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: Aug 31 – Sep 6, 2026

268 stable releases from 99 tracked agents and frameworks in the week of Aug 31, 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 6, 2026

  • Persistent team bot desktops, more reliable computer recovery, and cloud coding agents.
  • Multi-account integrations, optional integration catalogs, GraphQL tools, and signed GitHub routine triggers.
  • Configurable model reasoning, editable agent memory and skills, and protected credential handoff.
  • Auth-disabled servers stop returning 401 for a stale bearer; invalid
  • Cursor sessions are filed under their repo and the cursor agent instead of
  • reindex no longer aborts on a scope created after startup — the manifest is
  • New desktop UI, projects, images, model picker and editable plans.
  • Durable injections, FIFO after-turn tasks, accurate transcripts, clearer worker messages and retry controls.
  • Astra finish queues and Loop; Pro: no auto-compaction, ten-minute recovery window.
  • Windows on Arm: native ARM64 installer joins the stable release. If you run the x64 build on an Arm device, install this once and it will receive ARM64 updates automatically.
  • App: collapsible ribbon like Office in every editor; AI chat renders tables and code blocks; "New chat" fully clears history and attachments; RTL-aware AI panels.
  • Docs: long documents open much faster; spellcheck toggle; edit existing hyperlinks; per-side table borders; embedded DOCX fonts load.
  • Shuttle playback shortcuts: J and L step the active scene faster in the direction it is already going, or turn it around at 1x, and K parks it.
  • Layout constraints in the inspector: the transform panel now sets constrainX and constrainY per element, backed by named horizontal and vertical constraint sets in the runtime and documented in…
  • Timeline drag fix: dragging a clip (or a multi-clip selection) can no longer push it past the start of the scene — the earliest dragged clip pins at frame 0 and snapping keeps the spacing between…
  • cc39a9395698ca9cf6cf869c963a641c7c015a71 feat(auth): redesign the sign-in, sign-up and forget pages and fix dark mode
  • d353b156f4f075f25faa5895b50e698db78e9ba7 feat: group release notes by commit type
  • 75420fc5406672afb9a97f4541963a0e8d22ad75 fix(apple): use the real name from Sign in with Apple as the display name
  • Live widget freshness labels: snapshot and stale token-history ages keep advancing between timeline reloads (#3445; partial fix for #3339).
  • Reliable Linux caches: preserve model pricing and Pi/OMP cost history across repeated refreshes.
  • Clearer provider usage: retain Command Code monthly usage during subscription timeouts, show Moonshot balances in the correct currency, and explain ElevenLabs API key failures.
jcode
  • Run the local terminal UI against a remote Jcode server over SSH, keeping the workspace, tools, and agent execution on the remote host
  • Explicitly import a local Jcode-managed OpenAI or Claude login to a trusted SSH host after confirmation, without overwriting an existing remote credential store
  • Sign in to Novita with its built-in API-key login option
  • 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).
  • Fixed ChatGPT OAuth model discovery hiding GPT-6 Astra by advertising Codex CLI 0.153.4 instead of 0.147.0.
  • Sessions that hold up. Resuming a session restores what the agent did, not only what it said; compaction keeps the recent tail and can be triggered on demand with /compact.
  • Your models, declared. A dsh-style settings dialog with per-model declarations (context window, capacities, reasoning ladder), live model discovery per connection, a Session-level context-window…
  • A terminal that reads like an app. Every bare command opens a selector, the TUI has a status line and tool cards, and a new turn no longer discards the prompt cache.
  • Godot 4.2+
  • Active Ziva account
  • GPT 6 Astra: Hosted on Pro, or with Codex BYOK
  • Moved the canonical GitHub repository to advaitpaliwal/feynman. Native installers, skills downloads, package source/issue metadata, and public documentation now use the new owner directly.
  • The canonical npm package is now @advaitpaliwal/feynman. Migrate old npm installs with npm uninstall -g @companion-ai/feynman, then npm install -g @advaitpaliwal/feynman.
  • Historical release attestations retain their original source identity; new releases are verified against the new repository owner without relaxing provenance checks.
  • A native macOS engine delivers input to the target app and window, with a virtual cursor that shows the agent's actions independently of your own pointer.
  • Background window capture stays fresh when another app covers the target. Window movement, resizing, minimized targets and focus changes are handled more reliably, including repeated text entry…
  • App selection shows native icons and resolves the exact app bundle. Accessibility controls with duplicate groups or descriptive labels remain actionable.
  • Stream PTC tool calls natively.
  • Keep private state out of subagent propagation.
  • 6c89fe2 release(langchain-quickjs): 0.3.7
  • Bug fixes and reliability improvements
  • No notable changes
  • refactor(persistence): centralize scheduler database access (dc28c3d5e) — Anwesh
  • fix(scheduler): reclaim claims abandoned by crashed workers (11588056d) — Devesh
  • fix(scheduler): preserve recurring morning-report payloads (90ffbc591) — Devesh

