openclaw v2026.9.7 发布：更新安全回滚、OpenAI Agents API 插件与 Gateway 性能优化

Official release notes (https://github.com/openclaw/openclaw/releases/tag/v2026.9.7) ## 2026.9.7 ### Highlights - Update safety: updates now back up every state and agent database before migrations and restore them on rollback, take consistent snapshots while the Gateway keeps writing, and stop before schema changes when snapshot cleanup fails. Fixes #157846 and #157603. (#158163, #158317, #156265) Thanks @blyszcz, @Nickfost, and @fuller-stack-dev. - Upgrades from 2026.9.5: managed updates no longer always roll back with a stack overflow, Windows updates finish after state migration instead of stranding the new schema, verified macOS app handoffs activate, fresh npm updates load their HTTP client, and interrupted npm package updates recover. Fixes #157011, #157077, and #156878. (#157273, #157262, #156972, #157756, #160015, #158491) Thanks @yoyo837, @joshbunker, @hannesrudolph, and @fuller-stack-dev. - OpenAI Agents API: run agents on OpenAI-hosted or self-hosted environments through the new Agents API plugin, with streamed replies, steering, live web search, OpenClaw tools, attachments and hosted files, preserved tool history, OpenClaw persona and workspace context, self-hosted skill discovery, and accurate token usage. (#151176, #156092, #156167, #156306, #156182, #154208, #154229, #158328, #159375, #158775, #158307, #159363, #156549, #158744) Thanks @sjf-oa, @sjf, @vincentkoc, and @Patrick-Erichsen. - Sign in with ChatGPT: add Sign in with ChatGPT (Beta) as an OpenAI authentication choice beside Codex login and API keys, with clearer capability and limitation notes, restored first-run provider choices, and SIWC credentials kept out of unsupported media requests. (#148567, #160011, #160078, #160023, #159920, #159412) Thanks @stevenlee-oai. - Gateway responsiveness: busy chats no longer stall everyone else, because transcript writes and projections, cold history preparation, artifact reads and downloads, Control UI file reads, profile avatars, roster discovery, prompt hashing, and placement claims now run off the Gateway main thread, and large uploads, long streams, big message saves, file edits, and database maintenance stay responsive. (#147914, #154069, #156095, #158450, #158583, #158464, #158445, #158676, #158404, #158797, #159463, #158486, #158848, #155779, #158443, #158420, #158489, #158570, #158402, #158449, #159818, #158655) Thanks @joshavant, @fuller-stack-dev, and @vincentkoc. - Chat performance: long conversations now scroll smoothly, typing in the composer stays responsive in long chats, streaming replies no longer stall the page, and switching or reloading sessions shows fresh history sooner. Fixes #145777. (#159294, #159400, #159806, #154443, #159293, #159408, #159877, #158930, #156196, #156355) Thanks @Marvinthebored, @jalehman, @Peetiegonzalez, and @germanchi1. - Native Apple chat: macOS and iOS chat now shows message times, the model behind each reply, and agent identities, accepts file and audio attachments, keeps paragraph breaks before lists, and on macOS adds a command palette plus the web sidebar's roster defaults and view options. (#159366, #159528, #159667, #159517, #159350, #160141) - Restart continuity: in-flight worker turns and their edits survive a Gateway restart, cancelling a cloud worker turn during a runtime update no longer takes the Gateway down, ACP runs are no longer cancelled right after an in-process restart, and active turns keep working through configuration reloads. (#159565, #160062, #157658, #154462) Fixes #157442 and #154456 Thanks @MertBasar0, @obviyus, @olyapop, @TreyLawrence, and @VACInc. ### Changes - Worktree sessions: start a new chat in an isolated managed worktree from web, iOS, or Android so parallel sessions on one repository no longer collide. (#100788) Thanks @obviyus. - Voice: Talk, Discord voice, FaceTime, and Twilio voice calls share one agent context, so voice answers about other sessions delegate to the agent instead of denying that work, and read-only lookups no longer ask for spoken confirmation. (#159548, #159474, #156151) Thanks @obviyus, @VACInc, and @Peetiegonzalez. - Role model limits: administrators can give named operator roles a model policy that native agents, Visitor Access, model pickers, and plugin completions enforce, with denials keeping their documented authorization error code. (#156346, #154893, #154839, #156604) Thanks @shakkernerd and @obviyus. - Chat-driven setup: owners can hand their agent an API key, a config