical skill, plugin scripts, and supported IDE bundles, including OpenCode, Mastracode, and Hermes.
  • The PageIndex SDK, local or cloud — vectorless, reasoning-based RAG, end to end.
  • Much faster indexing — the PageIndex Flash engine gets the tree from layout stats: no LLM involved for the structure generation itself, LLMs only write the node summaries, and tree expansion…
  • Flash engine: the local default (mode="standard" keeps the classic LLM pipeline).
  • Faster Codex activity refreshes: reuse unchanged local activity data instead of repeatedly validating and decoding it.
  • More reliable account cards: keep identity, usage, credits, and cached costs tied to the right account, and prefer fresh subscription dates.
  • Clearer provider quotas: distinguish unavailable z.ai limits from unused quotas and remove duplicate Antigravity rows without hiding distinct limits.
  • Custom target connector sink API changes
  • Container sinks now fulfill child target providers through per-action child slots (TargetActionSink.from_fn_with_children / from_async_fn_with_children).
  • Sinks whose callbacks compare equal but differ in type no longer share one sink identity, and callbacks that can't be weak-referenced (e.g. NamedTuple instances) now raise TypeError
  • Bundle Astrid 2026.9.2, adding optional authenticated MCP Streamable HTTP
  • Update the exact runtime and client-library pins while retaining GNU, musl,
  • fix(ci-fix): merge a behind base before fixing and recover from post-push conflicts (c3846df95) — Vincent Hus
  • feat(tool): enforce one action per model response with ToolRole (ecada259b) — Vincent Hus
  • refactor(skills): organize runtime code into dedicated packages (663c0dacc) — Vincent Hus

Sep 12, 2026

AI Memory
No release notes published.
Graphify
  • Fix: graphify.serve now imports cleanly on Python 3.12 and 3.13. The chinese extra pins jieba-py from 3.12 onward…
  • Fix: the git hook's rebuild-root guard now rejects a symlink-loop or dangling .graphify_root on Python 3.13, whose Path.resolve() no longer raises on a loop — the saved root must resolve to a real…
  • Pending-edit capacity: an equal-length overwrite of committed
  • Distributed SDK-edit workloads: workspace-distributed-sdk-edit-500
  • Packed metadata: pack rows 3,457 → 1,058, pack overflow slack
  • Fixed read-only git commands in Bash unexpectedly asking for permission after a session had been running for a while (regression in 2.1.269)
  • Connector discovery excludes raw HTTP connectors that do not support MCP tools.
  • A single invalid connector no longer prevents the rest of your connectors from loading.
  • Automatic session titles use the active model on custom Mistral endpoints.
  • bb desktop 0.43.1
  • For simplicity, ! in the composer is just prompt text now. It no longer starts a terminal session. Commands go through the agent and the usual approval flow.
  • Subagents can use their own model and reasoning effort. Use GPT-6, Astra, or Fable 5.1 to steer, then delegate implementation to Kimi K3.
  • You can send feedback to a running named subagent without interrupting its current tool. The final result still comes back to the parent conversation.
  • perf(grouping): return file indices instead of paths
  • chore: synchronize glm-5.3 to from z-ai coding plan to z ai api
  • fix(tui-sidebar): long model names wrapping in the TUI sidebar
  • fix: trigger model fallback on content-policy moderation rejections
  • fix: block repeated wait tool calls within a turn
  • Linux desktop and Omarchy: a new Qt desktop app brings usage, spending, settings, notifications, and an optional tray icon to Linux, with a theme-aware Omarchy bar widget.
  • Provider credits: show Codex workspace balances, DeepSeek daily spend, and optional Copilot credit allowances.
  • Linux Cursor authentication: reuse the signed-in Cursor app without manually copying cookies.
  • Retries after a failed turn now back off instead of hammering, and sessions keep their titles when the model errors.
  • Web fixes: drag targets stay put during streaming redraws, transcripts measure correctly after markdown renders, prompt expansion controls work inside message bubbles, and runtime-default prompts…
  • Verified human replies in threads can use admin routes again, and the admin's internal member override editor is back.
  • Improved the startup warning for deprecated unsandboxed permission rules across CLI, shared, and project configuration files to list each affected file path, up to five offending rules, and…
  • Improved /resume startup responsiveness when opening a large conversation history with a cold or stale summary cache.
  • Fixed MCP servers bundled inside plugins colliding with each other or with user-configured servers in mcp_config.json when they shared the same server name by automatically namespacing plugin MCP…
  • SmallestTTSService now uses Smallest AI's continuation API: text fragments within the same LLM turn share a context_id so the server joins them into one continuous generation instead of resetting…
  • Added LiveKitParams.audio_out_queue_size_ms to configure the outgoing rtc.AudioSource buffer size. Defaults to LiveKit's 1000 ms, so existing behaviour is unchanged.
  • Added enable_turn_detection to GradiumSTTService. With it on, Gradium's server-side end-pointing signal decides when user turns start and end instead of the pipeline's VAD, and the service…
  • efba40b7bf88ca143587cbdebb74cf1b463c316a: feat: OpenAI oAuth (@andrinoff)
  • 4a74b9e095792217e95867bf74c1aa86fef44ee2: fix(sessions): keep list position after deletion (@ifsin)
  • 33be610debb1eaf33bac40e0b089e1aa84d05602: fix: @meowgorithm review (@andrinoff)
  • fix: isolate CI counter caches and preserve lazy exports (27a492b1e) — Vincent Hus
  • feat: show verified CI repair counts in banner and live prompt (ec8b1626f) — Vincent Hus
  • test: trim colocated plugin docstring to the contract (ddf7370da) — Vincent Hus
  • Add a first-run banner describing the run, and open clai sessions with it
  • Preserve text part boundaries after tool calls in OpenAIChatModel
  • fix(temporal): key tool opt-out checks on operation kind, not an MCP import

Sep 11, 2026

  • Custom environments. New environment APIs let plugins define how working areas are created for threads.
  • Machine plugins. New APIs let plugins provision and manage machines for development, including cloud machines that can pause and resume.
  • Browser control. A new plugin API lets plugins control bb's built-in desktop browser.
  • This release focuses on runtime MCP fixes, React hook performance, and improvements to the Inspector's onboarding experience.
  • Serve MCP Apps: AG2 can now serve MCP Apps, with structured tool output and extension support.
  • Input requests, both directions: An AG2 MCP server can ask the calling client for input, and an AG2 MCP client can answer a server that asks for it.
  • output_scan: A new policy that inspects what a tool returns, rather than what it was asked to do. Every policy so far ran before the call.
  • Admins can now manage models directly: add a model and its metadata gets looked up for you, each one is verified before it can be enabled, and tasks resume on whichever native runtime you pick.
