Secure OpenClaw is a self-hosted personal AI assistant that lives on your messaging apps. You message it over WhatsApp, Telegram, Signal, or iMessage and it replies via Claude (or an Opencode-backed model) with full tool access, persistent memory, scheduled reminders, and app integrations through Composio's tool router. It runs as a gateway process, either locally or on a small VPS via Docker, plus an optional local terminal chat.
Its security posture is opt-in by default: each platform has DM and group allowlists (messages from unrecognized senders are silently dropped), and tool use runs in a default permission mode where the assistant asks for approval -- in the terminal you confirm y/n, and on messaging platforms it sends a "Reply Y to allow" message, with approvals timing out after two minutes. Memory is file-based (a long-term MEMORY.md plus daily and topic notes) loaded at conversation start and written only when you ask it to remember something; reminders persist as cron jobs that fire while the gateway runs.
Integrations cover 500+ apps via Composio (Gmail, Slack, GitHub, Notion, calendars, and more) with per-app OAuth on first use. Two AI providers are supported and switchable: the Claude Agent SDK (Opus/Sonnet/Haiku models, requires Claude Code) and Opencode as an open-source alternative with its own model lineup.