change, or an edit to a skill they own directly in chat, agents stop refusing or arguing over ordinary requested work while tool policy, sandboxing, and approvals still decide what needs confirmation, and OpenClaw change approvals now complete in the requesting chat with Allow once and Deny buttons or `/approve `. (#158120, #157447, #157947) Thanks @obviyus, @jalehman, and @joshavant. - Desktop audio: listen to managed desktop application sound from the docked desktop panel or the standalone desktop view, with isolated per-application audio on managed Linux desktops, playback tied to screen authorization, and setup warnings kept in a tooltip. (#157718, #157720, #157722, #157723, #157724, #157726, #159487) Thanks @vincentkoc. - Working status: show the agent's latest commentary as a live status line while it works, clearing when the run ends and keeping approval and startup messages first. (#156340) Thanks @vincentkoc. - Chat media: preview PDF attachments inline, move between expanded videos within a chat turn, and copy your own messages as Markdown. (#148399, #154051, #156563) Thanks @jjjhenriksen, @vyctorbrzezowski, and @VACInc. - New sessions: set `gateway.controlUi.newSessionModelDefaults: "configured"` to start fresh sessions from the agent's configured model, runtime, and thinking defaults instead of the last-used choices; the default stays `"last-used"`. (#156866) Thanks @vincentkoc. - Sidebar and presence: organize session filters in one panel, show GitHub merge and closure dates beside sidebar badges, see Open and Running session counts beside online people, and show active collaborators below typing previews. Fixes #150456. (#150605, #158081, #160022, #159263) Thanks @vyctorbrzezowski, @vincentkoc, and @shakkernerd. - Activity: open the Sessions tab with a Today pulse of hourly activity and counts that follow the current filters, highlight whole session rows on hover and focus, and quiet the row actions. (#159611) - Profile and preferences: see what your connection is allowed to do in plain language with a Reconnect action after access changes, and choose to open links in an external browser. Fixes #159884. (#156304, #159887) Thanks @vincentkoc and @VACInc. - Model pickers: group decision models by provider and drop the hover tooltip from the model picker. (#155301, #158697) Thanks @jalehman and @vyctorbrzezowski. - Chat tables: move table controls below the content, size desktop tables to their content, avoid unnecessary scrolling, and keep column order when expanding right-to-left tables. (#159924, #156874, #156750, #156947) Thanks @vyctorbrzezowski. - macOS app: let the macOS app embed the web conversation pane next to its native sidebar, with the chat header sitting in the native titlebar row. (#159739, #159932) - Channels: set `requireMentionInBotThreads: false` to accept unmentioned follow-ups only in threads the bot created, across Discord, Slack, Telegram, Matrix, Mattermost, Microsoft Teams, iMessage, Feishu, and other supported channels. (#157713) Fixes #157698. - Android chat: operate the agent browser inside chat with collapse and close controls, use a streamlined composer and attachment menu, find models despite typos or multiple search terms, see tool outcomes in collapsed work, and open context usage from the session menu. (#159767, #157799, #159725, #156950, #159733) Thanks @IWhatsskill. - Android setup and Overview: give saved gateways persistent local names, grant optional setup permissions together, open Settings beside the gateway selector, and read clearer node availability and paired-device explanations with realigned icons that separate work pages from settings. (#159714, #159727, #159718, #157746, #159732, #159961) Thanks @IWhatsskill. - macOS: renew a saved Gateway's Cloudflare Access browser sign-in before it expires instead of signing the app out every session lifetime. (#159381) - CLI: `openclaw --help` returns faster, and commands that do not need channel options skip resolving them at startup. (#30975) Thanks @vincentkoc and @obviyus. - Model pricing: price each route by whoever bills it, so an OpenRouter promotion no longer lowers cost estimates for OpenAI direct or other gateways, and download the smaller v2 hosted catalog with ordered model recommendations. (#156518, #156530, #156535, #158863) Thanks @vincentkoc and @obviyus. - Anthropic: bare `opus` and new Anthropic API, CLI, and media setup select Claude Opus 5.5 while explicit Opus 5 selections stay pinned, and thinking-display preferences reach Anthropic