OpenClaw logo
Personal Assistants · Agents Orchestration Frameworks

OpenClaw

OpenClaw is a free, open-source personal AI assistant that runs on your own machine and answers from WhatsApp, Telegram, Discord or Slack. It handles email, calendars, browsers, files and shell commands, using whichever AI model you connect.

Active Free plan Free API available Verified by Guidaio
Overview

What is OpenClaw?

OpenClaw is an open-source personal AI assistant that you install and run yourself, rather than a service you log into. Its core is written in TypeScript and requires Node, because the canonical state store uses node:sqlite. The architecture centres on a Gateway running on a machine you control: a laptop, a Mac mini, or a VPS on Fly.io, Hetzner, GCP, Azure or Daytona. Chat channels and companion devices connect to that Gateway, and your data and context stay there instead of in a vendor cloud.

You talk to it through messaging apps you already use. The site lists 29 chat channels, among them WhatsApp, Telegram, Discord, Slack, Signal, iMessage, Microsoft Teams, Matrix and IRC, in direct messages as well as group conversations. Model choice, authentication and execution runtime are deliberately separated: 64 model and media providers are documented, so you can bring a direct API key from Anthropic, OpenAI, Google or Mistral, sign in with an existing ChatGPT, Claude or GitHub Copilot subscription, route through gateways such as Amazon Bedrock, OpenRouter or LiteLLM, or run models locally with Ollama, LM Studio or vLLM.

What the assistant can do goes well beyond conversation. It reads and writes files, runs shell commands and scripts, drives a browser to click and fill forms, generates images, video, music and speech, and keeps persistent memory across sessions. Eight automation mechanisms, from scheduled tasks and heartbeats to hooks, standing orders and webhooks, let it work while you are away. Companion nodes for macOS, Windows, iOS and Android expose camera, screen, voice and location, and a Chrome extension shares only the tabs you place in a dedicated tab group. The Gateway can also serve an OpenAI-compatible HTTP API, disabled by default.

Around the assistant sit roughly 70 satellite open-source projects and ClawHub, a public registry of community skills and plugins, alongside 142 official plugins counted on the integrations page. The project was created by Peter Steinberger in late January 2026 and is now stewarded by the OpenClaw Foundation, a 501(c)(3) American non-profit announced on 8 July 2026, which keeps the code MIT licensed.

What it does

  • Read, triage and send email from your inbox
  • Manage a calendar, schedule events and resolve conflicts
  • Read and write files, run shell commands and scripts on your machine
  • Browse the web, click, fill forms and extract data from any site
  • Reply from WhatsApp, Telegram, Discord, Slack, Signal or iMessage
  • Work in the background through scheduled tasks, heartbeats, hooks and webhooks
  • Generate images, video, music and spoken replies through the providers you connect
Audience

When to use OpenClaw / When not to

A quick filter to help you decide if OpenClaw is the right fit.

When to use OpenClaw

  • Developers and DevOps engineers comfortable installing a service and keeping a machine running
  • Founders, freelancers and consultants who want email, calendar and admin handled without a subscription
  • Anyone already paying for a ChatGPT, Claude or GitHub Copilot plan who wants to reuse it inside an agent
  • Privacy-minded users who want context, memory and files to stay on hardware they control
  • Small teams that would rather have the assistant inside Slack, Discord or Teams than in another browser tab

When not to use OpenClaw

  • Non-technical users looking for a hosted product with a sign-up form and no terminal
  • Organisations that need contractual guarantees: terms of service, a DPA or a written GDPR position, none of which exist
  • Anyone unwilling to grant an agent full file and shell access to their machine
  • Teams needing commercial support with a response commitment, since help comes only from Discord, GitHub and the docs
  • Users without a machine that can stay powered on, or without a model subscription or API key to connect
Get started

How to use OpenClaw

A typical end-to-end flow, from setup to results.