Sep 5, 2026

  • Allow operator admin login without email
Graphify
  • Fix: a module docstring preceded by a leading comment (shebang, # -*- coding -*-, or a license header) is now extracted instead of silently dropped — comments are skipped when locating the first…
  • Fix: two Python definitions whose ids differ only by leading underscores (e.g. _get_connection and get_connection) no longer collide and silently drop one — private/dunder members are salted while…
  • Fix: a ghost node whose source_file was set to the doc that merely mentions a file is now merged onto the real file node via a conservative unique-label fallback, so it stops surviving every rebuild
  • Claude plugin and standalone hook JSON preserve literal backslashes under POSIX awk, including Windows paths.
  • Attached Codex, Hermes and Pi sessions now require PLAN_ID when their project contains multiple plans. An attachment authorizes a session; it does not choose a plan.
  • Standalone skill hooks read the host's JSON session identity, send PreToolUse and PostToolUse instructions through additionalContext, and throttle the progress reminder per turn.
  • feat(review): preserve findings across review reruns
  • feat(improve): cap suggestions per file after chunk aggregation
  • feat(git-providers): add public register\_git\_provider
  • Faster, more responsive chat: keep chat, dashboards, and session interactions responsive while long transcripts and disk usage are processed, with direct dashboard lookup, less cold-load work, and…
  • Reliable upgrades and recovery: keep active settings, enabled skills, and default-agent ownership in automatic updates, restore Gateway restarts after Git updates, and report outcomes with…
  • GPT-6 Astra support: select openai/gpt-6-astra with an OpenAI API-key profile or an eligible ChatGPT/Codex account, with text and image input, Responses tool calls, and supported reasoning controls.
  • Fixed daemon session creation after macOS timezone changes
  • Fixed the stable installer failing under npm 12 when resolving verified release dependencies. (Discussion
  • Added native callable tools for MCP servers supplied by ACP clients.
  • fix(desktop): keep packaged frontendDist relative so Windows embeds assets (dad5a33865fc81a2e55b3b60746632f615ec1e3a)
  • fix(sidebar): simplify unread indicators and emphasize priority activity (f038cbbb0d4092a72ffd93f17916f84d2b39bb43)
  • feat(buzz-acp): update base prompt; add buzz context and skills to Pi agents (4beffef6979347f7ebdc760705865bdd04d23508)
  • feat(code_ast): wire tree-sitter Lua grammar for syntax-aware splitting
  • feat(engine): give target action sinks value-based identity
  • fix(targets): reconcile schema changes during full reprocess for neo4j and falkordb
No release notes published.
bb
  • bb desktop 0.42.1
  • 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
  • Faster Codex cost-history refreshes: skip raw token-history reads for unchanged sessions while preserving exact pricing, reasoning totals, and fork accounting.
  • Cached custom menu-bar text: reuse plain text layouts while preserving native highlighting, spacing, display scaling, and colored emoji.
  • More reliable usage displays: show exhausted automatic quotas correctly, accept Kiro plan summaries, and recover rejected Kimi web sessions.
  • fix(cssom): match empty media query lists
  • fix(cssom): normalize quoted font family names
  • fix(cssom-view): zero detached window screen geometry
  • GPT-6 Astra — Available through OpenAI API keys and OpenAI Codex subscriptions. See API Keys and OpenAI Codex.
  • Added GPT-6 Astra for OpenAI API keys and OpenAI Codex subscriptions.
  • Added five-times-faster mouse wheel scrolling while holding Alt in fullscreen mode
  • For the installer path, run CodeWhaleSetup.exe; it installs
  • Extract the archive for your machine: codewhale-windows-x64.zip or
  • Double-click codewhale.bat (not the raw .exe) to launch
  • ZeroRelay and ZeroRouter arrive together: ZeroRelay provides blind forwarding with native mTLS enrollment, while ZeroRouter adds a first-class hosted routing preset and public model catalog.
  • Chat becomes more capable: the dashboard can keep several conversations open per agent, open the same agent in multiple tabs, and attach or drop images through the new upload path.
  • Provider support expands: Anthropic gains opt-in live thinking display, Grok Build ACP and Atlas Cloud become first-class providers, and compatible providers gain stronger response, attribution…
Tutti
  • 修复了 Agent 与 Codex CLI 版本不兼容的问题,确保使用当前支持的 Codex CLI 版本
  • 改进了 macOS 打包流程,将构建环境固定到稳定 runner,提升打包可靠性
  • Fixed Agent compatibility with Codex CLI by requiring the current supported Codex CLI version
  • Conversation content is no longer emitted as span events — gen_ai.system.message,
  • Realtime inference moved off agent_turn onto a new realtime_inference child span,
  • gen_ai.provider.name is normalized to the registry spelling. Providers outside the
  • Added /model <name> <prompt>, which runs a single prompt on another model and then returns the session to the model it was using, so you can consult a different model mid-conversation without…
  • Added a conversation_title field to the JSON payload passed to custom status line and window title scripts, so they can show the active conversation's name and pick up renames immediately.
  • Added an agents list to custom agent Markdown frontmatter, letting an agent declare the subagents it depends on using the same workspace-relative, absolute, and agent-relative path rules as skills.
  • feat(viewer): mark comments as fixed/ignored and hide them
  • fix(deps): bump grpc, fast-uri and browserslist to patch high-severity advisories
  • refactor: retire per-file naming left over from grouped review
  • fix(agents): stop parents waiting on finished non-interactive children
  • fix(cloud): print top-up instructions on 402 and guide oversize or ar…
  • fix(viewer): show stopped run status
  • fix(scheduler): use valid task kind in store tests (b343b6e03) — Anwesh
  • fix(scheduler): make task-store writes crash-safe, preserve unreadable stores (0ec74cb8a) — ap9
  • feat(scheduler): add scoped read-only GitHub CI health skill (2892072c9) — Ceren Camkiran
  • Add pydantic_ai.prices.update_in_background()
  • Handle barge-in in the session: handle_barge_in=True, interrupt(played_bytes=...), and played_audio_bytes
  • Add provider_factory to infer_realtime_model