transports. (#156093) Thanks @fuller-stack-dev. - Gemini: add an opt-in `google-interactions` API backend for Gemini Interactions text, image, reasoning, and tool-call workflows. (#149880) Thanks @markmcd and @RomneyDa. - Telnyx: install Telnyx as an official external model provider through normal onboarding. (#116016) Thanks @dynamite-bud and @jalehman. - Codex: opt in to Ultrafast with `plugins.entries.codex.config.appServer.enableUltrafast`, used only when the selected model advertises it while other models keep their current speed. (#158703) Thanks @sjf-oa and @sjf. - Crabbox: run applications inside a Crabbox lease with `openclaw crabbox run --model` using a protected model credential that never leaves its original host. (#156207) - Browser: try opt-in Lightpanda semantic profiles with a portable deployment, and verify the Chromium headless shell across platforms. (#154359, #154360, #154396) Thanks @vincentkoc. - Browser sessions: let session writers use an isolated dashboard browser for their own session, show only the current session's tabs in the chat Browser panel, and use the Gateway `publicOrigin` as the default browser origin. (#156371, #159891, #157170) Thanks @vincentkoc. - Plugins: see declared plugin capabilities, setup guides, and tool inputs in detail previews, browse curated categories with a new Computer use shelf, and get consistent white icon tiles for bundled logos. (#157956, #157946, #158686, #155259) Thanks @Patrick-Erichsen. - Incognito sessions: expire Incognito sessions after 24 hours, keep their agent turns out of task records and logs, and stop their terminals when the session is reset. (#155596, #157543, #159207) Thanks @vincentkoc. - Uploads: let operators disable client file and image uploads to the Gateway. (#158567) Fixes #158556. - iOS connections: recognize when an external authorization provider requires browser sign-in before connecting, pause automatic reconnect, and offer a Retry action instead of treating it as a network failure. (#147070, #157454, #157457) Thanks @vincentkoc. - Subagents: clarify when research is worth delegating with `sessions_spawn` and how to follow up on a stalled child by checking execution and delivery status instead of loop-polling. (#141209) Thanks @hartmark and @jalehman. - Decision models: add the `decision_evaluate` tool for Boolean, Choice, and Score questions against an agent's configured decision model, with provider capabilities listed in `models.list`, and keep fallback working when a provider rejects the input. (#155134, #156746) Thanks @jjjhenriksen and @jalehman. - Portals: let collaborators open session-scoped interactive previews from dedicated worker attachments through a restricted `portal` tool, without global host-port access. (#156373) Thanks @vincentkoc. - Meetings: add experimental duplicate-safe Google Meet participation with retained caption context and speaker provenance. (#152327) Thanks @canvrno-oai. - Faster sessions and history: speed up session lists, chat history pages, archived session search, transcript appends, new conversation creation, subagent context loading, and automatic maintenance on busy stores, with fewer repeated scans, reloads, and refresh storms. (#155995, #156055, #156665, #157043, #157572, #157582, #157622, #158397, #158648, #158809, #158961, #159153, #159155, #159309, #159328, #159449, #159456, #159585, #159742, #159900, #158394, #159994, #159647, #159152, #159475, #156830, #158771, #158433, #158559, #158660, #159286, #159710, #159780, #159898, #158859, #158326, #158345, #156400, #158073, #158878, #158885, #159083) Thanks @vincentkoc and @chopin-op60. - Faster multi-client Gateway: reduce CPU and traffic when many browsers connect, reconnect, or watch the same chats by spreading reconnect attempts, streaming append deltas, reusing prepared identities, avatars, auth status, and message encodings, sharing pending approval replay, and skipping unchanged roster publications. (#158497, #158536, #158942, #160150, #158555, #158587, #155997, #158552, #158649, #158595, #158597, #156335, #156784, #156358, #158627, #158586, #158803, #158833, #159595, #158643, #158572, #159460, #159457, #158591) Thanks @vincentkoc. - Faster agent turns: keep buffered Anthropic streams responsive, serialize large model requests once, reuse Code Mode workers and sandbox classification, stop per-evaluation watchdog threads, unblock session lanes after no-progress loops, and cut tool catalog and validation