OpenClaw Personal Assistant
A personal AI assistant that runs on your devices and meets you in the messaging channels you already use.
- Source repo
- openclaw/openclaw
- Stars
- ★ 394k
- Last updated
- today
- License
- MIT
- Primary language
- TypeScript
- FA score
- 47/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformChatGPT · Codex · Claude.ai · OpenAI API · Claude API
- Cost
- Free software; you pay for model usage
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- A single user who wants to talk to a personal assistant from WhatsApp or Telegram while keeping the Gateway on their own device.
- Not a fit if
- Non-technical users who want zero-config setup without channel and device pairing
- Commercial adopters who need clear license assurance, since repo metadata says NOASSERTION
- Source review
- 47/100 · Major gaps
What does this agent do, and when should you use it?
OpenClaw is a personal AI assistant developed by the OpenClaw Foundation, with a local-first Gateway as its control plane. It receives and answers messages through 25+ channels—including WhatsApp, Telegram, Slack, Discord, Signal, iMessage, and Microsoft Teams—as well as built-in WebChat. The Gateway manages sessions, channels, tools, and events, and can route incoming traffic by channel, account, or peer to isolated agent workspaces and sessions. It offers macOS, iOS, and Android nodes, a Windows Hub companion app, voice features, and a live Canvas surface. It is designed for a single user running an always-on assistant on their own environment; tools for the default main session run on the host.
openclaw onboard --install-daemon configures the Gateway, workspace, channels, and skills, then installs a persistent service. The Gateway normally runs on port 18789; openclaw gateway status checks it and openclaw dashboard opens the Control UI. openclaw message send --target +1234567890 --message "Hello from OpenClaw" sends through a connected channel, while openclaw agent --message "Ship checklist" --thinking high submits a prompt to the configured model and produces a response that can be delivered to a connected channel. Workspace context includes files such as AGENTS.md and SOUL.md, plus skills under the workspace skills directory; sessions_list, sessions_history, and sessions_send operate on sessions. Unknown direct-message senders receive a pairing code by default and their messages are not processed.
- A single user who wants to talk to a personal assistant from WhatsApp or Telegram while keeping the Gateway on their own device.
- An operator who receives messages in Slack, Discord, and Microsoft Teams and wants to route accounts or peers into separate workspaces and sessions.
- A macOS, iOS, or Android user who wants to use wake words, continuous voice, or voice input with a paired Gateway.
- A terminal user who wants to run a one-off prompt with
openclaw agentand optionally deliver the reply through a connected chat channel. - A user who needs an assistant with browser, Canvas, node, cron, and session tools while retaining local workspace configuration.
How do you install or deploy this agent?
Requires Node 24.15+ (recommended), Node 22.22.3+, or Node 25.9+. On macOS/Linux, run curl -fsSL https://openclaw.ai/install.sh | bash; on Windows PowerShell, run iwr -useb https://openclaw.ai/install.ps1 | iex; or install with npm install -g openclaw@latest. Then run openclaw onboard --install-daemon and complete setup for model authentication, the Gateway, workspace, and channels. Authentication can use API keys, or OAuth for Anthropic Claude Pro/Max, OpenAI ChatGPT/Codex, and GitHub Copilot.
How do you use this agent?
Run openclaw gateway status and confirm that the Gateway is running on port 18789, then use openclaw dashboard to open the Control UI. Test the assistant with openclaw agent --message "Ship checklist" --thinking high, or send a test message with openclaw message send --target +1234567890 --message "Hello from OpenClaw". After an upgrade, run openclaw update followed by openclaw doctor; for foreground debugging, first run openclaw gateway stop, then openclaw gateway --port 18789 --verbose.
What are this agent's strengths and limitations?
- A local-first Gateway centralizes sessions, channels, tools, and events in one control plane.
- It supports 25+ messaging channels alongside macOS, iOS, and Android nodes, Windows Hub, and built-in WebChat.
- It supports Anthropic, OpenAI, Google Gemini, xAI Grok, OpenRouter, GitHub Copilot, MiniMax, and OpenAI- or Anthropic-compatible endpoints.
- Unknown DMs use pairing by default, and non-main sessions can be sandboxed with Docker, SSH, or OpenShell backends.