Sep 4, 2026

OpenCode
  • Allow Codex OAuth model filtering to recognize integer GPT versions like gpt-6
  • Fixes issue of gpt-6-astra not showing up for openai subscription users
  • @Peter267
QM
  • Pick GPT-6 Astra from the model registry and picker.
  • Slack cron deliveries now carry the cron's title and a link to its settings.
  • Approval fixes in the web UI: repeated submissions stay isolated, and the input lock releases at run handoff.
jcode
  • Linux x86_64: available
  • Linux aarch64: available
  • macOS Apple Silicon: available
  • Code export now works on mobile runs — test runs against mobile targets generate exportable Python automation code, the same way web runs do.
  • JavaScript export is blocked on mobile — passing --language javascript on a mobile export now fails immediately with a clear message instead of silently producing broken output.
  • The Python-only constraint is visible — the export UI surfaces the mobile limitation so you know what to expect before you run.
  • Version 0.2.1
  • Build the phones into the nightly and the releases
  • Export at the project's own aspect
  • 1.x rollback restored. The pre-migration safety archive is now taken before the SQLite schema is migrated, so it is a genuine pre-2.0 recovery point a 1.x binary can reopen — the documented…
  • Typed edges work on OpenAI-family providers. relations was an open map that OpenAI strict mode silently emptied.
  • bootstrap reliability. Tolerates a chunk that returns no pages; retries a chunk on transient LLM errors — 5xx/429/timeout — instead of discarding the whole run.
  • Added an "Organization policy" line to /status and claude doctor that says why your organization's policy could not be loaded, such as a proxy not passing the endpoint through
  • Added bashOutputMaxChars and taskOutputMaxChars settings to raise how much command and background-task output Claude receives inline before it is saved to a file, up to 128K characters
  • Added --append-subagent-system-prompt-file to read the subagent system prompt from a file, for prompts too large to pass on the command line
  • Redesigned Skills and Connectors
  • Browse and manage Skills in a library-style dashboard with dedicated detail views.
  • Inspect source and access details, enable or disable Skills, and review their files from one place.
  • More reliable Codex cost-history catch-up: avoid repeatedly rediscovering completed work and clear abandoned Refreshing activity after account or settings changes.
  • Clearer Codex extra-credit accounting: keep spend, monthly limits, and purchased balances consistent, including confirmed zero balances.
  • Stable cost-chart navigation: switching between Token and Cost preserves menu position and scrolling, even with tall charts.
  • Fixed AltGr / diacritic input being swallowed by the integrated terminal on Windows.
  • Fixed goToDefinition opening out-of-worktree entries
  • Fixed dev container environment variables being logged in full
Kilo Code
  • #13796 626d1cc - Keep codebase indexing status current and support workspace paths with non-ASCII characters.
  • #13795 2fd7123 - Pause Changes, Review, and chat badge polling while their panels are hidden, and reuse unchanged file counts to reduce CPU use in large repositories.
  • Docker image now installs the 'dlt' extra so common data-ingestion tools work out of the box.
  • API add endpoint accepts raw_data strings and git repo URLs to simplify adding datasets programmatically.
  • Multiple fixes to migrations, logging, and provenance to improve stability during upgrades and concurrent operations.
Langfuse
  • feat(experiments): rebuild the results grid around the score columns
  • feat(experiments): a worse-than-comparison filter and a score-by-run matrix
  • fix(web): Truncate SelectInput text
  • chore(release): release Buzz Desktop version 0.5.21 (75f101d8b4f5b4b26f9891b73b87c67fedac25a0)
  • fix(desktop): wrap message tables within the available pane (6cf514ed90981d1db4acdb97a272543ecca83993)
  • 🤖 fix(desktop): harden smoke E2E tests against Bestie overlay and toast timing (7a9a5233d9d755e715be0c585cf7850e935d28cf)
  • @mastra/factory: sandbox config is now a callback (ctx => new E2BSandbox({ id: ctx.sessionId })) rather than an options object; workdir/maxSandboxes and the sandbox fleet/reattach model are removed.
  • @mastra/playground-ui: Chip, ChipsGroup, and StatusBadge were removed in favor of a single Badge component (now a <span> with updated prop typing).
  • Added a review workflow status to observability feedback.
  • Stateless MCP serving. MCPConfig(stateless=True) serves /mcp without session tracking, so any replica can answer any request and a multi-instance deployment needs no session affinity.
  • MCPTools(protocol_mode=...) selects which MCP protocol era to negotiate. Defaults to "legacy", preserving today's session-based behaviour.
  • AgentOS MCP Server Card at GET /mcp/server-card, listing the served tools and carrying a configurable name, version and instructions.
CrewAI
  • Fix legacy platform tool alias discovery
  • Update snapshot and changelog for v1.15.19
  • feat(tool): include arguments in failure details
  • fix(rules): route Objective-C++ files to Objective-C rules
  • fix(tool): repair serialized code_comment arguments instead of losing the batch
  • #3423 b199e33 Thanks @RealKai42! - Remind the model of its context budget before automatic compaction, and after compaction point it at the session's event log for exact details.