  • Sandboxes grew up. They're independent resources now, with sensible defaults, one set of tools, and native provider recovery. We qualify them by actually running code on each provider.
  • Big files just work: uploads go straight to object storage and can resume if interrupted, and attachments stick around in completed chat history.
  • No more second writers. Profile gateways wrote hosted-room state into the root state.db every 5 seconds; the dashboard opened a writable handle on startup.
  • Healthy WAL databases stop wedging. OpenZFS (deleted) dentries and a close() racing an append_message both produced a sticky DeletedWalGenerationError on a perfectly good store.
  • FTS damage no longer kills your turn. An error scoped to the full-text-search index was classified as whole-file corruption and fail-closed the conversation.
  • Added claude plugin eval: run a plugin's eval suite against Claude Code and get scored, reproducible results (JSON + HTML report); see claude plugin eval --help
  • Added /output-style [name] to list and switch output styles, including over Remote Control and in cloud and other headless sessions
  • Added a diff of the files a Bash command changed to the Bash tool result when the Bash tool handles file edits (setting bashEditDiffEnabled)
  • feat: click an image attachment to open it full size
  • feat: only the creator can turn an automation back on
  • feat: allow editing automations on cloud backends
Claude Mem
  • fix(supervisor): jail SDK subprocess cwd
  • fix(hooks): anchor project names to Claude project dir
  • .NET: Replace deprecated AWSSDK.Extensions.Bedrock.MEAI with AWS.Bedrock.MEAI
  • .NET: Remove Azure.AI.OpenAI dependency
  • .NET: Track and update A2A task state
  • b36906e6d512264c008f2fe5c1262c8e21b09c20 MUL-6967 perf(inbox): derive the unread badge from the summary endpoint, not the full list
  • 24e200dca365fb978509dd376426c6ff9c6fecc5 MUL-7000: fix(agent): record Antigravity token usage
  • 676245af632db866dcba2afb47f5e7d99abf53b5 MUL-7034: fix(server): keep the parent workdir when an automatic retry must start fresh
No release notes published.
  • Windows: .exe installer or portable. macOS: .dmg (Apple Silicon aarch64 or Intel x86_64). Linux: .AppImage (universal) or .deb (Debian/Ubuntu).
  • Debian 12+ ships without FUSE 2 by default, which AppImage requires. If ./omniget.AppImage fails with a libfuse error, run
  • bash sudo apt install libfuse2
  • #224 — stale ChatGPT thread after a reclaimed address
  • #427 — doctor minting a pairing code too early
  • #419 / #420 — INIT checkpoint before the message is visible
  • feat(tls): support custom ca and client certificates
  • fix(xhr): resolve open URLs with credentials
  • feat(websocket): use native curl transport
  • New /branch command forks the current conversation into a new resumable session, leaving the current session unchanged. Resume the copy in another terminal with vibe --resume <id>.
  • Bump the local harness runtime to v0.4.5.
  • Make @ file mentions use Git-aware discovery and accept standalone pasted files and folders.
  • Code mode combines local and plugin tools in one JavaScript call, with saved task plans and automatic background command results.
  • More reliable message delivery, workflow retries, Compact & Resume, worker reuse, and idle browser-tab handling.
  • Improved Windows desktop control and window capture, plus faster recording and clearer tool results.
Langfuse
  • feat(filters): unify sidebar search and saved views
  • feat(experiments): name the expected variables when a dataset does not match
  • feat(experiments): colour-code the compared runs, align a score cell's icons, and lighten the dark-mode red
  • feat(viewer): add session compare page
  • fix(llm): add deepseek-flash model
  • fix(agent): make --preview apply the same selection the review applies
  • 点击动画资源改为从运行中的原生 helper 目录解析,不再依赖 SwiftPM 生成的资源访问器;后者在正式安装包里会回退到构建机路径,导致资源加载失败并让后台进程退出。
  • 动画帧缺失或不可读时保留程序化点击反馈,可见点击仍然正常返回,不再影响后续操作。
  • 加固原生回归与打包门禁:纯原生改动纳入 macOS 检查,在最终发布包中验证重定位后的资源确实可加载,并拒绝解析到应用目录之外的资源。
  • #2193 87bd594 Thanks @mattzcarey! - Extract facet ("sub-agent") machinery into packages/agents/src/dynamic-agents/, add the this.dynamicAgents capability facade, and reposition facets as an…
  • #2175 8ffb3ad Thanks @mattzcarey! - Lifecycle owns a durable job queue, driven as an alarm event loop.
  • #2198 99e5e2e Thanks @mattzcarey! - Add RoutedAgents (agents/routing), a Lifecycle capability that codifies the "user hub with one Durable Object per chat" topology: an owning Agent keeps a…
  • A run refuses to start on an unresolved variable — kane-cli run, kane-cli testrun, and kane-cli testmd run scan the resolved test for {{name}} references before anything is dispatched.
  • The receipt says what to do — unresolved names are grouped by the action they need, each pool file is named once, and the entry file is the one you gave on the command line.
  • What is never gated — values a test produces while it runs (store … as 'x'), replay-only steps, and runs dispatched to the HyperExecute grid go through as before.
  • A standalone template-center service enables multi-replica CubeMaster deployment.
  • The lifecycle management service now supports active/standby deployment.
  • Snapshot support for Host Mount and Volume Plugin: snapshot, restore, rollback, and clone now work for sandboxes with external mounts.
  • None called out in this changelog.
  • Added trace filters for span names, model providers, timing, outcomes, identity, and version lineage.
  • Added deleteFeedback() and deleteScores() to observability storage. Both accept a batch of ids and optional organizationId / resourceId tenant scope, are idempotent, and are implemented by the…
  • Agent Teams ReliabilityRail – Added repeat-tool, model-error, tool-error, output-length, compaction, and ping-pong detectors with configurable severity actions, leader escalation, local…
  • Third-party Agent reliability – Added retry and user-guidance handling for transient connection/reading timeouts, rate limiting (429), and service-unavailable (5xx) failures, with unified final…
  • SwarmFlow runtime controls, budget governance, and TUI commands – Added runtime control and budget-governance capabilities, together with a TUI command set for operational workflows…
  • The composer now shows the current branch's GitHub pull request — PR number, merge status, changed-line totals, and CI checks.
  • The Customize view's Tools, Skills, and Rules tabs now read as one consistent list instead of three different ones, matching the pattern Plugins already used.
  • Deleting a queued prompt no longer leaves it in the transcript as a message that never ran.