- It requires a supported Node.js release; source development requires pnpm, and root-level
npm installis not a supported source setup. - Tools in the default main session run on the host with full access, so remote exposure requires careful security configuration.
- Messaging, voice, and node capabilities each require their own channel setup, device pairing, and model authentication.
- The supplied repository metadata says
NOASSERTION, while the README badge says MIT, leaving inconsistent license evidence for adopters to resolve.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| OpenClaw Personal Assistant This agent | 47 · Major gaps | CLIFree + model costs | ★ 394k | today | TypeScript | ChatGPT · Codex · Claude.ai · OpenAI API · Claude API |
| GAIA — Personal AI Assistant | 73 · Some gaps | Hosted serviceFreemium | ★ 324 | today | Python | — |
| LISA Autonomous Personal Assistant | 59 · Major gaps | CLIFree + model costs | ★ 175 | 28d ago | TypeScript | Codex · Claude Code · OpenAI API · Claude API |
| Kody Personal Assistant Platform | 70 · Some gaps | CLIFree + model costs | ★ 910 | today | TypeScript | — |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows security policy, pairing approval, sandbox options, and trust model, but lacks implementation details. Deductions: insufficient evidence for least privilege, user confirmation, data flow transparency, sensitive data handling, dependency security, external effects, rollback, and source attribution.
README and package.json are consistent, but no specific evidence for error handling or dependency availability. Deductions: lack of failure messages and dependency availability specifics.
Documentation describes multiple use cases and platforms, but capability boundaries and trigger precision are not clearly defined. Deductions: insufficient description of capability boundaries and trigger precision.
Clear documentation structure, install notes, naming, and examples, but known limitations and changelog are not present in provided files. Deductions: missing known limitations and changelog.
Output usability and marginal value are documented, but cost-benefit is not quantified. Deductions: lack of specific cost-benefit data.
Claims are documented, but lack independent verification and separation of facts and inferences. Deductions: insufficient claim traceability, cross-source corroboration, and fact-inference separation.
- Publisher identity is unverified; treat as unknown.
- Security policy explicitly states no multi-tenant isolation; be cautious when sharing a gateway.
- Default exec behavior is host-first; enable sandbox for isolation.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
v2026.11.1 openclaw 2026.11.1
Release notes excerpt
2026.11.1 Highlights - Safer Mac settings: failed config saves no longer overwrite secrets, and turning off Tailscale exposure keeps Gateway authentication intact. (#187734, #187664) Fixes #187600, #187597 Thanks @anmol0b. - More reliable channel replies: Telegram, Discord, and webhook conversations keep reaching the agent after request handlers or earlier turns finish. (#179200, #179385, #179704, #189457) Fixes #178554 Thanks @steipete, @LiuwqGit, @jalehman, and @vincentkoc. …View on GitHub ↗ -
v2026.10.10 openclaw 2026.10.10
Release notes excerpt
<!-- openclaw-release-publication:docs-v1 --> OpenClaw v2026.10.10 4,825 commits · 3,028 pull requests · 454 contributors Changes included in this release The release notes and changelog contain the same content, presented in two formats: - [Release notes](https://docs.openclaw.ai/releases/2026.10.10) — formatted for people. - [Changelog](https://raw.githubusercontent.com/openclaw/openclaw/main/CHANGELOG/2026.10.10.md) — plain Markdown for AI agents and tools. Thanks …View on GitHub ↗ -
v2026.9.36 openclaw 2026.9.36
Release notes excerpt
This is a gateway-only extended-stable release, which is our current equivalent to LTS. This release is OpenClaw from the end of September 2026, plus critical security updates, reliability and performance fixes, and features like new model support. The latest version of OpenClaw at the time of this release is [2026.10.9](https://github.com/openclaw/openclaw/releases#release-v2026.10.9) 2026.9.36 Highlights …View on GitHub ↗
FAQ
Can it use an existing model subscription?
What happens when an unknown person sends a DM?
dmPolicy: "pairing": the sender receives a pairing code and their message is not processed. Approve access with openclaw pairing approve <channel> <code>.Is it automatically safe for group use?
agents.defaults.sandbox.mode to non-main so non-main sessions run in sandboxes.How do I check for risky configuration or setup issues?
openclaw doctor; the README states that it surfaces risky or misconfigured DM policies.