overhead. (#156054, #156066, #156872, #157021, #158410, #158428, #158610, #158805, #158837, #158872, #159082, #159464, #159574, #152066, #157233, #156800) Thanks @obviyus and @vincentkoc. - Faster startup, cloud, and automation work: reduce startup placement work, prepare first cloud dispatch artifacts in parallel, stop polling checkout and PR state for idle sessions, and reuse unchanged artifact list summaries, and speed up concurrent automation lists, new chat bursts, and pairing commits. (#157079, #158065, #158525, #159301, #159390, #159520, #159102, #158941, #157113, #156801, #158077, #158665, #156387, #158867, #158779, #157776, #159537) - Diagnostics: identify agents across message and model-call traces, trace completed harness commentary, and report blocked CPU profiler starts, waiting or failed lane tasks, live worker counts per pool, heap allocation attribution, and guarded heap snapshot diffs. (#156018, #156343, #158361, #158549, #158561, #159892, #160040) Thanks @Patrick-Erichsen, @ryan-dyer-sp, and @vincentkoc. - Session access: people with session-only grants can read visible conversations, receive updates, copy history as Markdown, and rename, pin, archive, and restore their own sessions, while other people's sessions stay read-only. Guests can now create and control their own sessions. (#155184, #155565) Thanks @shakkernerd. - Discord and Slack: accept bot-authored messages by default when `allowBots` is omitted, still subject to existing access and mention rules (explicit `false` and `"mentions"` are unchanged), and keep parent context when someone replies to a bot's Discord message. (#157091) - People and presence: agents can use the `presence` tool to list who is online and which device produced a person's latest activity, idle people stay shown as online, administrators can merge duplicate profiles with `openclaw users merge`, and members of multi-user Gateways can edit their personal `USER.md` in Profile settings. Fixes #159423 and #155255. (#159117, #159493, #159889, #155256) Thanks @VACInc. - Ultra mode: use Ultra as a turn-scoped harness mode on OpenClaw and Claude Code runtimes independently of the model's maximum reasoning effort, while native Codex keeps its own Ultra behavior. Fixes #155391. (#155393) Thanks @vincentkoc. - Terminal access diagnostics: `openclaw exec-policy show` and the agent Tools panel explain why an agent profile hides `exec` and `process`, including configuration sources and the global-to-agent profile hierarchy. Fixes #155914. (#157014) Thanks @joshavant and @Pegorim. - Managed updates: OCM-managed installations can request updates from the Control UI and `/update` and follow the OCM job through completion, and failed-update diagnosis now starts only after you click. (#158932, #157359) Thanks @shakkernerd, @fuller-stack-dev, and @vincentkoc. - Subagents: scope subagent concurrency to each spawning session, report capacity waits as queued execution, and tell models the wait timeout limits. Fixes #156370. (#156381, #156681, #156542) Thanks @vincentkoc. - Background commands: show native background commands in Tasks, warn agents when a background command cannot wake them, and keep quiet heartbeats quiet after tool failures. (#158038, #158884, #156174) Thanks @VACInc and @pash-openai. - Cloud workers: show the live cloud worker pool in Settings and start warm cloud workers and snapshots faster. (#157200, #156380, #156767) - Agents API: run isolated restricted sessions so Memory dreaming can write real diary narratives through the Agents API backend. (#159246) Thanks @sjf-oa and @sjf. - Side chat: remove automatic summaries, accept dropped images, and report instead of silently dropping images on text-only models. (#156892, #156840, #156900) Thanks @Patrick-Erichsen. - HTML widgets: allow widgets up to 10 MiB. (#159970) - Faster sessions: reduce Gateway CPU for session listing, concurrent viewers, activity broadcasts, chat history catch-up, streamed replies, and mention suggestions, and parse streamed SSE incrementally with CR-framed events accepted. (#156221, #157120, #158346, #158408, #158469, #159182, #158481, #158503, #157552, #154929, #158395, #158444, #156817, #156922, #158792) - Faster databases: reuse prepared statements and schema facts, shorten agent database admission checks, keep idle agent databases open for thirty minutes, preserve database witnesses across managed restarts, and reduce host CPU for agent worker writes. (#154946, #158973, #156643, #157030, #159445, #158405, #159753, #157666, #158750, #158843, #158763, #159876, #158385, #157333) - Faster tasks and tools: reduce CPU