  • #3525 eba23ed Thanks @wbxl2000! - Turn-level file history is now always on; the experimental file-history flag has been removed.
  • #3516 6013658 Thanks @Grapedge! - Add an occasional session rating prompt above the input box.
  • An Install update button. Next to Connect, and in the update notice, whenever a verified
  • A download is fetched once. The staged artifact is kept under the release it belongs to and
  • The update notice no longer breaks the window. The banner between the header and the panels
  • Structured MCP outputs
  • Read-only MCP tools now advertise structured output schemas
  • Successful calls return matching "structuredContent" while keeping the existing text JSON content for compatibility
  • refactor(providers): share isAbortLikeError, compactObject and encodePathSegment
  • refactor(providers): share the 400/502 ProviderRequestError factories
  • refactor(providers): default createProviderTimeout to 30 seconds
  • Persistent Claude thinking effort — Supported Anthropic transports preserve per-turn effort and recover safely from signed-thinking mismatches. See Model Configuration.
  • Fullscreen transcript controls — Jump to the latest message from a scrolled transcript and use the embedded working indicator. See TUI Fullscreen Viewport.
  • Restorable in-memory sessions — Resume externally stored session entries through the SDK. See Session Management.
  • 54e641aaa377905a3c2be3596106cdbf4a9e934a MUL-6665: rename agent task records to runs
  • 2aad9842a9febc559666186a0e24d01c340fe933 MUL-6891, MUL-7048: fix(agents): preserve presence for hidden private runtimes
  • 2efe1b2a43f743d7c7a4edb7a70534e13496caee MUL-6984 fix(daemon): let workflow step 2 alone decide the comment scan
CodeBurn
  • macOS Apple Silicon: CodeBurn-0.9.24-arm64.dmg (or the .zip to drag to Applications)
  • macOS Intel: CodeBurn-0.9.24.dmg (or -mac.zip)
  • Windows: CodeBurn-Setup-0.9.24.exe, bundles the tray app and the new Capacity Dock. The Microsoft Store build is the signed, self-updating route; this installer is unsigned, so SmartScreen asks once.
  • /skills browser for session skills, with search and filter; shared skills sync on session start. Experimental, off by default.
  • Browser SSO completes behind split-horizon reverse proxies; opt-in browser_auth_allow_origin_rewrite rewrites sign-in/poll URLs onto configured origins
  • Discovery of Mistral's built-in web-search connector
  • fix(auth,desktop): stop moving people into the organization they joined last
  • feat(pages): usercontent origin — Cloudflare Worker + R2, view tickets, server thumbnails (spike)
  • feat(files): the files layer — schema, router, media route (SUPER-2077)
  • [Settings] Settings gains a sidebar search: find any setting row across all sections (title/description/aliases, Chinese pinyin included), jump to it with scroll + highlight, full keyboard navigation.
  • [Settings] Settings search now covers all sections with static settings rows, including MCP tabs.
  • [Translation] The translate model now has its own temperature, top-p and reasoning effort settings, editable from the translation settings panel and the translate page toolbar.
  • feat(fetch): pass through raw responses by default
  • feat(fetch): add disable-css mode
  • fix(cli): read wait scripts from stdin
  • Add OpenAI gpt-6-astra model support
  • Fix the context_subtree() exporter cache and its span-processor leaks
  • Restore the Instrumentation spec option include_model_request_parameters and sharpen the undeclared-attribute merge error
  • commit: 5b2803a7262e4c502fc37a419b14305f49c6c33c
  • source tag: beta2026-09-04_11-40-00
  • regions: cn global
Browser Use
  • Upgrade Browser Harness to 0.1.13.
  • Exact-pin all declared runtime, optional, development, and build dependencies.
  • Upgrade and migrate to MCP Python SDK 2.1.1.
  • Added half-page scrolling with Ctrl+D and Ctrl+U to the artifact viewer.
  • Added the pickerGrouping setting to /config and settings.json to configure the default conversation list view in /resume (flat or grouped by workspace).
  • Improved terminal ASCII diagram rendering for Mermaid flowcharts in the artifact viewer and conversation.
Browser Harness
  • Pin the MCP extra to mcp 2.1.1.
  • Pin release build tooling to build 1.6.0 and setuptools 84.0.0.
  • Preserve the Chrome permission and macOS accessibility reliability improvements from 0.1.12.
  • chore(monitoring): remove Watchdog (a0705aa46) — Awokoya Olawale Davidson
  • fix(shell): close the transcript holes left by inline menus (fcadc1b94) — Yauhen Bichel
  • feat(shell): quiet status bar and composer chrome (Droid/Cursor parity) (19f3f955e) — Yauhen Bichel
  • Faster worktree, renderer, editor, browser, terminal, and Git operations across local and remote workspaces.
  • More resilient agent sessions, native chat, SSH relay recovery, and Windows/WSL execution.
  • Expanded workspace, terminal, CLI, cloud, and cross-platform reliability improvements.
  • Avatar accents extract straight from your photo, Gemini 3.8 Flash joins the catalog, and the desktop companion learns hands-free dictation.
  • Your assistant dresses in your colors. Upload an avatar image and Vellum automatically pulls out its dominant accent color.
  • Gemini 3.8 Flash arrives. Google's newest Flash model joins the provider catalog across web and desktop, bringing low-latency responses and high token limits to your fast and balanced workflows.
Caveman
  • Signed Caveman runtime companions for macOS, Linux, and Windows.

