OpenClaw

A free, self-hosted AI assistant for technical users who want an agent inside their chat apps, but it needs hands-on setup, careful security hardening, and separately paid model usage.

• Self-hosted Gateway that connects chat apps such as WhatsApp, Telegram, Slack, Discord, Signal, and iMessage to an AI agent
• Persistent memory that keeps preferences, projects, and people across conversations
• Agent can browse the web, fill forms, read and write files, and run shell commands, with full or sandboxed access
• Works with hosted, subscription-backed, gateway, or local models, with provider failover
• Community skills and plugins from ClawHub, scanned before listing, plus custom skills
• Shared team Gateway so members can open and steer the same sessions
• Exec approvals for shell commands and an `openclaw security audit` command to check for config drift
• Desktop apps for macOS and Windows, a web dashboard, and iOS and Android companion apps
• Install by script, npm, Docker, Nix, or from source

• Built for one trusted operator, not hostile multi-user use

  The security guide assumes a single trust boundary per gateway. Tools run on the host unless sandboxing is set up, and third-party ClawHub skills sit outside the project security scope.

  Source

• Credentials can be stored as plain text

  API keys can sit in plain-text config files that the agent can read. SecretRefs reduce this, but they are opt-in and must be set up for each credential.

  Source

• Running costs depend on your model and hosting

  The software is free, but model usage is billed by your provider through your own API key, and always-on use may need a server. The docs give no cost figures, so budgets are hard to predict.

  Source