# OpenClaw

Self-hosted, MIT-licensed AI assistant you message from the chat apps you already use

**Verdict: conditional.** OpenClaw is a credible free way to run a personal AI agent on your own hardware: MIT licensed, no paid tier, reachable from 29 chat channels, and able to browse, read and write files and run shell commands. Conditional because that reach is the risk: its docs disclaim it as a boundary between mutually adversarial users, and Unit 42 found malicious ClawHub skills, warning that installing a skill grants complete control over the agent's identity. Right for technical users who will harden and audit it; wrong for unmanaged shared deployments. Last verified 2026-11-12.

OpenClaw is an open-source AI assistant that runs on your own computer and is reached through chat apps: its README names Discord, iMessage, Slack, Teams, Telegram and WhatsApp, openclaw.ai counts 29 channels in total, and there are native apps for macOS, iOS, Android, Windows and Linux. Architecturally it is a self-hosted gateway, a single process its docs call the single source of truth for sessions, routing and channel connections, which connects those chat apps to AI coding agents. OpenClaw's site says the agent browses the web, fills forms, reads and writes files and runs shell commands, with full access or sandboxed operation as the user's choice, and remembers preferences, projects and people from one conversation to the next; the README states that state, memory and credentials live on the user's hardware. Models are bring-your-own: the site lists hosted, subscription-backed, gateway and local models, and the providers page names Anthropic, OpenAI, Google, DeepSeek, Moonshot AI, OpenRouter and LiteLLM alongside local runtimes Ollama, LM Studio, llama.cpp and vLLM. Community skills and plugins, including those on ClawHub, extend what it can do. The project is MIT licensed, stewarded by the OpenClaw Foundation, an independent 501(c)(3), and has no paid tier, hosted service or token. It installs through a shell or PowerShell script or npm and, per the README read on November 12, 2026, runs on Node 24.16.0+ on the 24.x line or 26.1.0+, with 24.22.1+ or 26.12.1+ strongly recommended. It ships releases several times a week: on November 12, 2026 the releases page showed v2026.11.3 (November 10) as the latest stable release and v2026.9.37 (November 10) as a Gateway-only extended-stable line, and the repository showed 394.4k stars. Check the releases page for the current version.

## Strengths

- Running a personal AI agent on hardware you control: the README says state, memory and credentials live on your hardware, and the site says state lives on your machine, not a vendor cloud
- Talking to an agent from apps you already use, with openclaw.ai naming WhatsApp, Telegram, Discord, Slack, Signal and iMessage among its 29 channels, in DMs and group chats
- Choosing your own model provider, from Anthropic, OpenAI, Google and DeepSeek to local runtimes such as Ollama, LM Studio, llama.cpp and vLLM
- Zero software cost: MIT licensed, stewarded by an independent 501(c)(3), with no paid tier, hosted service or token
- Hands-on computer errands, which OpenClaw's site describes as browsing the web, filling forms, reading and writing files and running shell commands
- Starting from restrictive defaults: on a regular host install the gateway binds to loopback, most channels answer unknown DM senders with a pairing code, and group access is allowlisted, with an openclaw security audit command to detect drift
- Sharing one agent or gateway between people who do not trust each other: the security docs state OpenClaw is not a hostile multi-tenant security boundary for mutually adversarial users
- Installing community skills without auditing them: Unit 42 identified five malicious skills on ClawHub that had not been blocked between February and May 2026, and warns that installation results in complete control over the agent's identity; OpenClaw deleted all five after Unit 42 reported them
- Deployments nobody will keep patched: CVE-2026-25253, rated 8.8 High, let versions before 2026.1.29 take a gatewayUrl from a query string and open a WebSocket connection without prompting, sending a token
- Non-technical users: installation runs through a shell script, PowerShell or npm and requires Node 24.16.0+ or 26.1.0+, with 24.22.1+ or 26.12.1+ strongly recommended
- Anyone who wants a vendor-hosted service, since the README states the project has none

## Pricing

- Model: open-source
- Open source (MIT): $0

Website: https://openclaw.ai

## Sources

- [OpenClaw GitHub releases (official)](https://github.com/openclaw/openclaw/releases)
- [CVE Program record, CVE-2026-25253](https://cveawg.mitre.org/api/cve/CVE-2026-25253)
- [OpenClaw docs gateway security](https://docs.openclaw.ai/gateway/security)
- [Palo Alto Networks Unit 42, OpenClaw skill marketplace supply chain research](https://unit42.paloaltonetworks.com/openclaw-ai-supply-chain-risk/)
- [OpenClaw docs overview](https://docs.openclaw.ai/)
- [OpenClaw docs model providers](https://docs.openclaw.ai/providers)
- [OpenClaw website](https://openclaw.ai)
- [OpenClaw GitHub README](https://github.com/openclaw/openclaw)

Canonical: https://neomanex.com/directory/openclaw