Sep 3, 2026

  • Added a diff panel that opens beside the conversation in fullscreen mode and shows your uncommitted changes as Claude edits; toggle it with /diff
  • Added a likely cause for prompt-cache misses (e.g. tool definitions or system prompt changed, idle past the TTL) to /cost and the status line's prompt_cache field
  • Added /reload-plugins to headless sessions, so it appears in the Claude Code Desktop and SDK command lists
  • One SDK range-edit operation covers overwrite, append, prepend, insertion,
  • Localized edits reuse unchanged content through the shared piece-tree engine;
  • A daemon close/remount race is fixed: watcher and mount reservations retire
CopilotKit
  • Release angular/v0.5.1
  • Amazon Q for VS Code 2.7.0
  • Auto-updater for desktop application
  • Linux ARM64 desktop release packages
  • Load extensions in the background so the CLI prompt is immediately usable
Deep Agents
  • Defaulted general-purpose subagents to fork mode.
  • Defaulted the Auto classifier by provider.
  • Added support for binding conversations to a recorded workspace.
  • d8ec44919265b3641b7898a02038cbfe14590ee5 .NET: Bump package version to 1.80.1
  • 3438d882147b6e3e19a161a5f0dcb64e23e181db .NET: update SDK to 10.0.303
  • 437336175b8ef841a8ccace6c6cf909f34f0cad6 .NET: Remove retired OpenAI Assistants integration tests
  • Agent Plugins are now discovered and run by the shared Hub. Packages under ~/.agents/plugins are validated from their plugin.json, their valid Agent Skills become available to the agent, and their…
  • The "Cline Hub was updated" dialog no longer appears on every launch and reconnect.
  • Signing in now shows the device confirmation code in the app while you wait on the browser, so you can match it against the code the browser asks you to confirm — in onboarding, Account settings…
  • Diagrams in every chat: Mermaid blocks now render as diagrams in the Control UI and in the native macOS, iOS, and Android apps, with enlarge previews and a retry when a diagram fails to render on…
  • From install to chat in one prompt: fresh installs (including npx openclaw@latest) get a quick-start lane that detects existing Claude Code or Codex logins and API keys, verifies them live, and…
  • Personal skill libraries on shared Gateways: keep your own skills beside the workspace set with openclaw skills library, import them from ZIP archives, and share or publish them per identity on…
Langfuse
  • chore(pricing): flag shut-down Gemini 3 Pro/3.1 Flash-Lite previews, refresh audit snapshot
  • chore(deps): bump qs to 6.16.0
  • chore(deps): bump @humanfs/node to 0.16.8
  • Windows: the OKF migration no longer aborts on the .serve.lock file the same server holds under an exclusive lock (os error 33). The backup walk now skips it.
  • libgit2 crash-loop fixed: the OKF migration could fail with invalid object specified … class=Tree and never start.
  • OpenCode Go: requests now send x-opencode-session + User-Agent: ai-memory/<version> — required before OpenCode Go starts rejecting header-less requests on 2026-09-06.
  • Changes since langchain==1.3.18
  • docs(langchain): runnable langchain.mcp examples feat(langchain): langchain.mcp namespace, MCPAdapter perf(anthropic,langchain): omit middleware trace inputs fix(langchain): include model…
  • fix: clean up unused browser recovery launch
  • docs: retry hidden-tab scrolling after activation
  • docs: prevent duplicate tabs across harness calls
Rakazo
  • Keep macOS controls clear of the setup title
  • Move web UI to shadcn (Base UI) on shared semantic tokens
  • Fix connector card spacing and authorize hover feedback
  • Codex: let cost-history catch-up finish while active rollout files keep growing, preserving complete session and subagent accounting without publishing partial tails. Thanks @LeoLin990405!
  • Antigravity: restore token history from newer local sessions whose timestamps moved to the steps table, while rejecting missing, duplicate, or conflicting timestamp evidence instead of inventing…
  • Codex: keep each managed account's selected workspace authoritative across stacked refreshes, credits, history, menu rows, reconciliation, and System Account promotion instead of reverting to or…
  • AI: Gemini models work again; OpenCode Zen and Go added as providers; output token cap is now a setting (default 32k) and truncated replies are flagged instead of "(no reply)".
  • Docs: Word-style dark page in dark theme — print, PDF export and clipboard keep the original colors.
  • Slides: compressed embedded fonts (MTX) now decode; text fidelity fixes for default run properties, Symbol bullets, theme colors, CJK/Latin font selection and slide-number fields.
  • ci: make local plugin release notes resilient to large releases
  • fix: normalize numeric local-plugin source refs
  • Preference memory fixes
  • d276991c603e6041d20fb5c1e65df8756d35d2ae MUL-6743 refactor(issues): remove the assignment handoff note
  • d4a712abf3880dfbd3daeac5daac1bd4bfb39b6f MUL-6771 feat(cli): filter and sort issue list by custom property
  • 043121806176be751c345c7d336c37927248bb08 MUL-6833 fix(web): localize auth callback and error pages
  • Persist sessions created with a title before the first visible message
  • Hold the idle-agent reservation through the wake turn and terminal status fanout
  • Price OpenRouter @endpoint-pinned models from the pinned endpoint
  • samples: Add an end-to-end Foundry-hosted Telegram agent sample
  • agent-framework-foundry-hosting: Add explicit Agent Server or agent-owned model history selection while preventing duplicated conversation replay
  • agent-framework-core, agent-framework-foundry: [BREAKING] Restore sequence-only agent middleware inputs and remove the experimental agent-hooks core extra
  • The OpenAI Chat Completions and Responses converters no longer raise on an assistant message with no content parts, which a Chat Generator returns when it discards a malformed tool call.
  • The Hugging Face message converter no longer raises on an assistant message with no content parts, and sends it with empty content instead.
  • Give one-off capabilities a default id and a combine rule for repeats
  • Add context_window to ModelProfile and context_window_used to RunContext
  • Add profile flag to reject streams without finish_reason
  • [Cherry Cloud] Sign in to Cherry Cloud for entitlement-aware Agent models, automatic local Gateway startup, quota-aware model selection, and durable Product Session handling.
  • [Context Management] Configure the compression trigger threshold per assistant or globally in Settings.
  • [Mini Apps] Improved installation and navigation with drag-and-drop packages, stable tabs and icons, and transparent Translate and Painting views.
  • Full Changelog
  • Log in with your Claude/ChatGPT subscription: per-user accounts drive each turn, plus dynamic OpenRouter catalogs and a model picker built for large ones.
  • Search across everything QM can see: principal-aware core search lets the agent reference conversations across scopes, and ⌘K search got much faster.
  • Deploy to Porter and Fly with new providers, guided onboarding, and safe-by-default app publishing (path-based serving with zero config, one-var subdomain upgrade).
  • Protected notebook kernels launch directly on Windows. Windows kernels start under the network sandbox without the indirect launch path that could leave them unprotected or failing to start.
  • File exports publish atomically. Exported files appear only once fully written, so a failed or interrupted export can no longer leave a partial file behind, and package exports keep valid…
  • Generated image previews are restored. Images the agent generates show their previews again instead of falling back to placeholders.
  • Security: Fixed a security vulnerability where Chat file uploads could spoof APPLICATION attribution and be read anonymously, allowing content injection into the application namespace.
  • Security: Fixed a security vulnerability where document/paragraph migration only validated source knowledge base permissions, leading to cross-tenant content injection and RAG poisoning.
  • Security: Fixed a security vulnerability where XLSX embedded image decompression lacked a pixel count limit, allowing authenticated users to trigger memory-exhaustion denial of service (DoS).
  • 修复了外部 Claude 命令丢失的问题,自定义命令现在可以正常保留
  • 改进了 Claude 二进制文件的安装与状态检测流程,Agent 启动更可靠
  • 优化了 Windows 平台上的 Agent 运行环境准备流程
  • feat(raw): capture raw LLM traffic via SDK middleware (opt-in)
  • fix(tool): report tool call failures
  • fix(diff): propagate untracked file listing errors
  • Self-host a multi-user QwenPaw Hub with local-process or Docker runtimes, workspace-level access controls, credential management, and reverse-proxy support
  • Deploy QwenPaw Data as a native PawApp with a durable analysis workspace, packaged runtime, and one-command Docker Compose example
  • Add an intelligent mail assistant with real-time monitoring, search, thread-aware actions, sender allow/deny rules, and human approval for unknown recipients
  • commit: 2a039176935c71ee0e1a6377defccbea738b034b
  • source tag: beta2026-09-03_11-10-18
  • regions: cn global
  • feat(dsh-plugin): add Browser tool parity and dedicated Harness skill
  • feat(vom): refactor semantic structure generation for grouped observations | 重构语义结构生成,支持分组式 VOM 观察
  • fix(dsh-plugin): stabilize skill loading and browser sessions
  • Added an opt-in workspace-grouped view to the /resume conversation picker, allowing users to toggle between a flat list and conversations grouped by directory with Ctrl+F.
  • Added Gemini 3.8 Flash to the model catalog when connecting with a GEMINI_API_KEY.
  • Changed custom agents defined in Markdown to inherit ambient skills, rules, and subagents by default, matching the configuration of default agents.
  • Faster worktree switching, Git metadata scans, terminal startup, and renderer updates across local, WSL, and remote workspaces.
  • More reliable native chat and agent sessions, including large command results, live tool progress, image attachments, and delivery recovery.
  • Broader SSH, Windows/WSL, GitLab, updater, startup, and release reliability improvements.
  • fix(messaging): remove outbound send approval gates (34b77c820) — Vincent Hus
  • feat(shell): polish terminal identity and startup animation (0c55702fb) — Vincent Hus
  • docs: drop the orphaned Hermes surface-attribution runbook (b9e75be67) — Vincent Hus