  1. Install with the one-line script, with npm i -g openclaw, or from a git checkout using pnpm
  2. Run openclaw onboard to set up the assistant, its persona and its configuration
  3. Connect a model: an API key, an existing ChatGPT, Claude or Copilot subscription, a gateway, or a local model
  4. Add one or more chat channels, each of which installs in a single command or during onboarding
  5. Install the Gateway as a service with openclaw onboard --install-daemon or openclaw gateway install
  6. Message the assistant from your chat app and let it build persistent memory as you go
  7. Optionally pair a companion node: the macOS menu bar app, Windows Hub, or the iOS and Android apps
  8. Optionally install skills and plugins from ClawHub, or ask the assistant to write its own
  9. Optionally enable the OpenAI-compatible HTTP endpoint, which is off by default, for programmatic access
  10. Use openclaw doctor to diagnose or repair the install, and openclaw update --channel to switch release channels
Quick read

Pros & Cons

Pros

  • Free and MIT licensed, with code and signed releases published on GitHub
  • Runs on your own hardware, so context and state never reach a vendor cloud
  • No model lock-in: subscriptions, API keys, gateways and local models all work
  • Reachable from chat apps you already have, with no new interface to learn
  • Very large documentation, translated into twenty languages
  • Stewarded by a 501(c)(3) foundation with a stated neutrality commitment and named industrial partners
  • Android app on Google Play, plus signed APKs carrying verifiable GitHub Actions provenance

Cons

  • Command-line installation, a mandatory Node runtime and a machine that has to stay on
  • No hosted version: everything is self-operated
  • No terms of service, legal notice, DPA or GDPR statement published anywhere
  • The privacy policy covers only the mobile apps and the Chrome extension, not the core runtime
  • No postal address published, despite 501(c)(3) status
  • No support email address; help runs through Discord, GitHub and the documentation
  • Companion apps for macOS and Windows are still labelled Beta, and no Linux companion app exists
Pricing

Pricing & Plans

OpenClaw is distributed free of charge under the MIT licence, and no paid tier of any kind is published: the site carries no price, no plan comparison and no billing page. A permanent free offer is therefore the only one available. The genuine cost sits elsewhere, with the AI model provider you connect, whether an existing subscription, a pay-as-you-go API key or a locally hosted model that costs nothing, and with the machine or VPS that runs the Gateway. Neither amount is quantified by OpenClaw.

Plan 1
Free
  • the complete product under the MIT licence
  • with no paid tier
  • no seat limit and no feature gating
Prices and plans listed above may evolve. Always check the official pricing page before subscribing.
Trust & Privacy

Data, GDPR & hosting

A consolidated view of how OpenClaw handles your data.

GDPR overview

There is no GDPR content on this site at all. Across the marketing pages and the full English documentation, the words GDPR and General Data Protection Regulation never appear once. This is silence rather than a refusal: OpenClaw neither claims compliance nor denies it. There are no terms of service, no legal notice, no data processing agreement, no subprocessor list and no Article 27 representative. The single privacy policy, effective 2 August 2026, covers only the iOS and Android apps and the Chrome extension, and names security@openclaw.ai for privacy questions. One structural point softens the picture: in a standard self-hosted deployment the Foundation receives no user data, so the operator of the gateway, usually you, is the party actually handling personal data and carrying the obligations.

Who owns the data?

OpenClaw takes ownership of nothing you produce with it. The published privacy policy, which covers the mobile apps and the Chrome extension, states that data goes to the gateway you configure rather than to a central OpenClaw cloud, and that there is no OpenClaw cloud in the extension's path and no third-party data collection. In a standard self-hosted deployment the vendor therefore holds nothing: files, transcripts and memory live on your machine, while mobile secrets sit in the iOS Keychain or Android encrypted preferences. The parties that actually see your content are whoever operates your gateway and the AI model provider you chose to connect.

Reuse rights

Because OpenClaw runs on your own hardware and the Foundation publishes no terms of service, there is no contractual restriction on what you may do with the data the assistant produces or processes: no licence is granted to OpenClaw over your content, and none is claimed back from you. The software itself is MIT licensed, which permits commercial use, modification and redistribution. In practice the limits come from elsewhere, namely the terms of the AI model provider you connect and those of any third-party service your gateway forwards to. The privacy policy says plainly that such services then handle data according to their own policies.