  • Verify that a result reproduces. Re-execute a captured artifact version from its sealed recipe — recorded inputs, environment lock, and execution evidence — in an isolated environment, and compare…
  • One search across the whole workspace. Global search reaches projects, sessions, message bodies, uploaded and generated files, and the literature library — grouped, filterable results with a…
  • Remote media and model endpoints are private by default. Images, audio, and video embedded in model output load only after explicit activation, images sent to models are normalized, and remote…
  • Added support for excludeDefaultComponents: true in custom agent Markdown frontmatter, allowing custom agents to opt out of default prompt sections and built-in tools while preserving…
  • Improved model API error resilience and diagnostics: transient genai.APIError failures (502, 503, 504, per-minute 429 rate limits, and mid-stream interruptions) automatically retry in-process with…
  • Improved MCP and provider tool schema validation to preserve open object schemas (such as {"type": "object"} or explicit additionalProperties: true) instead of rejecting undeclared arguments on…
  • fix(browser): load newly opened background tabs
  • Unify tab surface selection across workspace activation
  • fix(native-chat): keep worktree active during chat creation
  • Added send_message support for progress updates.
  • Added targeted conversation deletion.
  • Added support for managing tool approvals through tools.json.
  • Plugin switcher tabs: give custom providers their own tabs, refresh the selected plugin directly, and use plugin-only menus.
  • Optional pace colors: show pace indicators in green when behind pace and red when ahead, while retaining their signed values.
  • Better Codex cost tracking: restore shared spend in multi-account menus and read activity totals without decoding unused event history.
  • commit: abcf92c2b34e99209e505662a3fe4e11868e8aa1
  • source tag: beta2026-09-11_12-32-49
  • regions: cn global
  • Configure model routing separately for each Agent, including provider preferences and fallback behavior
  • Add Auto Fin proactive memory review and upgrade ReMe for more reliable long-term-memory processing
  • Add PowerContext as an optional long-term-memory backend with configuration in the Console
  • Release notes — formatted for people, with expandable sections.
  • Changelog — plain Markdown for AI agents and tools.
  • Signed source: 3a9d69db306cd7f081e06254cb89c4bcc14a7107.
  • Added LatencyBreakdown.contributions, a timeline of the user-to-bot interval whose durations sum to the measured latency.
  • Eval scenario turns can play an audio file as the user instead of synthesizing text.
  • Added a cancellable_by_llm argument to LLMSwitcher.register_function(), which forwards it to every LLM the switcher fronts.
  • Team management: create and maintain Agents, configure name, responsibilities, default model, skills, and knowledge, and copy the setup from an existing Agent.
  • Resource isolation: each Agent has its own workspace, memory, and sessions; the knowledge base and skills can be shared or kept independent.
  • Group collaboration: add multiple Agents to one session and use @ to direct a message to a specific member, with each Agent using its own configured model.
  • fix(scheduler): require full delivery for completion (934303849) — Anwesh
  • fix(scheduler): persist successful manual task runs (511224081) — Devesh
  • fix(ci-analytics): every analysis reads GitHub (ca099d85a) — Yauhen Bichel
  • Your companion now draws directly on your screen with a live pencil cursor, Link by Stripe is generally available for seamless wallet checkout, and independent tasks run in parallel across…
  • Link by Stripe is now generally available, fully supported as a first-class payment provider across all installs.
  • Your companion draws live on your screen. Annotating during a companion call now gives you a dedicated pencil cursor to sketch directly over shared windows…

Sep 10, 2026

  • Fix: a call to a Python function defined nested inside another function now resolves to that inner definition per lexical scope, instead of leaking to a same-named function elsewhere.
  • Fix: submodule imports inside a PEP 420 namespace package (a directory with no __init__.py) now resolve to the target module instead of being dropped
  • Fix: a bare-name import of a module sitting next to the importing file (a flat script dir with no package) now resolves to that sibling — matching CPython's sys.path[0] behavior — without…
  • chore: isolate ruff dependencies in ci
  • fix(telemetry): describe option objects in the session report
  • chore: remove commented-out keepalive stub from Deepgram V2 STT
  • Models: Safeguard agent uptime by automatically failing over to backup models when the primary model encounters errors. (c300b8d)
  • Integrations: Build rich voice and telephony agents using the newly added LiveKit runner. (f277d91)
  • Workflows: Simplify graph workflow management by authoring and loading ADK 2.0 graphs directly from declarative YAML configurations. (2a9461b)
  • Added to the Claude apps gateway: with pricing: set in gateway.yaml, signed-in Claude Code clients receive the same rates through managed settings, so /cost and telemetry match the spend meter
  • Added a startup warning for gateways when access_control.allow_cidrs is empty, and a one-time warning the first time a request arrives from a public address
  • Added the gatewayInternalNetworks managed setting, letting administrators allow /login to a Claude apps gateway on their organization's own public IPv4 block
Kilo Code
  • #14013 d2381d7 - Enable the experimental shared agent board (Kilo Swarm) with the KILO_EXPERIMENTAL_SHARED_AGENT_BOARD environment variable, or the umbrella KILO_EXPERIMENTAL, in addition to the…
  • fix(xlsx): synchronize VML note targets during sheet shifts
  • fix(pptx): preserve equation order in HTML previews
E2B
  • 5e418dc: Document that onResume / on_resume needs a control plane that knows the option: an older self-hosted or BYOC control plane drops the memory field and restores memory while reporting…
  • 9136603: Retry control-plane HTTP requests up to three times after 429 responses using the server's delta-seconds Retry-After delay.
  • 以前显式关掉过的用户不受影响——设置优先,更新不会替你打开。
  • 极简模式(minimal)依旧只保留 model + 工作目录,不会凭空多出进度条。
  • 鼠标点这一行(工具卡点卡面)即可展开,再点一次收起;键盘用 Ctrl+O。
  • feat(rules): add Rego policy review support
  • fix(llm): request stream usage by default with extra_body override
  • fix(scan): align preview selection with execution
  • VS Code extension: debug submenu with log level picker, open log file, and a togglable session status panel showing agent statistics (steps, tokens, tokens/sec, context, cost)
  • Open the session picker faster by building the merged session listing once per cursor walk instead of on every page.
  • Renamed SessionBackendKind literals from "python"/"rust" to "legacy"/"unified" to match backend identity naming.
  • 新增持久化的隔离 JavaScript 工作进程执行原生应用操作,可批量完成已知动作,动作之间不再需要模型往返。
  • 对齐 macOS 的手势、按键、应用清单、截图、滚动和剪贴板行为,并补齐已签名原生接收端的回归测试。
  • 坐标点击改为走精确的辅助功能命中;渲染期属性可复用;键盘宏被限制在已验证的原生边界内。
  • Bundle Astrid 2026.9.1 for initialization, resumed grants, durable capsule
  • Allow Community and Oracle host setup without an OpenAI API key. Provider
  • Read principal-scoped loaded capsule metadata for system inspection instead
Ziva
  • Godot 4.2+
  • Active Ziva account
  • The PageIndex SDK, local or cloud — vectorless, reasoning-based RAG, end to end.