for task maintenance, trajectory tool-schema redaction, image discovery in text-heavy conversations, meeting transcript lookups, and frontmatter recovery, give swarm collector groups independent lanes, and defer TypeSafe evaluator startup until its first decision. (#156321, #157527, #159271, #156183, #158471, #158871, #158371, #159074, #158374, #158616, #157380, #159004) Thanks @vincentkoc and @joshavant. - Config: support `${VAR:-default}` fallback values in config environment substitution instead of passing the expression through unresolved. Fixes #149551. (#155164) Thanks @etzelm, @jalehman, and @z2mi. - Package updates: let the new release run its own read-only admission checks for configuration, database compatibility, and plugin availability, so an older installed updater no longer refuses an update the candidate can handle; `--admission installed` forces the installed checks, and Git/source updates are unchanged. (#155632) - Update status: show quiet per-step progress while staging Git runtimes and after the Gateway startup check, date recorded outcomes, report the preferred Git release target and known scheduler policy, and clear the applying status once a handed-off update finishes. (#153407, #155671, #156227, #157329, #158926, #158301, #158006) Thanks @RomneyDa, @vincentkoc, @shakkernerd, and @joshavant. - System services: updates now refresh system-scope Gateway services and hand the final privileged restart to the operator instead of failing. Fixes #156409. (#156584) Thanks @ion-developing. - Accounts: show the ChatGPT sign-in email and the selected account's identity in provider settings and the composer, release deleted credentials from model defaults and conversations, and offer existing saved accounts during recovery. (#157629, #157745) Thanks @stevenlee-oai and @vincentkoc. - CLI: reject unknown root options instead of launching onboarding, list the non-interactive setup command in root help, keep the process title as `openclaw`, fail `reset` when state cannot be removed, and never fall back to local state when a configured remote Gateway is unreachable. (#159213, #159214, #157118, #159250, #159223) - Video generation: add Kie AI, Z.AI, and Novita video generation providers, including bundled Kie plugin artwork. (#160080, #160529) Thanks @vincentkoc. - macOS: show the web conversation in native chat windows. (#159946) - Turn control: people with the same permissions can steer each other's turns, and write operators can control their own sessions through agents. (#159850, #160296) Thanks @VACInc. - Chat performance: throttle sidebar narration updates, describe each session once per load, skip needless markdown parsing on every render, stop app-shell style rules from restyling the whole app on each insertion, render less offscreen work when presenting a conversation, avoid roster reloads after read acknowledgments, omit duplicate assistant text for chat clients, defer session titles until reply progress, and shorten the CLI startup banner animation. (#160002, #160148, #160294, #160241, #160298, #160328, #160186, #160251, #160250) Thanks @IWhatsskill. - Faster runtime: shared-state reads no longer re-read the ownership lock every time, and worker launch helpers start faster on session hosts. (#160139, #160163) - Desktop control: agents can resume after you take manual control of a desktop. (#159923) - Faster sessions and nodes: loading older chat history reuses normalization and tool turns, channel config reloads keep prepared state, remote session turns finish faster when files did not change, prompt caching stays effective for sessions on nodes, and cross-session preparation and registry maintenance wait less. (#160215, #160274, #160231, #160179, #160183, #160510) - Faster sessions and nodes: session list reads reuse registry relations, and remote session turns finish without a status polling delay. (#160198, #160162) - Session sharing: show the assigned session owner beside sharing controls, include the signed-in user in the Online roster, and keep PR and issue identity in failed previews. (#160673, #160618, #160219) Thanks @Patrick-Erichsen, @VACInc, and @vyctorbrzezowski. - Voice and speech: join Slack huddles as a signed-in Slack user, speak Gemini dialogues with two voices, and speak with Gemini 3.8 Flash TTS. (#159879, #157465, #157331) Thanks @saariuslystoned and @IWhatsskill. - Plugins: show installed accounts and credential status for plugins. (#160092) Thanks @Patrick-Erichsen. - Composer: remove large pasted text from the composer