Data retention & training

Retention summary
Retention is entirely in your hands, because nothing is stored on the vendor's side. Data on your device stays there until you delete it, clear the app data or uninstall the app. Anything handled by your gateway or a connected third-party service is kept according to that service's own configuration and policies. Temporary files may be written to the app cache while features run. On iOS the app keeps a small read-only offline cache of recent chat sessions and an outbox of up to 50 queued messages that expire after 48 hours offline, and a reset or forget action clears both. You can disconnect from a gateway, clear app storage or uninstall at any time. No vendor-side retention period is published, because there is no vendor-side storage.
Trains on customer data
No
GDPR contact

Hosting summary

There is no vendor hosting to describe. OpenClaw publishes neither a hosting country nor a region, because it does not host user data: the Gateway runs wherever you put it. The documented options are a personal laptop, a Mac mini, a Raspberry Pi, a machine on your tailnet, or a VPS on Fly.io, Hetzner, Google Cloud, Azure, exe.dev, Daytona or EasyRunner, so the jurisdiction that applies is the one you choose. The privacy policy states that the gateway may be local, on your tailnet, self-hosted, or operated by someone else. On mobile, connection secrets are stored locally using platform security features, namely the iOS Keychain and Android encrypted shared preferences. The marketing site itself is served from Vercel infrastructure, but that concerns the web pages rather than user content. If you route your assistant through a hosted model provider, that provider's own hosting terms apply to whatever you send it.

Watch-outs

Things to keep in mind

Risks and trade-offs to weigh before adopting OpenClaw.

  • The assistant can read and write your files and run shell commands, so a misunderstood instruction can do real damage
  • The Gateway's HTTP endpoint is described in the documentation as equivalent to owner-level credentials and must never be exposed to the public internet
  • Skills and plugins installed from ClawHub are third-party code running on your machine with your agent's permissions
  • No terms of service, DPA or GDPR position exists, so there is nothing contractual to fall back on if something goes wrong
  • An assistant that reads email, watches calendars and acts on its own invites over-delegation: check what it did rather than assume it was right
  • The privacy policy leaves the core runtime undocumented, and your data's real path depends on the model provider you chose
  • The project is barely six months old and its foundation only weeks old, so governance maturity is still unproven
Setup

Setup & Integrations

Technical difficulty

Moderate to high, and clearly aimed at technical users. Installation is a command line affair: a curl one-liner, npm i -g openclaw, or a git checkout with pnpm. Node is mandatory, and on macOS the first run may need administrator rights for Homebrew. You then run openclaw onboard, connect a model provider, add at least one chat channel and install the Gateway as a system service. A machine has to stay available for the assistant to keep working. Companion apps for macOS, Windows and Android ease daily use, and openclaw doctor repairs a broken install.

Deployment

Web appDesktop appMobile appIOS appAndroid appChrome extensionBrowser extensionAPIPluginSlack app

Apps stores

Integrations

WhatsApp Telegram Discord Slack Signal IMessage Microsoft Teams Matrix Mattermost IRC LINE Google Chat Nextcloud Talk Twitch WeChat Zalo Feishu Nostr Twilio Synology Chat Anthropic OpenAI Google Gemini XAI MiniMax Moonshot Qwen Z.AI Mistral DeepSeek GitHub Copilot Amazon Bedrock OpenRouter Groq Cerebras Together AI Fireworks NVIDIA Hugging Face Ollama LM Studio VLLM LiteLLM Cloudflare AI Gateway Vercel AI Gateway Cohere Perplexity ElevenLabs Deepgram Runway ComfyUI Fal Azure Speech Gmail GitHub Obsidian Notion Apple Notes Spotify Sonos Philips Hue Home Assistant 1Password

Supported languages

EnglishChineseJapaneseSpanishPortugueseKoreanGermanFrenchHindiArabicItalianVietnameseDutchTurkishUkrainianIndonesianPolishRussianPersianThai
Company