  • Much faster indexing — the PageIndex Flash engine gets the tree from layout stats: no LLM involved for the structure generation itself, LLMs only write the node summaries, and tree expansion…
  • Flash engine: the local default (mode="standard" keeps the classic LLM pipeline).
  • Agentic RAG: Refactored the Agentic RAG retrieval framework, significantly improving reasoning speed and benchmark performance.
  • Knowledge compilation runtime settings
  • Enforce a rate limit on the knowledge compilation pipeline
  • Added answer forms for Codex questions asked while the agent continues working
  • Added Mark as unread to finished workspaces in the sidebar menu
  • Added an in-app What's new sheet in the help menu, Settings → About, and desktop update callout
  • Knowledge base service-API keys were scoped to the whole workspace: one key could read and write every knowledge base in the tenant, so giving an integration access to a single knowledge base…
  • The workflow canvas required a mouse to reposition anything. Nodes, multi-node selections, comment markers, and draft comments can now be moved with the arrow keys — hold Shift for larger steps —…
  • Plugin authors now have public profile pages at /marketplace/creator/<handle>, listing their creations with a publisher sidebar, reachable from a new Creator Center entry in the sidebar help menu.
  • samples: Add Magentic custom-manager prompts and MLflow observability examples
  • agent-framework-core: Add shared vector-store abstractions, portable filters, and an in-memory vector store
  • agent-framework-core: Add a maximum-duration bound and stop-reason signal to the tool invocation loop
  • fix(server): let pi be the only pre-session model resolver so a new w…
  • feat(subagents): add global and workspace controls
  • fix(web): stop the drag-canceled toast firing with no drag in progress
  • release: merge dev into main
  • fix(ego-browser): contain fixture and subscriber failures
  • ci(security): prevent shell injection via github.head_ref in Main PR Source
No release notes published.
  • Safer provider and channel boundaries: bundled providers and channel adapters now bound untrusted response bodies, reject oversized inputs before expensive work, and preserve safe recovery when…
  • More reliable long-running delivery: agent, gateway, retry, and channel paths now handle cancellation, retries, partial sends, and process-stream failures without losing work or replaying unsafe…
  • Stronger bundled-plugin resilience: local-model, browser, media, and collaboration plugins recover cleanly from malformed payloads, timeouts, and transient upstream failures.
WrenAI
  • core: support ordered cube queries (8d23753)
  • #810 65dbd57 Thanks @jkennedyvz! - Modernize runtime and development dependencies, upgrade pnpm, and pin the patched qs transitive dependency.
  • #794 3e180e9 Thanks @Yashwanth-Kumar-Kotla! - fix: unescape .env values in a single atomic pass
  • #796 4364f25 Thanks @Yashwanth-Kumar-Kotla! - fix: allow the google fonts origins the visualizer page requests
  • ChatGPT Subscription (Codex) now lists only the models your plan can actually use.
  • Windows updates no longer fail with "Error opening file for writing". The compiled sidecar re-executes itself as the detached Cline Hub daemon, which outlives the app by design, and Tauri's NSIS…
  • Your prompt is no longer lost when a send fails before the turn starts — switching to Codex and having the OAuth refresh throw, for instance.
  • ci: cap build-test at 60 min and stream prek hook progress
  • ci(deps): bump taiki-e/install-action from 2.87.0 to 2.87.5
  • fix(litellm): align embedding responses to inputs by index
  • Daily spend ledger: inspect tokens, requests, and spending day by day in Usage & Spend, with your selected time zone and clear labels for unavailable amounts.
  • More useful cost charts: hover over a daily bar in the menu to see its date, cost, and token count.
  • Menus that fit your workflow: choose visible usage rows, select session or weekly percentages, and add explicit reset countdowns or clocks to menu-bar layouts.
  • Added the remote-control start, remote-control status, and remote-control stop subcommands to run the CLI as a background service registered with your operating system's service manager so your…
  • Improved half-page scrolling across all scrollable views — including altscreen mode, the diff viewer, and read-only detail panels where ctrl+d no longer triggers the exit prompt — and set shift+up…
  • Fixed prompts or model responses blocked by content safety filters failing with a spurious "no candidate found" error, ending silently with an empty turn, or repeatedly retrying the rejected input.
  • fix(skills): the scheduled-agent demo asks one declared question (962aa8187) — Yauhen Bichel
  • fix(shell): the onboarding question is asked once, unless you ask for it (5cbeb8b4c) — Yauhen Bichel
  • fix(scheduler): migrate legacy task kinds (11e71ad31) — Anwesh
  • Recommended GLM 5.3 models are now available in code workflows.
  • Added pricing support for Fireworks GLM-5.3 and GLM-5.3 Flash usage.
  • Workspace switches now ask for confirmation before restarting the server.

Sep 9, 2026

  • fix(webeye): allow raw-CDP screenshot rescue on persistent-context browsers (24d08f8ad)
  • Gate passkey 2FA behind the enterprise plan with a visible book-a-demo upsell tile (SKY-14535) (92976292e)
  • Run each Copilot build test in a browser that matches where the test starts (SKY-16164) (a93ad7831)
Traycer
  • Traycer Host 1.3.1.
  • Add telemetry to track checkpoint runtime and CLI usage.
  • Fix gateway errors reported inside an HTTP 200 response.
  • Keep deploy push on the AMP create source.
  • feat(vended-tools): port notebook tool to Python (545d9ed07)
  • feat(bidi): simplify Google and OpenAI model configs (6d4f61f28)
  • feat(python): add backgroundTasks to Agent (42e5a61bd)
  • fix(clipboard): only report copy success if it happens
  • fix: stop long reasoning traces from running away with the heap
  • fix(config): honor providers.json when auto-update is disabled
  • This release converges React Native's render-tool hooks onto react-core, improves Copilot context timing during page navigation, and adds provider-level agent configuration.
  • Added maxEffortLevel setting (top-level or per model under modelSettings): caps the effort level on every provider, including Bedrock, Vertex and Foundry; users can still pick a lower level
  • Added --system-prompt-snapshot off to render the system prompt fresh on every request instead of reusing the conversation's recorded prompt (for iterating on prompt text)
  • Fixed Cowork scheduled tasks in the cloud failing at startup for organizations whose managed settings require sandboxing
  • author keeps the test's identity — every step is re-authored on the existing test and the run commits a new version on the same test case, instead of minting a brand-new test on each --author run.
  • testrun run --author reaches every member — the flag is forwarded to members locally and on --remote, where the grid now installs the same kane-cli version you dispatched from.
  • A mid-run lock conflict no longer abandons a heal — under the readonly policy the heal keeps shrinking and re-authoring locally, the result is not committed, and stderr points at --push.
  • feat: show manifest-declared value statements on dashboard cards and rows
  • feat: Make automation run UI task-outcome aware
  • feat: enable LLM provider connections on cloud
No release notes published.
