- openclaw 2026.9.2 — 2026.9.2 Highlights - Faster, more responsive chat: keep chat, dashboards, and session interactions responsive while long transcripts and disk usage are processed, with direct dashboard lookup, less...
- Cloud sessions should start without a Gateway checkout — Summary Start a cloud session from a GitHub repository URL and ref without cloning the source or creating a working tree on the Gateway. Problem to solve Cloud session creation currently clones the...
- fix(firecrawl): report an unresolvable self-hosted baseUrl as a DNS failure — What Problem This Solves Fixes the DNS-diagnostic bug reported in #135333, following the self-hosted Firecrawl work discussed in #63877. When a configured self-hosted baseUrl cannot resolve,...
- fix(exec): block /approve smuggled behind shell line continuations — detectUnsafeExecControlShellCommand missed /approve split by a backslash-newline (echo hi; /appr\ove allow-once): the shell joins the word before executing, but tree-sitter-bash word-breaks there, so...
- fix(doctor): replace unavailable TUI cleanup advice — Related: #134031 What Problem This Solves Fixes an issue where operators following WhatsApp responsiveness advice were told that doctor --fix would stop TUI clients, although current repair mode...
- Gateway restart runs paced cron jobs before their accepted next-check deadline — Bug type Behavior bug (incorrect output/state without crash) Beta release blocker No Summary Restart catch-up can execute a paced cron automation before its accepted future next-check deadline. A...
- fix(test): preserve IRC import smoke under Bun — What Problem This Solves Fixes the IRC public-module import smoke test under Bun, where Node-only tsx preload hooks do not provide the required source loading behavior. Why This Change Was Made Keep...
- feat: start cloud sessions without a Gateway checkout — What Problem This Solves Starting a repository session in the cloud currently requires a Gateway clone and local managed worktree. This change prepares the checkout on the selected cloud or paired...
- feat(transcripts): browse and export saved meeting archives — Related: #130860 Builds on #139572 and #139658. What Problem This Solves Gateway clients can read recent meeting notes, but cannot page through a large archive, search saved utterances, download a...
- fix: native worker turns fail to load WebSocket transport — What Problem This Solves Fixes an issue where starting a native OpenClaw turn on a cloud or paired worker fails before its first reply because the transferred worker cannot find ws/package.json....
- feat: add WhatsApp phone-code login — Closes #85867 What Problem This Solves Fixes an issue where users connecting WhatsApp from a headless, remote, or QR-unavailable environment had no first-class alternative to scanning a QR code. It...
- fix(codex): deliver provider-only question prompts — What Problem This Solves Fixes #139809. Codex-backed runs dispatched directly to Telegram can carry messageProvider: "telegram" with no explicit messageChannel. The regular message tool path already...
- refactor(plugins): unify install config snapshot preparation — Plugin installation built its authorized config snapshot twice: once for valid config and again after plugin-owned recovery. The recovery branch also split install-record and persisted-location path...
- fix: session group defaults show the operator.admin requirement instead of a Git error — Closes #138955 What Problem This Solves Fixes an issue where operators opening New session defaults for a sidebar group and selecting a folder outside the configured agent workspace, on a connection...
- fix(test): close Telegram ingress HTTP fixtures reliably — What Problem This Solves Fixes cleanup failures in Telegram ingress outcome tests under Bun, where manually closing accepted connections before the HTTP listener produced ERR_SERVER_NOT_RUNNING. Why...
- chore(fleet): exercise numeric-user cells in Docker — Related: #137151 and #137211 What Problem This Solves Fleet's existing tests do not exercise an official cell image through the host CLI when the invoking user ID differs from the image user. This...
- [Feature]: Feature Request: Pre-send Message Validation for Feishu Group Chat — Summary Feishu group chat messages require XML tags for mention notifications. Agents frequently use wrong format, causing missed task assignments. Problem to solve In multi-agent setups using Feishu...
- ci: avoid screenshot capture for iOS unit-test-only changes — What Problem This Solves Changing an iPhone or Watch unit-test Swift file currently selects the full iPhone, iPad and Watch screenshot pipeline even though those files are not inputs to the capture...
- fix(cron): complete timeout test observation phase — Related: #139957, #138411 What Problem This Solves Resolves a problem where the cron process-group timeout test hangs after process cleanup gains a separate exit-observation phase. This caused a...
- Feishu plugin WebSocket doesn't receive im.message.receive_v1 events — title: "Feishu plugin: WebSocket mode doesn't subscribe to im.message.receive_v1 → no message events" Describe the bug OpenClaw Feishu plugin (v2026.6.9) connects via WebSocket successfully (ws...
- [Bug]: Compaction fails with 'Thread group size (1024) exceeds maximum threads per threadgroup (896)' — Summary Auto-compaction fails on local MLX inference with the error: The compaction recovery chain exhausts all attempts and blocks the session, suggesting a /reset. Steps to reproduce - Run an...
- fix(plugins): reject hollow installs and repair missing dependencies — Closes #101049 What Problem This Solves Fixes an issue where users installing an npm channel plugin could receive a successful install even though a required runtime dependency was missing. The...
- [Bug]: Session rotation creates duplicate sessions with broken delivery context — Bug type Behavior bug (incorrect output/state without crash) Beta release blocker No Summary When the Mattermost channel is reloaded (e.g., after config change, gateway restart, or channel restart),...