Behind OpenClaw

Company name
OpenClaw Foundation
Founded
30/01/2026
Country of origin
🇺🇸 United States
UBO
INFORMATION_NOT_FOUND
UBO country
INFORMATION_NOT_FOUND
Domain registrar country
🇺🇸 United States

Fundraising

No funding rounds: OpenClaw is stewarded by a 501(c)(3) non-profit funded by donations rather than equity
Major donors named on 8 July 2026: Offline Holdings, Lobster Computer Company, University of Michigan and OpenAI
The University of Michigan is described as the largest donor and launched an Institute for Agentic Computing
OpenAI is described as a major donor, supports inference and runs an internal team called Claw Labs

Social

Official links

Resources

All the official URLs gathered for verification and reference.

FAQ

Frequently asked questions

Is OpenClaw free?
Yes. The software is open source under the MIT licence and there is no paid tier at all. You pay only your AI model provider, and the machine or VPS that runs the Gateway if you use one.
Where does my data actually live?
On the machine hosting your Gateway. The privacy policy states that data goes to the gateway you configure rather than to a central OpenClaw cloud, and that this gateway may be local, on your tailnet, self-hosted or operated by someone else.
Which AI model does it use?
Whichever one you connect. The integrations page lists 64 model and media providers, including Anthropic, OpenAI, Google Gemini, xAI, Mistral and DeepSeek, gateways such as Amazon Bedrock and OpenRouter, and local runtimes like Ollama, LM Studio and vLLM.
Which apps can I talk to it from?
Twenty-nine chat channels are documented, among them WhatsApp, Telegram, Discord, Slack, Signal, iMessage, Microsoft Teams, Matrix, Mattermost and IRC, in both direct messages and group chats.
Is there a mobile app?
Yes. The Android app is published on Google Play and an iOS app is documented; both are companion nodes that require a running Gateway. The site does not publish an App Store link.
Does OpenClaw offer an API?
Yes. The Gateway can serve an OpenAI-compatible HTTP surface with chat completions, models, embeddings and responses endpoints. It is disabled by default, and the documentation insists it stay on loopback, tailnet or private ingress.
Is it GDPR compliant?
The site never addresses the question. GDPR is not mentioned once on the marketing pages or in the documentation, and there is no terms of service, no data processing agreement and no Article 27 representative.
Who is behind the project?
The OpenClaw Foundation, a 501(c)(3) American non-profit announced on 8 July 2026. The project itself was created by Peter Steinberger and started in late January 2026.
How technical do I need to be?
Fairly technical. You install from the command line, need a Node runtime, register the Gateway as a system service, and keep a machine available for the assistant to run on.
How do I get help?
Through the online documentation, the Discord community and the GitHub repository. No support email address is published, and the only contact address on the site is for privacy and security questions.
Conclusion

Should you pick OpenClaw?

OpenClaw is unusually substantial for a project barely six months old. The code is public and MIT licensed, the documentation runs to hundreds of pages and is translated into twenty languages, an Android app ships through Google Play with signed builds, and the HTTP API is documented endpoint by endpoint. Microsoft, NVIDIA, OpenAI, Red Hat and the University of Michigan are all named as partners or donors. This is a real product with a real community, not a landing page.

Its appeal is straightforward: the software costs nothing, it is tied to no model vendor, and your context stays on hardware you control. For a developer or a technically confident founder, that combination is hard to find elsewhere.

The reservations are just as clear, and they concern governance more than capability. OpenClaw publishes no terms of service, no legal notice, no data processing agreement and no postal address, and the word GDPR appears nowhere on the site or in its documentation. The single privacy policy covers only the mobile apps and the Chrome extension, leaving the core runtime undocumented in that respect. An organisation that has chosen 501(c)(3) status has an address somewhere, and not publishing it is a gap a European buyer will notice.

There is also the question of what the tool is allowed to do. Full file and shell access is the point of the product, and the documentation is refreshingly blunt that the HTTP endpoint amounts to operator credentials and must never face the public internet. That honesty is welcome, but it does not remove the risk. In short: excellent for technically comfortable individuals and small teams who want an assistant they own; organisations that need a contract, a DPA or a compliance statement should wait until the Foundation publishes one.