Sep 2, 2026

Planning With Files
  • The PostToolUse nudge never reached the model (closes #239, reported
  • The nudge is now once per turn rather than once per tool call. The string is a constant, so every repeat after the first carries no information: it names no tool, no file and no phase.
  • Bash came off the PostToolUse matchers. ls, git status and grep were tripping a "you changed something, record it" reminder.
  • bounded initialization with eight existing producers, four fixed slabs, and
  • no canonical object-segment spool or parent payload copy on the admitted
  • exact operation-local metadata reuse without a persistent cache;
  • Updated bundled Claude CLI to version 2.1.259
  • Added managedMcpServers managed setting: organizations can provide HTTP/SSE MCP servers to every user (same entry shape as .mcp.json); entries that name a command to run are skipped
  • Added --permission-prompts none for unattended headless hosts: anything that would prompt is denied automatically while the active permission mode (including auto mode) keeps deciding
  • Added recognition of glab mr create/merge/close/reopen/note/update so GitLab merge requests show as MR !N in the collapsed tool summary and refresh the footer MR badge
  • Default provider header timeouts to five minutes so slow model startups fail less often.
  • Default streamed chunk timeouts to five minutes, with false supported to disable them.
  • Let Anthropic thinking.blockBinding opt out via config when you need to keep the provider default. (@dkindlund)
  • The Loop. Goal with the exit removed: every finished turn gets its next message until you switch it off.
  • Improved stability and reliability through auto reloads. A chat that goes silent, shows an error or loses its tab is reloaded or reopened on its own evidence, and a turn that is still running is…
  • macOS computer use. The optional Desktop connector (screen, mouse, keyboard, clipboard) now ships on macOS, off by default.
  • Fixed the inline assistant doing nothing when the model emitted reasoning before its tool call. (#61220; thanks aviatesk)
  • Fixed some Pyright and BasedPyright settings not being applied properly. (#62673; thanks lingyaochu)
  • Fixed bookmarks not disappearing from the gutter immediately after clearing all bookmarks. (#62847; thanks austincummings)
AI Memory
  • One lint report per project. The lint pass supersedes a single _lint/report.md (history stays in the version chain) instead of writing a dated page every day — a long-lived store had accumulated…
  • Knowledge first in the web UI. The project view moves machinery (lint reports, session captures, monthly logs, bundle indexes) into a collapsed System sidebar section, and Recent Activity lists…
  • Calmer homepage. The "LLM-optimised memory" explainer is dismissible (remembered per browser), and the redundant always-on backup banner is gone — the one-time migration dialog and ai-memory…
  • Renamed the SDK subagent mode from handoff to isolated.
  • 1aae374 release(deepagents): 0.7.13
  • 15454a8 fix(sdk): rename handoff subagent mode to isolated
  • Artifacts you can edit. Markdown, plain text, scripts, and source-code artifacts and uploads can now be edited as raw text — every save publishes a new version that preserves its source lineage…
  • NVIDIA Build joins the built-in providers. A curated NVIDIA catalog with six agent-capable models, led by NVIDIA Nemotron 3.5 Lightning.
  • Runtimes you can repair. App-managed Python and R runtimes can be safely reinstalled from Settings, and a new toggle decides whether the agent may create runtime environments on its own.
  • skills: reference new cloud cli
  • MCP call errors no longer quarantine the connection
  • fix(tui): run environment and model checks on the no-target start screen
  • Docs: AI edits now preserve formatting — rewritten paragraphs and tables keep their fonts, widths, borders and shading.
  • Docs: comment balloons and page borders now appear in print preview and PDF export; large-document export is faster and no longer stalls or produces blank pages.
  • Docs: layout fidelity — cross-page tables, per-section margins, CJK typed grid and punctuation compression, VML pictures, doughnut charts.
  • Maintenance release — packaging/pipeline fixes only, no user-facing changes.
Caveman
  • Signed Caveman runtime companions for macOS, Linux, and Windows.
Ziva
  • Godot 4.2+
  • Active Ziva account
Mem0
  • OSS notices: Notice configuration now comes from a static, cacheable repository file with a bundled disabled fallback and deterministic rollout assignment, instead of calling PostHog's…
  • fix: prevent OTel prototype-chain clobbering on v3
  • fix(blobstorage): suppress sub-second remainder chunk to prevent silent key collision (v3 backport of
  • security(api-keys): gate organization API key creation on the admin-api entitlement (v3 backport of
  • #13501 9231c25 - Preview local applications in Agent Manager with embedded developer tools, grouped diagnostics, and review-style element feedback for precise frontend changes.
  • #13556 52d4247 - Edit queued messages in VS Code before they are sent, while preserving their text and attachments.
  • #13565 49692a6 - Advertise optional instance kind, process start time, and current Git branch in CLI heartbeats.
  • feat(dsh-plugin): add Browser tool parity and dedicated Harness skill
  • feat(vom): refactor semantic structure generation for grouped observations | 重构语义结构生成,支持分组式 VOM 观察
  • fix(dsh-plugin): stabilize skill loading and browser sessions
  • 2e424e06a1d9077c9a09f3e439f6dae4c7416a12 MUL-5154: fix(agent): surface Kimi provider API errors
  • 5bc021c369ca8ad5d24ccdf7a02588a54b2e5835 MUL-6542 fix(mobile): stabilize recycled markdown layout
  • 11861145abc59a0d39c8c8f24ad837d4584e664f MUL-6758 feat(core): add truncateWithEllipsis string helper
  • 3659a08ac65b15c1e6c67378df1377e75e73f8f9 ci: release from a manually pushed tag instead of semantic-release
  • 0c426322c176a773fc97f2c0c0c5c6c28a688e38 feat: add OIDC provider type
  • babdb25bf438b55d1bd53a4064194ba2c3978b14 feat: add a command palette, bulk row selection and inline form validation
Kimi Code
  • #3469 979baad Thanks @sailist! - Fix the condition for showing the kimi-cli migration prompt.
No release notes published.
  • commit: 5a595a25fffc369bce61852217a40cdf37b373e8
  • source tag: beta2026-09-02_15-03-04
  • regions: cn global
Cline
  • ClinePass is now surfaced across the app: a card on the account page describing what the plan covers, a hint in provider settings, and a banner on the home screen. Dismissed banners stay dismissed.
  • Fixed the background Hub process ballooning in memory during long sessions. Session status updates broadcast a full copy of the conversation transcript to every connected client, so on a large…
  • Hook scripts that fail to spawn no longer crash the extension's core process and take the running task down with them.
  • Initial release
  • release(langchain): 1.4.0a4 test(langchain): cover mixed-era ClientGroup and group elicitation Update libs/langchain_v1/langchain/mcp/adapter.py fix(langchain): drive MCP elicitation via member…
Paperclip
  • Onboarding no longer gets stuck on the organization step. On the first onboarding screen ("What is the name of your organization?"), clicking Continue created the organization on the server but…
  • No migrations, no configuration changes, no API changes. Upgrade in place.
  • Faster worktree creation and smoother performance across terminals, large diffs, Git operations, and the app UI.
  • A more capable native chat experience, with image paste and attachment previews plus improved send and response reliability.
  • More reliable remote/SSH and Windows/WSL workflows, including safer terminal recovery, path handling, and updater downloads.
WrenAI
  • pydantic: forward session properties to the engine (9bd52fc)
  • Improved /mcp panel navigation so up and down arrow keys strictly navigate between MCP servers and plugins, left and right arrow keys cycle available actions, and Esc exits action mode or the panel.
  • Fixed duplicate agent entries appearing in the /agents picker when an agent with the same name is discovered across multiple configuration sources or plugins.
  • Fixed tool initialization and agent startup failing when launched from an inaccessible or deleted working directory by using absolute URI schemes for in-memory parameter schemas.
  • fix(test): patch GitHub package API in account-gate tests (9c7d44589) — Yauhen Bichel
  • fix(shell): clear a stale pinned plan when a new turn starts idle, always cap command output to a short head in the transcript (bce5b128b) — Yauhen Bichel
  • feat(shell): collapse tracebacks, frame output, soften text further (9170c5430) — Yauhen Bichel
  • Preserve confirmed finding identities across scans and comparisons, and show related findings with their reasons while keeping distinct findings separate.
  • Improve deduplication with separate screening and pair reviews, validated pair assignments, and groups that respect explicit DISTINCT decisions. Invalid submissions receive one corrective turn.
  • Generate synthetic Standard scan results with scan --mock or the SDK's mock: true, without authentication or model calls. Mock results support normal reports, exports, history, and reruns.
  • Added active-turn steering for pi agents so new messages reach the running turn without interrupting it
  • Added declared build commands and monorepo source paths for Git-hosted plugins
  • Reduced JS stalls while watching a streaming agent on mobile