  • fix(desktop): relaunch agents on the host when the default account switches
  • chore(desktop): bump version to 1.27.0
  • feat(desktop): open the PR summary as a workspace pane from the Changes pane
  • Unify sidebar create actions into single dropdown menu
  • fix(sidebar): reveal collapsed workspaces without clearing filters
  • fix(ui): ignore the persistent workspace list when detecting overlays
  • Added automatic language detection for untitled buffers. (#61412; thanks amtoaer)
  • Added support for rendering LSP 3.18 Markdown diagnostic messages from language servers. (#61030; thanks aviatesk)
  • Fixed searching for full SHA-256 commit hashes in the Git Graph. (#63371; thanks albertbogusz)
  • Import PDFs in batches. The literature library imports many PDFs in one pass — selection, destination, duplicate policy, per-file progress, and retry of unfinished files — and the sidebar shows…
  • Run long work in the background. Notebook, REPL, and shell jobs can run in the background: the agent turn is released while exact run identity, cancellation, and provenance are kept, and results…
  • A broad literature stability pass. Library changes sync across clients immediately, stale collection edits surface a conflict instead of overwriting saved values, and data-location relocation…
  • Signed macOS filesystem app installation through the common installer,
  • Native Linux musl product archives for x86_64 and ARM64, authenticated platform
  • aos distro apply --principal P --yes for applying the signed Community
  • feat(public-api): optional startTime on GET /observations/{id}
  • feat(filters): surface has:/-has: while typing a nullable field name
  • feat(web): add source promt name to playground window content
  • 8e48365a05a87df485ce242ec7088fd11f79b76a feat: fall back to user/list_id in WeCom syncer
  • a61599b1852a024c15430aa454c4d6a772c64b88 feat: support department sync in WeCom syncer
  • 311f4d17a65bbb33ec30762e9bf61935c0e8997e fix(auth): drop the panel card on a phone
  • 🐙 A new GitHub panel in the session workspace rail shows the branch's PR — summary, comments, CI checks, and a stacked diff with a file-tree navigator — refreshing live as the agent pushes.
  • 🎨 Canvas is now a built-in page: a per-project board of draggable session cards with live status, worktree branches, and PR links, where you can open transcripts or start new sessions right from…
  • 📦 New managed-sandbox providers — run each host in a local libkrun microVM (sandbox.provider: microsandbox) with sub-second boots and in-place resume, or as self-reclaiming Kubernetes…
Mem0
  • Reuses shared conversation preparation, memory formatting, project/session/global scope utilities, and telemetry while preserving Pi's native extension API and @mem0/pi-agent-plugin package name.
  • Pi loads the built dist/entry.js extension instead of executing source TypeScript from an installed package. Builds also publish the library entry point and declarations.
  • Removed Dream consolidation and pin commands, skills, configuration, types, and exports. The remaining commands are remember, search, forget, tour, scope, and status.
  • Godot 4.2+
  • Active Ziva account
  • Reference images: Paste or attach an image in chat and pass it straight to image generation as a reference. Previously the model could only describe it in words, which sends words rather than pixels.
  • 261522e3e3d71516d6b34d7c5ab1ccc8a4669718 MUL-6788: perf(server): batch claim-path attachment and runtime lookups
  • 074f5ab237f1c42cc3315199a7214d66d24ca722 MUL-6834: fix(codex): preserve token usage on cancelled turns
  • cc758a74e149bb51d7e32581a811093340594306 MUL-6941 feat(daemon): record task phase latency
  • Fix: an incremental rebuild no longer wipes cross-file project AST nodes — re-extracting one .csproj/.sln was dropping package/framework nodes of a referenced project…
  • Fix: when duplicate nodes merge, the richer (more complete) node is now kept as the survivor and the losers' non-empty fields are folded in, instead of a shorter-id passing mention winning and…
  • Fix: a C# generic call site with explicit type arguments — Get<int>(...), unqualified or through this — now resolves to the method definition instead of capturing Get<int> as the callee and…
  • fix(worker): disable piece cache prewarm
  • Revert "fix(connections): stop oauth2 connections failing to save when the code has special characters"
  • Desktop keeps 15 recent managed Code worktrees by default, configurable in Local Coding settings; cleanup skips active, dirty, untracked, or unpushed worktrees.
  • Background process tools (process.start, process.stop, process.output, process.list, process.write) now show the command or process ID in the call header and a typed result widget instead of raw JSON.
  • Tool call groups now fold into a single compact summary line by default. Click to expand the full list. Ctrl+O toggles all groups and result bodies.
  • Enforced negotiated WebSocket payload ceilings with a symmetric checksummed transfer protocol for large operations, including bounded transfer state, disk spooling, and automatic cancellation on…
  • Made workspace HTTP file transfers atomic with streaming uploads into partial files, size verification, fsync, and SHA-256 download receipts
  • Bounded remote connector tool payloads with explicit continuation metadata for large content, rejecting oversized inline artifacts before base64 expansion and routing bulk files to authenticated…
  • [Agent] Agent sessions now label self-started replies — “Goal round N” for DSH goal continuations, or “Continued after background work” — so users understand why a new message appears.
  • [Models] DeepSeek V4.1 Flash Expires-On-0910 vision model is now available in the catalog.
  • [Cherry Cloud] Cloud sessions are now bound to the current computer, preserving free-grant history across credential resets.
  • fix(output): write reports atomically
  • fix(mcp): bound the MCP shutdown phase
  • fix(review): exit immediately on the second signal
E2B
  • f842aa8: Expose a configurable minimum free-disk target with minFreeDiskMb in JavaScript, min_free_disk_mb in Python, and --min-free-disk-mb in template create.
  • Updated dependencies [f842aa8]
  • e2b@2.49.0
  • feat: support current image generation tool options
  • fix(sandbox): prevent UnixLocal file API symlink races
  • fix(sessions): reset compaction response chain after pop
  • The Files and Changes panel now has a single top-bar Refresh button that does the right thing per tab: on Files it re-mounts every root and revalidates project detail.
  • A git worktree created inside an already-open project now shows up on the Changes tab after a single refresh — no more switching projects away and back to make it appear.
  • The Refresh button tracks real in-flight state (busy/spinner) and ignores repeated clicks until the refresh completes.
  • v0.1.4 adds packed SQLite storage, native FULL/PREFIX representations, bounded shared Init/Commit publication, and reuse of authenticated Init comparison bytes.
  • The accepted full157 observation uses 184,582,144 allocated bytes, 15.374% less than its 218,116,096-byte supplemental control.
  • Validation covers 198 performance cases, 226 routine proofs, supplemental small-file/SDK/FUSE workloads, 157 performance states and 157 retained-history proofs.