preview. (#160834) Thanks @Patrick-Erichsen. - Faster chats and sessions: streaming replies stay fast in long chats, node session turns start as fast as local ones by reusing warm workers, child-session lists load faster on large rosters, session lists stay warm when changing the Talk model, the Gateway stays responsive while compaction loads history, SQLite entry writes yield under contention, and Gateway broadcasts, narration deadlines, prompt diagnostics, Code Mode tool search, stream payload inspection, subagent registry patches, cron delivery lookups, canvas leases, and page styling all do less repeated work. (#160837, #160265, #160404, #159545, #160314, #160801, #160787, #160769, #160807, #160811, #160208, #160800, #160216, #160808, #160833) - UI polish: Online people activity rows are more compact, GitHub reference backgrounds are softer across themes, and failed-message inbox alerts clear after review. (#160140, #160844, #160822) Thanks @vyctorbrzezowski, @VACInc, and @stevenlee-oai. - Models: support Claude Sonnet 5.5. (#160847) - Approvals: enforce scoped Slack plugin reviewers. (#159525) Thanks @stevenlee-oai. - Session hosts: show worker runtime install progress on paired session hosts, explain and recover session-host setup problems, and let Ask in side chat stage selection comments. (#160110, #160188, #160747) Thanks @Patrick-Erichsen and @TeaCup404. - Faster sessions: metadata reads skip co Repository README (current repository documentation, not release changes) # OpenClaw 🦞 — Your assistant, on your devices, in your chats OpenClaw is an open-source AI assistant that runs on your own computer and meets you in the channels you already use: Discord, iMessage, Slack, Teams, Telegram, WhatsApp, and 20+ more, plus native apps for macOS, iOS, Android, Windows, and Linux. One Gateway runs it as a personal assistant on a laptop or as a shared [team deployment](https://docs.openclaw.ai/start/teams); configuration is the only difference. **Yours, with no catch.** State, memory, and credentials live on your hardware. Models and agent harnesses (Claude, Codex, local models) are plugins you can swap without changing anything else. Your prompts go to the model provider and chat platforms you configure, plus any diagnostics export you enable yourself; by default OpenClaw itself phones home for nothing but a daily version check, anonymous feature statistics are opt-in, and `update.checkOnStart: false` disables both ([what OpenClaw sends](https://docs.openclaw.ai/gateway/telemetry)). OpenClaw is stewarded by the [OpenClaw Foundation](https://openclaw.org), an independent 501(c)(3), and has no paid tier, hosted service, or token. The architecture case — trusted gateway, untrusted execution, deterministic policy — is in [Why OpenClaw](https://docs.openclaw.ai/start/why-openclaw). [Website](https://openclaw.ai) · [Docs](https://docs.openclaw.ai) · [Getting started](https://docs.openclaw.ai/start/getting-started) · [Why OpenClaw](https://docs.openclaw.ai/start/why-openclaw) · [FAQ](https://docs.openclaw.ai/help/faq) · [Vision](VISION.md) · [DeepWiki](https://deepwiki.com/openclaw/openclaw) ## Install The installer supports macOS, Linux, and Windows. It provisions a supported Node.js runtime when needed. ```bash # macOS / Linux / WSL2 curl -fsSL https://openclaw.ai/install.sh | bash ``` ```powershell # Windows PowerShell iwr -useb https://openclaw.ai/install.ps1 | iex ``` Already manage Node.js? Install the published package instead (Node 24.16+ or 26.1+; Node 26 recommended): ```bash npm install -g openclaw@latest --allow-scripts=openclaw ``` That command is for npm 12 or npm 11.16+. On npm 11.15 and earlier, omit `--allow-scripts=openclaw`. See the [installation guide](https://docs.openclaw.ai/install) for the lifecycle script contract, Docker, Nix, and other deployment paths. ## Quick start On a fresh install, the installer scripts start onboarding automatically. Complete the wizard they open. If you installed the package directly with npm, pnpm, or Bun, run: ```bash openclaw onboard --install-daemon ``` After onboarding: ```bash openclaw gateway status openclaw dashboard ``` Onboarding verifies model access, creates the workspace, and configures the Gateway. The last command opens the Control UI; send a message there to confirm the assistant is working. See the [getting started guide](https://docs.openclaw.ai/start/getting-started) for channel setup and troubleshooting. ## How it fits together - The [Gateway](https://docs.openclaw.ai/gateway) is the local control