Sep 1, 2026

PageIndex
  • 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).
Graphiti
  • 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
Claude Agent SDK
  • Updated bundled Claude CLI to version 2.1.258
Claude Code
  • Fixed Claude Code failing to launch on macOS 12 (Monterey), a regression introduced in 2.1.255
  • Fixed remote and scheduled sessions failing with "user messages must have non-empty content" after a re-sent permission approval could not be applied
  • Discord goes official, Claude Fable 5.1 and Gemini 3.7 Flash arrive, and guardian approvals land in one tidy inbox.
  • Discord is an official channel. Connect your bot through a setup wizard that walks you through every step and hands identity verification to your assistant with one "Verify me." Your assistant…
  • Claude Fable 5.1 and Gemini 3.7 Flash are here. Both bring 1M-token contexts: Fable 5.1 across Anthropic, OpenRouter, and Vercel AI Gateway with adaptive thinking, and Gemini 3.7 Flash straight in…
  • 📥 Import your existing local CLI sessions: pull recent Claude Code, Codex, and other CLI conversations straight into Omnigent from the web UI, keeping each one's original title.
  • 📁 Projects now drive new sessions: name a project_id at creation to inherit the project's agent, workspace, and git defaults, move a session into a project from its title's folder tag, and…
  • 📱 Refreshed mobile and wide-screen UI: a ChatGPT-style mobile drawer, full-screen mobile search, better use of ultra-wide displays, appearance export/import, a heavier code-font option, a markdown…
  • Claude and Vertex AI: reduce background CPU spent reading transcript metadata and looking up model prices during local cost scans, preserving provider detection and token/cost totals.
  • Local costs: skip unnecessary parsing work for discarded oversized log records, preserving complete-record validation and cost totals.
  • Codex: avoid repeated full scans after trace-log pruning, while retaining the latest validated cost history through temporary trace-database failures. Thanks @brzvsk!
  • bb desktop 0.41.0
  • Claude 5 sessions now tolerate stale thinking blocks instead of failing after prompt or tool changes.
  • Bedrock GPT-5.6 models now accept none reasoning effort.
  • Bedrock reasoning and replay handling is more reliable. (@pengzh1)
Langflow
  • feat(models): add pluggable provider policy
  • feat: make langflow-core the shared bundle-free distribution
  • feat: add scoped authorization visibility prefilter
  • Fix LiteLLM cost model resolution
  • feat: place caller-provided files into the sandbox workspace (extra_files, --workspace-file)
  • feat(reporting): contextual CVSS breakdown on dependency reports
Prime Agent
  • Fixed a v0.9.0 regression: the agents view's Inactive section was empty on a fresh view until a search was typed. The saved-session catalog now loads (progressively) when the view opens.
  • Added subagent conversation forking in the SDK.
  • Fixed glob sorting to tolerate missing modified_at values in the SDK.
  • a197807 release(deepagents): 0.7.12
Kilo Code
  • Keep Agent Manager run/action buttons available while the IDE is indexing, instead of disabling them and showing "waits for analysis" tooltips.
  • registerComponent — an Angular component can now be a tool the agent calls.* Previously a component could only be rendered *in response to a tool call; it can now be registered as the tool itself.
  • Managed Intelligence entitlement authority. CopilotKit exposes two new readonly signals sourced from the connected runtime's /info response — runtimeEntitlements and runtimeEntitlementRetryPending.
  • Human-in-the-loop results resolve without the bus envelope.
  • Added Claude Fable 5.1 with a 1M context window
  • mode agent now pauses on every question — kane-cli no longer auto-accepts recommended actions; every decision waits for your explicit input, so nothing runs without your approval.
  • Variable placeholders no longer produce trailing garbage — typing into a {{var}} field no longer appends stray characters after the substituted value.
  • if_else steps with no sub-checks are now anchored correctly — branches authored without nested checks previously could be skipped or mis-executed; they now behave as expected.
  • MCPAdapter adapts any target fastmcp.Client accepts — a URL, a local script, an in-process server, an MCPConfig naming several servers, or a pre-built client — as well as a FastMCP ClientGroup for…
  • MCPAdapter.list_tools(*, cache_mode="use") discovers and adapts tools, with optional client-side response caching (SEP-2549): use serves a cached list within the server's TTL hint, refresh…
  • as_langchain_tool(tool, client, *, elicitation=None) converts a single MCP tool for callers managing their own client.
  • Your Home agent, beside your work: open Home in a right or bottom dock with Cmd/Ctrl+Shift+H, keep your current page in view, and preview or remove its work-context snapshot or attach selected…
  • A desktop companion for Linux: install the .deb or AppImage on x86-64 Linux, connect to a local or remote Gateway, and open Quick Chat from the system tray or an X11 keyboard shortcut.
  • Start work without switching pages: create and run a background session from New Session, keep its selected local, cloud, or paired-device placement, and open it from the completion notice.
  • #720 6ba64c9 Thanks @colifran! - feat: add durable page-level resumability across local, CI, and host runs
  • #743 0492b5e Thanks @Christian-Sidak! - fix: set maxTokens on Bedrock Converse API calls to avoid 4096-token default cap
  • #414 e280c17 Thanks @bikeusaland! - fix: correct git-repo incremental diff and isolate connector ingestion failures
No release notes published.
  • feat(cmd, viewer): open viewer URL in default browser
  • fix(rules): route .cxx and .hxx files to C++ rules
  • fix(rules): route JavaScript module files to JS rules
  • Custom Babel plugins: the desktop app now picks up plugins from a project's own Babel config (babel.config.js / .babelrc) and runs them alongside the built-in transforms.
  • New useResolution hook (@diffusionstudio/jsx): a reactive accessor for the host's rasterization density, so a <surface> can size its bitmap to stay sharp at any export resolution.
  • New examples: TypeGPU caustics (examples/10-typegpu.tsx) and Redraw integration on a <surface> with resolution-aware rendering (examples/11-redraw.tsx), plus expanded JSX lifecycle and…
  • Turn-completed notifications now include the conversation's name, so you can tell at a glance which conversation finished instead of seeing identical "finished responding" messages…
  • When a conversation pauses waiting on you — a tool permission request or an agent question (AskUserQuestion) — the sidebar now shows a distinct attention icon instead of the generic generating…
  • A desktop notification also fires for the waiting state (when the window is unfocused), naming the conversation when its title is known, so a paused agent no longer waits unnoticed.
Moli
  • ~10% smaller binaries through build optimizations, dependency cleanup, and the migration from OpenSSL to aws-lc-rs.
  • ~20% lower memory usage through more compact DOM, renderer, task queue, and style data structures.
  • Faster rendering and DOM operations, with optimized layout probing, font resolution, style invalidation, DOM mutation, fragment parsing, and custom property inspection.
  • feat(noramlized-io): hook up parser trace io behind feature flag
  • feat(api): add self-hosted API specification renderer
  • feat(trace-table): add cached input tokens and cost
  • correct --with-stat flag description (ebd79ac)
  • validate namespace exists before running core analyzers (c7e91c0)
  • deps: update docker/setup-buildx-action digest to 37fe631 (92c0091)
  • Embedding failures are surfaced, not masked. Ingestion now reports embedding failures instead of returning success.
  • Embedders raise EmbeddingError instead of returning an empty vector on failure. Calling a vector DB's search() directly now raises where it previously returned [].
  • AWS Bedrock embedding failures raise EmbeddingError, not ModelProviderError. Existing except ModelProviderError handlers around Bedrock embedding will stop catching — switch to except EmbeddingError.
  • Memory Recovery and Capture Boundaries: Fixed dirty-data recovery, orphan trace writes, and LLM empty-response retry stability issues.
  • Memory Recovery and Capture Boundaries: Fixed dirty-data recovery, orphan trace writes, and LLM empty-response retry stability issues.
  • Hermes and Bridge Experience: Combines Hermes, provider, and bridge-flow improvements to reduce startup and connection friction.
  • 修复了认证过程中托管运行时丢失的问题,确保 Agent 认证后仍能正常使用托管运行时
  • 更新了 Agent 故障排查文档,新增 Agent 提供者配置指南
  • Fixed an issue where the managed runtime was lost during authentication, ensuring Agents continue to work correctly after auth
  • replace the Ant Design console with the shadcn/ui frontend (38ecc6e)
  • web/ now holds the shadcn/ui frontend built with Vite. Its
  • Faster, smoother everyday use through Markdown, worktree, terminal, and renderer performance improvements.
  • More reliable agent sessions, native chat, remote/SSH connections, and Windows/WSL workflows.
  • Refined browser, automation, mobile, onboarding, and cross-platform infrastructure.
  • Improved /model <name> autocompletion to accept the proposed model name ghost text with Tab.
  • Reduced subagent streaming overhead by sending subagent trajectory metadata once per subtrajectory instead of with every step.
  • Fixed commands with subcommands (such as models or agents) hanging on an inherited, unclosed standard input pipe instead of executing immediately.
  • Added experimental WebMCP support for discovering and invoking tools provided by the current page, including frame-aware tool selection, detached results, cancellation, bounded metadata and output…
  • Added a WebMCP generation skill that helps agents expose existing page workflows as validated page tools while recording safety constraints, deterministic checks, and fallback behavior.
  • Updated the eve integration to 0.47.3 and raised its compatibility floor to 0.39.1.
  • npm: bash npm install -g base44@latest
  • Homebrew: bash brew install base44/tap/base44
  • Version: 0.1.14 NPM Tag: latest
  • One SQLite Store for LayerStacks, immutable Layers, named Branches,
  • Zero-copy Branch Fork from an immutable Layer or eligible Commit.
  • Ephemeral materialized or real-FUSE Workspaces with explicit Commit and End.
  • Add glm-5.3-flash and rework the Z.AI test suite onto cassettes
  • fix: preserve all conditions in pruned span queries
  • Accept and map Z.AI's non-standard finish_reason values
  • feat(shell): fix reply table rendering and add focus-aware sound notifications (8eaf2bf2d) — Yauhen Bichel
  • fix: canonical main-build versions and raw ANSI prompt rows (ca4d652a1) — vaibhav upreti
  • fix(shell): paint submitted prompt body with a concrete Rich Style (027380248) — vaibhav upreti