  • New app: GenOffice HTML. Create and edit .html files with AI in two modes: AI Design builds visual pages (landing pages, dashboards, slides-style layouts) from a brief with style directions, and…
  • AI: separate BYOK keys for web search, image generation and media analysis; new Codex App Server provider; Ask AI quotes and highlights the selected text.
  • App: global AutoSave; AI panel text size and spellcheck settings; Czech UI.
  • @mastra/factory: defineBoard() phases must add kind, working phases must add role, and initialPhase must be resting.
  • @mastra/factory: global rules object removed (FactoryRules, defaults/merge/assert helpers, new MastraFactory({ rules })).
  • @mastra/factory: GET /web/factory/projects/:id/attention response shape changed (tier removed; counts/latest fields moved under kinds[kind]...).
  • Fixed the live chat stream doubling text and dropping messages mid-turn. The sidecar has two Hub sockets that both receive a session's events — ClineCore's own client and the observer client — and…
  • Fixed a queued prompt's own message vanishing from the chat. When you queue a prompt behind a running turn, the runtime drains the queue just before it answers the previous send, so the previous…
  • Cline no longer stops silently mid-task when a model gets stuck repeating itself.
  • UK equity market
  • Zerodha Kite Connect
  • Read-only multi-broker portfolio
BrowserSkill
  • fix(extension): support UUID generation on insecure origins
  • fix(extension): keep user-opened tabs free & fix agent-tab leak on session stop (issue
  • fix(bsk): fix Windows self-update failing to replace binary
  • #3613 d4d20d2 Thanks @liukx0205! - Add read-only tools to the /btw side agent.
  • #3524 f6a9c39 Thanks @RealKai42! - Add an experimental Updates panel with paginated progress messages from the main agent and subagents; enable it with KIMI_CODE_EXPERIMENTAL_NOTIFY_USER=1.
  • #3634 e831fd1 Thanks @7Sageer! - The subagent model pool ([secondary_model]) is now always on.
  • @sunznx, for reporting the duplicate skills, tracing the fallback, and proposing the scoped manifest fix in #241.
  • Added the Astra system prompt for GPT-6 models.
  • Preserved Bedrock DeepSeek model IDs, including ARN-based IDs, so they resolve correctly. (@YeEmrick)
  • Updated the Azure provider SDK to pick up compatibility fixes.
  • Reject invalid DeferredToolResults.approvals values
  • Add a GitHubCopilotProvider for GitHub Copilot's OpenAI-compatible API
  • Honor anthropic_disallows_sampling_settings in BedrockConverseModel
  • commit: 0d1a5c7da95e22a1215dc5ad637bf60549453ee8
  • source tag: beta2026-09-09_10-52-41
  • regions: cn global
  • Added GPT-6 Astra support.
  • Reuses the server across workspace switches.
  • Limits thread resume age.
  • Memory v3 switches to pinpoint section retrieval, GPT-6 Astra lands in the catalog, and your companion shares your screen and points at what you are looking at.
  • Pinpoint memory with Memory v3. Your assistant now retrieves and injects individual sections rather than loading whole pages at once.
  • GPT-6 Astra arrives. GPT-6 Astra joins the model catalog across OpenAI and OpenRouter with full Responses API effort support, ready for both standard conversations and code assistance.
  • feat(skills): pre_execute hook opens the onboarding menu before any model step (0d3a9a6d9) — Vincent Hus
  • ci: remove redundant interactive-shell workflow (95a745e65) — Devesh
  • fix(ci): drop stale github-ci-fix-onboarding references; fix morning-report ordering test (7f3743d4f) — Vincent Hus
  • Improved resilience to transient model API errors: the agent now retries errors such as 503 Unavailable for much longer with exponential backoff, so a brief service hiccup no longer aborts your…
  • Improved sign-in and startup speed by reading your signed-in identity from the stored credential instead of making a network request on every launch and after browser consent.
  • Improved headless (-p) runs to exit promptly once the final answer is delivered: the CLI waits for running background tasks and scheduled timers to finish, bound by --print-timeout, and leaves…

Sep 8, 2026

Claude Code
  • Fixed a 2.1.265 regression affecting LLM-gateway and proxy setups: the undocumented CLAUDE_CODE_USE_GATEWAY environment variable, previously ignored unless ANTHROPIC_BASE_URL and…
CopilotKit
  • Remove the inspectorLearning option from Fetch, Express, and Hono handlers. Inspector Learning now requires an Intelligence runtime with a configured Learning container, not debug mode or a…
  • Provider-level humanInTheLoop tools now wait for respond, keep parallel calls independent, and report an error when aborted instead of an empty success.
  • Custom useDefaultRenderTool renderers can return null to hide selected tool calls without a type error.
  • fix(frontend): render no icon for plain select trigger
  • fix: detect watsonx models behind LiteLLM proxy
  • fix(cli): clean up failed serve inputs
  • A Python kernel that dies after a successful startup is restarted on the next use instead of every call being handed the dead kernel forever, and skill-MCP tools advertise their real input schemas…
  • Moved the semantic-edge ledger's append and replay IO onto the shared event-log substrate.
  • Made every durable JSON/JSONL state write crash-safe through one shared atomic-write owner (temp file + rename, Windows rename retry): auth.json is no longer written in place…
Agno
  • fix: clean startup logs and initialize durable queue storage
  • fix: preserve Control Plane access with public surfaces
  • Release: v3.0.9
  • fix(cssom): read used sizes from existing layout snapshots
  • perf(style): reuse cached owner import metadata
  • fix(layout): omit whitespace-only flex and grid items
  • Changelog for v0.58.0-preview.0
  • chore(release): bump version to 0.59.0-nightly.20260825.g812f7a2bc
  • fix(core): prevent SSRF in MCP OAuth metadata discovery and authentication
  • fix(ci): bump gh-action-pypi-publish for metadata 2.5
  • Bump graphiti-core to 0.30.1
  • Fix Neo4j database routing and honor NEO4J_DATABASE in MCP server
  • 会话内缩略图:用户附件、助手回复、工具结果里的图片(PNG/JPEG/WebP/GIF)在转录区显示为有界等比缩略图,只有图片的消息也带本地化标签;缩略图随滚动按可见区域裁剪,不覆盖输入栏(#691 #775)。
  • 居中大图预览:全屏下点击输入框里的 [Image #N] 或转录区缩略图打开预览,标题为 Image #N — 格式 · 宽×高 · 体积 · 文件名,卡片最多占对话区约 95%;预览打开时对话区内容变淡(前景/背景向终端底色混合一半),输入栏与状态栏保持清晰(#697 #775)。
  • 缩放 / 平移 / 看图导航:「适应」「100%」与 +/- 在 100%、200%、400%、800% 之间切换,拖动、滚轮或方向按钮平移;←/→ 或底部 ‹/› 切换图片(首尾不循环,显示「当前张/总张数」)(#779)。
  • Classify finding severity with custom rubrics and supporting context through the CLI and SDK.