plane for sessions, tools, events, and channel connections. - The [Control UI](https://docs.openclaw.ai/web/control-ui), CLI, and [TUI](https://docs.openclaw.ai/web/tui) connect to the Gateway. - [Channels](https://docs.openclaw.ai/channels) bring the assistant to WhatsApp, Telegram, Slack, Discord, Google Chat, Signal, iMessage, and other messaging services. - [Companion apps and nodes](https://docs.openclaw.ai/platforms) add voice, Canvas, camera, screen, and device-local actions on supported platforms. OpenClaw works with hosted and local [model providers](https://docs.openclaw.ai/concepts/model-providers). Its [tools](https://docs.openclaw.ai/tools), [skills](https://docs.openclaw.ai/tools/skills), and [plugins](https://docs.openclaw.ai/plugins) extend what an assistant can do. ## Security Treat inbound messages as untrusted input. DM-capable channels pair unknown senders by default; approve a pairing request with `openclaw pairing approve `. Tools run on the host for the main session unless you configure sandboxing. Read the [security guide](https://docs.openclaw.ai/gateway/security), [exposure runbook](https://docs.openclaw.ai/gateway/security/exposure-runbook), and [sandboxing guide](https://docs.openclaw.ai/gateway/sandboxing) before connecting other users or exposing the Gateway remotely. ## Documentation | Goal | Start here | | -------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Configure models and auth | [Models](https://docs.openclaw.ai/concepts/models) · [Model providers](https://docs.openclaw.ai/concepts/model-providers) | | Connect a messaging service | [Channels](https://docs.openclaw.ai/channels) | | Add tools, skills, and plugins | [Tools](https://docs.openclaw.ai/tools) · [Skills](https://docs.openclaw.ai/tools/skills) · [Plugins](https://docs.openclaw.ai/plugins) · [ClawHub](https://clawhub.ai) | | Run apps and device nodes | [Platforms](https://docs.openclaw.ai/platforms) · [Nodes](https://docs.openclaw.ai/nodes) | | Use the CLI and chat commands | [CLI reference](https://docs.openclaw.ai/cli) · [Slash commands](https://docs.openclaw.ai/tools/slash-commands) | | Configure or operate the Gateway | [Configuration](https://docs.openclaw.ai/gateway/configuration) · [Architecture](https://docs.openclaw.ai/concepts/architecture) · [Updating](https://docs.openclaw.ai/install/updating) · [Release channels](https://docs.openclaw.ai/install/development-channels) | ## Development The repository is a pnpm workspace. Plain `npm install` at the repository root is not supported. ```bash git clone https://github.com/openclaw/openclaw.git cd openclaw pnpm install pnpm build pnpm ui:build ``` See [CONTRIBUTING.md](CONTRIBUTING.md) for the contribution workflow and the [source setup guide](https://docs.openclaw.ai/start/setup) for the development loop. ## Governance OpenClaw is developed in the open by the [OpenClaw Foundation](https://openclaw.org), an independent 501(c)(3). The Foundation employs the core team and signs releases. Donors and infrastructure sponsors support the Foundation; none of them own or direct the project. OpenAI is a donor, not an owner. ## Community See [CONTRIBUTING.md](CONTRIBUTING.md) for maintainers and contribution guidelines; AI-assisted PRs are welcome. Use the [issue chooser](https://github.com/openclaw/openclaw/issues/new/choose) for bugs and feature requests, ask setup questions in [Discord](https://discord.gg/clawd), and report vulnerabilities through [SECURITY.md](SECURITY.md). New capabilities usually belong in plugins built on the [plugin SDK](https://docs.openclaw.ai/plugins/building-plugins) and shared through [ClawHub](https://clawhub.ai). OpenClaw was built for **Molty**, a space lobster AI assistant, by Peter Steinberger and the community. Explore the [project lore](https://docs.openclaw.ai/start/lore), [soul.md](https://soul.md), [Peter's site](https://steipete.me), [Star History](https://www.star-history.com/#openclaw/openclaw&type=date&legend=top-left), and [@openclaw](https://x.com/openclaw). Special thanks to [Mario Zechner](https://mariozechner.at/) for his support and for [pi](https://github.com/earendil-works/pi), and to Adam Doppelt for the lobster.bot domain. ## Donors and sponsors The Foundation is funded by donors including Amazon, Lobster Computer Company, Offline Holdings, OpenAI, Red Hat, and the University of Michigan, with infrastructure support from Blacksmith, Convex, GitHub, NVIDIA, and Vercel. ## Contributors Thanks to all clawtributors: <img src="https://avatars.githubusercontent.c