Aug 31, 2026

No release notes published.
  • Marketplace is now a two-pane explorer: a browsable list on the left and full catalog metadata for the selected item on the right, with category tag filters that collapse behind a "more" toggle
  • Stopping a session now actually stops everything it started. Stop stays available while child agents are running, and an abort propagates to delegated subagents and to teammates instead of leaving…
  • Fixed the ask-a-question tool's option text overflowing instead of wrapping
  • a83480eb156fd75cd96b4fecaf3cc40570bb3142: feat: centralize tui option defaults and extract exit banner rendering (@taciturnaxolotl)
  • ee455a9966c33086494d897b2c8b4b73d70c0ff5: fix(config): resolve provider base URL before custom validation (@joestump-agent)
  • 1e5be92e2ac9100959b34f90347522b2557f477a: fix(mcp): bound MCP init waits so a wedged server can't blank the app (@joestump-agent)
  • Release v1.70.0
  • Updated bundled Claude CLI to version 2.1.252
  • Fixed Bash commands failing with "task output swap refused (tasks dir moved or linked)" on some Macs
  • Fixed "always allow" not saving in a project that has no .claude/settings.local.json yet
  • Fixed Remote Control sessions hosted by Claude Desktop or VS Code stalling for minutes after a tool finished when the connection to claude.ai was degraded
  • Bot Mode — your agents become a society, built in — Bot Mode is now a bundled, default-on part of the desktop app: every agent profile gets a name, a deterministic avatar face…
  • hermes peer — bot-to-bot DMs between your agents — Any Hermes agent can now message any other by handle, across profiles and gateways, from the CLI or from inside a conversation.
  • Cron jobs that remember — Scheduled jobs stopped being goldfish. Cron agents now load and update persistent memory like every other agent, continuity=true carries each run's output into the next…
  • .NET: Bump AWSSDK.Extensions.Bedrock.MEAI from 4.0.6.10 to 4.0.101.8
  • .NET: Stabilize Foundry recovery tests
  • .NET: fix: preserve Responses logprobs field
Langfuse
  • feat: add API spec previews for pull requests
  • feat(traces): link evaluator execution traces
  • fix(evals): avoid reasoning collisions in structured output
  • Changed the project license to Apache-2.0
  • Changed Cmd/Ctrl+E to toggle Explorer visibility without selecting a view
  • Changed the default mobile content text size from 15px to 16px
  • feat(py): support multiple continuation inputs (4705f0c16)
  • feat(ts): add backgroundTasks to Agent (c1d5e4ef2)
  • feat(mcp): added server/discover for SEP-2575 (00dd3c560)
  • feat(xlsx): report numeric display overflow
  • @y4ho0 made their first contribution in
  • SKY-15908: Let Copilot edit pending workflow drafts (90dee482f)
  • Log page-analysis deadline exhaustion at warning consistently
  • Stop the CDP proxy aborting healthy clients with WebSocket 1013 under upstream bursts
  • Run and manage build/run configurations directly from the worktree editor.
  • Render all Mermaid diagram types in chat, not just flowcharts.
  • Add worktree transfer safety checks, unified change summaries, setup scripts, and GitHub integration controls.
  • Move between projects in seconds. The workspace project menu now lists your other active projects with title and description previews, so you can switch projects without leaving the session you…
  • Notebook code stays inside your network boundaries. Notebook and compute runtimes can only reach Open Science defaults and the domains you approve, and when code tries a new destination, you…
  • Enter credentials once, reuse them everywhere. API keys, access tokens, and OAuth sign-ins can be stored device-wide and bound to any custom connector's environment variables, headers, or sign-in.
  • Hardened dashboard origin validation and reverse-proxy access with same-origin provenance enforcement that defends against DNS rebinding, form/header smuggling, and cross-origin requests.
  • Fixed root remote CDP WebSocket URLs with query strings to insert the required slash before the query while preserving the encoded query
  • First-time setup is now a real choice
  • Before the first ChatGPT connection, C2C asks once: AI automated setup (preview) or guided manual setup
  • The choice is remembered on this machine and reused across workspaces
  • Local costs: reduce background CPU spent parsing native session timestamps and validating appended Codex history, preserving timestamp precision, daily totals, and fork accounting.
  • Codex: reduce repeated decoding of cached cost history during scans, while preserving stored totals and checking for database and filesystem changes.
  • Claude and Vertex AI: reduce CPU spent reading local transcript metadata, preserving provider detection, Unicode handling, and token/cost totals.
  • Running an attestation helper inside .planning/<slug>/ now updates the slug .attestation instead of creating a legacy file.
  • macOS system aliases, Windows active-plan pointers, Unix PowerShell refusal, and trusted interpreter fallbacks now behave consistently across the supported matrix.
  • PyPI: bash pip install crawl4ai==0.9.3
  • Docker: bash docker pull unclecode/crawl4ai:0.9.3 docker pull unclecode/crawl4ai:latest
  • Note: Docker images are being built and will be available shortly. Check the Docker Release workflow for build status.
  • npm: bash npm install -g base44@latest
  • Homebrew: bash brew install base44/tap/base44
  • Version: 0.1.13 NPM Tag: latest
  • 7c0e750452599ef5ead020c8c787b32c512f2cd4 MUL-6569 feat(wecom): give every undelivered reply a reason an operator can count
  • 1dd6b9ecdbaa991bfd51b48ef5c269056045d547 MUL-6570 feat(wecom): route a reply to the replica that holds the bot's socket
  • 64ec7f54163d918d5d7fd4dcae857f241b7842d0 MUL-6737: fix(timeouts): close the two gaps the 2h inactivity budget left open
  • mempalace search reuses a healthy live Hub. On the reported 4.99-million-vector palace, two direct CLI searches held about 18.6 GB of private RSS.
  • Hub reads run concurrently. Searches no longer serialize behind one exclusive HTTP dispatch lock; reads share a writer-preferring lock while mutations remain exclusive.
  • Search results are clearer and faster to enrich. Public similarity is the raw vector score, closet boost remains in effective ranking, source lookups are indexed, and duplicate enriched passages…
  • feat(allowlist): add Pug template support
  • feat(grouping): skip the grouping LLM call for small change sets
  • feat(allowlist): add Verilog, SystemVerilog, and VHDL support
  • commit: 2310a13333f5c5f98a77357651208a339f639669
  • source tag: beta2026-08-31_16-26-37
  • regions: cn global
  • More reliable agent sessions, native chat, terminal recovery, and workspace state across local, remote, and SSH connections.
  • Smoother onboarding and everyday editing with faster rendering, improved review workflows, and clearer agent status.
  • Korean localization for onboarding and the Agent Dashboard, plus mobile, Windows, GPU, and CI reliability improvements.
  • chore(desktop): bump version to 1.25.0
  • feat(marketing): tier icons and factory backdrop for leaderboard pages
  • fix(panes): keep resize handles from painting above dialogs
  • hotfix: adapt to AnyDoc 0.2 document model
  • Find past conversations: search visible conversation text by exact words or phrases and reopen the surrounding messages from a matching result. Thanks @hercial61.
  • Sessions beyond your Gateway: run work on paired devices or cloud workers, move the session workspace with it, and reuse warm machines and project seeds for later cloud sessions.
  • Follow work as it happens: keep a durable session progress card through reloads and follow subagent activity and accumulating edits across web and native chat.
  • feat(runtime): propagate request cancellation
  • fix(ssrf): use shared assertPublicHttpUrl on self-hosted base URLs
  • fix(ssrf): keep DNS validation on region- and credential-resolved hosts
  • FRED macro requests now pin the data vintage to the as-of date, so a backtest no longer sees later revisions.
  • Social sentiment (StockTwits, Reddit) is trimmed to the analysis window instead of showing today's chatter, via one shared UTC window rule reused by news.
  • Decision-log memory records when each outcome became known and only injects lessons resolved by the trade date; a decision is no longer settled before its holding window fully trades.
WrenAI
  • bindings: build against wren-core 0.3.2 (2b0543d)

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.