- [Feature]: 输入文本错别字检测 — Summary 在聊天输入中自动检测并提示常见的错别字、语法错误 示例： "非好常好" → 提示可能是"非常好"（"好"字多余） "请帮我查一下" → 无错误 Problem to solve 在聊天输入中当存在错别字，或语法错误时，可能造成OpenClaw费解甚至理解错误，从而引发上下文加长、反馈错误等。希望能自动检测文字、语法错误，并高亮提示以便及时纠正。提升答准率和对话效率。...
- fix(daemon): discover supported versioned Node runtimes — Summary - Recognize supported versioned and distro-specific Node executables during daemon runtime discovery. - Replace the duplicate node/node.exe-only classifier with the existing canonical...
- fix: sessions_yield keeps long transcript history available — What Problem This Solves Fixes an issue where sessions_yield cleanup on long SQLite-backed sessions temporarily made transcript history and bounded context unavailable while a full transcript...
- fix(channels): keep ACP configured bindings authoritative over stale non-ACP runtime bindings — > AI-assisted: yes Closes #115354 What Problem This Solves - Fixes an issue where users who have a per-peer ACP configured binding (e.g. {type: "acp", agentId: "claude", match: {channel: "feishu",...
- fix(macos): show the full automation count — What Problem This Solves Fixes an issue where the Mac Automations menu shows an incomplete enabled-job count when the catalog spans a list page. With 201 enabled jobs, the rendered menu reported 200....
- fix(ci): preflight job outputs exceed 1 MiB on full-plan pull requests — Related: - https://github.com/openclaw/openclaw/issues/139929 - https://github.com/openclaw/openclaw/issues/136820 - https://github.com/openclaw/openclaw/pull/140036 What Problem This Solves...
- refactor(browser): separate executable probes and share classification — Related: #135868 What Problem This Solves Browser executable discovery combined platform selection with version and command probes in a grandfathered oversized module, and maintained overlapping...
- fix: native plugin stylesheets fail in macOS over local HTTP — Closes #139708 What Problem This Solves Fixes an issue where users opening Workboard in the macOS app see Could not load plugin stylesheet: workboard when their authenticated Gateway runs over...
- OpenClaw repo stats for 2026-09-06 — 389006 stars, 81741 forks, 3280 contributors
- OpenClaw sponsors (87 total) — 87 sponsors supporting OpenClaw
- fix(workers): disambiguate skill and plugin scan manifests (#3617) — Use the claimed target kind for dual-manifest scan inputs while retaining full-directory scanning. Share the resolver across both workers and keep bundled SkillSpector directory resolution separate.
- fix(publish): recover failed staged release artifacts (#3616) — Authorize a successor through current publisher credentials, retain the failed attempt and artifact history, and require fresh scans before publication. Bind initial attempts before scanning and...
- fix(security): retain scanner error tail (#3606)
- fix(ci): stabilize the UI proof process-tree reaping test (#3597) — The reaper test's leader exited on a fixed 100 ms timer, so under load TERM could land before the grandchild installed its SIGTERM handler. TERM then killed the grandchild outright and the test...
- openclaw v2026.9.2 on npm — Multi-channel AI gateway with extensible messaging integrations
- openclaw/openclaw on Docker Hub — Official openclaw/openclaw docker images.
- mradermacher/Qwen3.5-9b-Opus-Openclaw-Distilled-i1-GGUF — text-generation
- OpenClaw — Deploy your own personal AI assistant with OpenClaw on a DigitalOcean Droplet® server, a powerful open-source platform that runs entirely on your infrastructure. OpenClaw connects to the messaging...
- Ask HN: Who is using OpenClaw?
- OpenClaw surpasses React to become the most-starred software project on GitHub
- Your Agent Queue Needs Backpressure, Not Just More Workers — Long-running agents rarely fail because a single request is too large. They fail when work arrives...
- Your Agent Queue Is Not a Workflow Until It Tracks Execution State — An AI agent can receive the same job twice even when your queue is behaving correctly. A worker may...
- An OpenClaw Agent ‘Made $8K/mo.’ A Human Closed All 17 Clients. — $8,000 a month, built by an AI agent from a Mac Mini in Provo, Utah. Two clients sent me that clip in the same week — the same May episode…Continue reading on Medium »
- When to pay for a managed AI agent and when to just self-host — KiloClaw shut its doors to new users, so your managed-agent decision just got simpler. The primary tool in this duo-pick is KiloClaw, the…Continue reading on Medium »
- Is KiloClaw Worth $55/Month After Anaconda’s Acquisition? The Hosting Decision Nobody Expected — KiloClaw is end-of-life — the managed AI agent hosting option you wanted is closing to new users, but the bigger decision is still open.Continue reading on Medium »
- Local Agents: Hermes vs OpenClaw — Run the best open-source AI Agents on your computer for freeContinue reading on Towards AI »
- The OpenClaw Setup Nobody Talks About: Local, Offline, and Actually Private — Everyone’s OpenClaw runs on a cloud VM they rent from someone else’s computer. Read that sentence again.Continue reading on Medium »
- MiroThinker — MiroThinker is an open-source search agent model, built for tool-augmented reasoning and real-world information seeking, aiming to match the deep research experience of OpenAI Deep Research and...