  • Match repeated findings across scan history, preserving confirmed identities and related-finding relationships. Automatic matching restores batching, and confirmed-finding lookups are faster.
  • Open draft GitLab merge requests for verified patches with the existing --create-pr option, including self-hosted GitLab. See patch publication.
  • Support GPT-6 Astra models
  • Tool calling for goose-agent
  • Kotlin callers can now configure Databricks AI Gateway path
  • Godot 4.2+
  • Active Ziva account
  • Revamped image generation: We now have a single generate_image tool with multiple backends, including RetroDiffusion and ChatGPT, and Ziva intelligently uses it to cover every art style. Try it out!
  • GateGuard and governance capture recognize destructive PowerShell commands,
  • Relative GateGuard exemption globs stay within the project root. Explicit
  • Installer writes reject collisions with untracked user-owned files. Failed
  • New Plugins workspace with an independent MCP connector, a reviewed catalog, custom servers and MCPB imports.
  • Blender, Memory, Playwright, Fetch and Unity integrations, plus browser sign-in for hosted HeyGen and Recraft services. Enabled plugins restore across app restarts.
  • Desktop-launched macOS/Linux apps find runtimes in standard Homebrew, Node and user uv installation locations while preserving the user's configured PATH precedence.
Strands Agents
  • feat: export context manager types as experimental (ec6eb54a5)
  • feat: accept cache_config on Ollama, Writer, and SageMaker (42aedfc03)
  • feat(context-py): port stash integration (a642b4177)
agent-browser
  • Fixed Windows headless Chrome desktop artifacts by isolating owned headless Chrome on a private desktop and ensuring its process tree is cleaned up when the daemon exits or is forcibly terminated.
Kilo Code
  • Move a session to a worktree directly from the worktree editor.
  • Fork a session from the worktree editor.
  • Add Ctrl+1/2/3/0 shortcuts to cycle mode, model, and reasoning effort.
  • serve no longer re-archives the whole data dir on every start. The
  • Promptless internal sessions no longer flood the wiki. A session is now
  • Wiki commits stop re-hashing the entire tree. Since the #594 guard,
  • Claude cost breakdowns: inspect daily usage and top models with the new opt-in CLI cost --breakdown output.
  • Faster, more accurate cost history: reuse Claude reports across launches, resume interrupted Codex scans, and recover usage missed by older parsers.
  • Clearer menus and fresher dashboards: refresh spend charts when returning to the app, show money or points for balance-only providers, and honor segmented claude-swap account menus.
  • Safer updates: rehearse core and plugin changes in isolated candidate state before activation, support eligible 2026.9.2 migrations, and recover abandoned update records without stopping a healthy…
  • Performance: preserve warm prompt caches, reduce unnecessary work during cold session updates and memory search, and reuse worker builds between sessions. Related #140681.
  • Skill Workshop: keep skills in one persistent agent-owned collection across workspaces, compare complete skill instructions, and retire missing-draft suggestions safely through Doctor.
  • Realtime Voice Agent: Introduced a realtime voice agent with DashScope backends and local transport support.
  • A2A Protocol Support: Added A2AAgent class to the agent module, enabling agent-to-agent communication.
  • New Pipeline Module: Added a new pipeline module with a GoalPipeline class for goal-driven agent workflows.
  • Fixed OpenClaw Installation: Restored the installation and automatic memory hooks for OpenClaw.
Langfuse
  • feat(preview): alert when a labeled preview is not serving
  • feat(worker): emit export data-freshness lag distribution
  • fix(datasets): read v4 dataset metrics when preview is enabled
  • Revert "fix(connections): stop oauth2 connections failing to save when the code has special characters"
  • ROB-1161 Fetch the Supabase api key from relay with caching and local fallback
  • Patch all High CVEs (and Mediums >60 days) in the holmes image
  • ROB-1258 Live-refresh git-synced skills via first-class skill_repos
OpenViking
  • feat(web-studio): share user skills and switch users
  • Revert "feat(web-studio): share user skills and switch users"
  • feat(web-studio): add Agent Experience module
  • [Models] Add GPT-6 Astra support with context limits, reasoning controls, Fast mode, native web search, tiered pricing, and the official model icon.
  • [Models] Add TokenDance as a built-in model provider with browser-based API key authorization.
  • [Scheduled Tasks] Daily, weekdays, and weekly presets now support multiple times per day (e.g. 09:00 and 18:00).
  • Multiple named plans in one project now require PLAN_ID, even when .planning/sessions/ does not exist.
  • UserPromptSubmit explains the missing pin. Per-tool and PreCompact hooks remain quiet, while Stop and attestation refuse to use an unrelated root plan. The shell, PowerShell and Python paths agree.
  • Set PLAN_ID in the environment that launches each host. Setting it inside a tool subprocess does not update an already-running host.
  • feat(core): support image results in web search tools
  • feat(core): customize output guardrail blocked messages
  • feat(mcp): add server-wide guardrails to MCP tools
  • Improved Codex chat UI is available when Structured Chat is enabled in Experimental settings, with inline file diffs, pane actions, clearer tool output, individual task stops, and structured…
  • GitHub Projects Roadmap views now render as timelines; the desktop also adds tab scrolling, narrower tabs, and refreshed notifications.
  • Faster workspace, renderer, editor, terminal, Git, and remote operations, alongside stronger SSH relay and Windows/WSL recovery.
  • LLM-guided multi-round episode retrieval.
  • Independent decider configuration. The new [decider] section selects
  • Unified benchmark harness. One runner and four adapters cover LoCoMo,
  • wren: add wren cloud to bind a local project to a Wren Cloud git remote (8e532b4)
  • wren: add a torch-free onnx embedding backend for wren memory (308a0dd)
  • wren: connect MSSqlConnector with autocommit so statements are not left in an open transaction (4d167fc)
  • Two-pane Settings shell; leaner account menu icons
  • Fix desktop server switching at the welcome screen
  • Add local server settings to the desktop app menu
  • Shared live Workspace core for FUSE operations and SDK edits, with commands and handles continuing across live-FUSE Commit boundaries.
  • In-place checkpoint installation, ordered namespace finalization, shared backing segments, and owned object admission remove repeated reconstruction, file cleanup, copying, and spill readback.
  • Payload, CAS/CDC, and Workspace-reuse scaling improvements with separately recorded qualification campaigns.
  • Deprecate fallback_model in favor of fallback_subagent_model on ImageGeneration and XSearch
  • Add openai-codex provider for ChatGPT/Codex subscription authentication
  • Add a direct image generation API with ImageGenerator
  • Inherited upstream presets: gpt-6-astra, gpt-5.6-sol,
  • Google AI Studio IDs: gemini-3.8-flash, gemini-3.7-flash,
  • Anthropic IDs: claude-fable-5-1, claude-opus-5, claude-sonnet-5, and claude-haiku-4-5-20251001. These are current active provider-documented IDs surfaced in the model guide.
  • feat(scheduler): deterministic CI reliability loop, background service, raw snapshots (751e7da2e) — Yauhen Bichel
  • feat(onboarding): CI reliability agent demo schedules a local loop and runs it once (724d3c783) — Yauhen Bichel
  • feat(ci-analytics): measure blocked time per merged PR as expected vs actual green (1a81e1a1b) — Yauhen Bichel

Sep 7, 2026

  • fx now uses a smaller shell tool with 75% fewer actions (3 vs 12), replacing the old terminal tool.
  • Enter now steers active turns instead of queuing follow-ups, and ctrl+enter does the same.
  • The subagent interface now has 67% fewer commands (2 vs 6), with direct delegation replacing the ctrl+x manager.
Rakazo
  • feat: improve Pi runtime and tool completion auditing
  • Simplify the first-run computer choice copy
  • Show real setup progress and keep the panel alive during long pulls
  • Existing install: hermes update
  • Fresh install: curl -fsSL | bash
  • Managed deployments should update through their deployment tooling using the new tag.
  • Added Discord channel support, including improved gateway failure reporting after startup.
  • Added chat-scoped conversation history with configurable storage and optional hybrid search.
  • Added MCP configuration tools, channel-based MCP server authorization, hot reload for MCP configuration, and OAuth/device authentication support for Slack and GitHub MCP integrations.
  • Fix: Rust trait method declarations (signature-only, and default-bodied) are now extracted as nodes — trait bodies were never walked, so both were silently dropped.
  • Fix: the atomic-write temp filename is now bounded, so exporting to a path near the Windows MAX_PATH / 255-char component limit no longer fails with a temp-file FileNotFoundError
  • Fix: when graphify's git hook decides to skip (no graph, rebase/merge/worktree, GRAPHIFY_SKIP_HOOK), it no longer terminates the whole hook — the block runs in a subshell so chained hooks and…
  • fix(batch): preserve per-item warnings across CLI and MCP
  • fix(xlsx): exclude phonetic guide text from cell values
  • Manage Local and saved SSH machines from one Herdr window, with a combined agent list, machine-scoped navigation, notifications, and automatic reconnects.
  • Multiple clients can now view different workspaces and tabs independently. Different tabs fit their viewing clients; when clients share a tab, the last one to interact with it controls its size.
  • Added Muse agent detection for idle, working, approval, and question states.
  • chore(relay): one relay URL everywhere, drop the relay-url-override flag
  • fix(relay2,host-service): fail fast on failed dial-backs and place tunnel objects at the host
  • chore(relay): delete the v1 relay and call the Durable Objects relay the relay
  • Account-specific model and reasoning discovery now handles localized and nested ChatGPT pickers through native metadata. Model availability follows the signed-in account.
  • Fewer browser tabs: one operation retains one opening attempt across navigation, closure and restart. Sleeping workers release browser memory while keeping their reusable conversations.
  • Durable Goal/Loop and finish delivery preserves the exact task and turn, ordered queues and delivery receipts across recovery. Editing an objective re-evaluates an already completed answer.
  • More reliable account handling: preserve Codex permission errors, retain Claude quota-warning history across credential refreshes, and keep Copilot accounts distinct.
  • Accurate provider history: count Poe weekly usage over the last seven days and keep valid Poe, MiMo, and OpenRouter usage available when optional history contains malformed data.
  • Better display support: preserve menu-bar positions on monitors left of the primary display and translate Claude's model-specific weekly quota labels.
  • perf(worker): compute prewarm pieces and code steps server-side ( #15310 )
  • > 📌 Upgrading a self-hosted instance? Review the Breaking Changes page for any actions required before you upgrade.
  • Optional Milvus derived-index backend. Install with
  • Vector search now uses cosine distance on LanceDB too. The generic
  • everalgo's transitive layer is pinned. everalgo-boundary,
  • Unified session & message handling: chat messages are consolidated into a single message model, removing inconsistencies between multiple handling paths.
  • Hardened login flow: cross-process token refresh is serialized to prevent token-chain failure; tokens are refreshed proactively before expiry to avoid forced re-login.
  • Model selection validation: the /model command now validates that the selected model is supported by the current endpoint, giving clear feedback when an invalid model name is entered.
  • 728d26f0b5ccdc17398c9c61ff26077d8200372f MUL-6211: feat(agents): expose usage in task history
  • 5f1f297d575c348e8b80d9dae8944fb2e131f529 MUL-6966 chore(brief): stop teaching agents to use issue metadata
  • 8c4e0d2985354503c0c3693fe7ac5179675734bd MUL-6994 feat(cli): add --fields to issue list JSON output
  • The events now run in one Python process. scripts/inject-plan.py is a byte-identical twin of inject-plan.sh plus the dispatcher logic of hooks/claude-hook.sh.
  • The standalone skill route uses the same twin once a plan is accepted, with a fork-free interpreter discovery. A directory without planning state answers before any fork on both routes.
  • Both routes share one turn-marker slot and one progress-guard slot per plan: the launcher hands the twin the shell's own $PWD spelling.
  • feat(allowlist): add OCaml/ReasonML language support
  • feat(action): first-class review controls (effort, max_tokens_budget, llm_reasoning_effort) and live progress
  • feat(viewer): number the lines inside the Existing Code block
  • Connecting to an SSH host now asks a plain yes-or-no question before offering to import a local OpenAI or Claude login
  • The remote login picker lists the providers the host actually supports, with local imports shown as explicit choices
  • Remote login status is sanitized before display, so host credential details are never echoed back
  • fix(telegram): redact bot tokens from gateway poller errors (cd7e1b913) — Ceren Camkiran
  • Run on HPC clusters with Slurm. Remote compute hosts gain a per-host execution mode — direct SSH or Slurm — so notebook runs can submit as Slurm jobs, with durable submission, polling, recovery…
  • A reference library for your literature. Import references by identifier or file, organize them into collections, link them to projects, compare and bulk-merge duplicates, attach full-text PDFs…
  • New providers and fresher model catalogs. Apodex joins the built-in providers, and the latest OpenAI and Anthropic models — GPT-6 Astra and Claude Fable 5.1 — are selectable out of the box.
  • Anthropic SDK 1.x — AG2 adapts to the breaking changes in the Anthropic SDK 1.x line, with Anthropic model name references updated to match.
  • OpenAI SDK 3.x — Support for the OpenAI SDK 3.x line, including hosted shell and MCP events.
  • AnthropicBashTool — New built-in tool, with improved event handling for Anthropic.

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.