At a Glance:
OpenClaw is a local-first personal AI assistant gateway that unifies multi-channel messaging across platforms like WhatsApp, Telegram, Slack, and Discord, runs on your own devices with macOS, Linux, Windows, iOS, and Android support, and includes agent routing, voice wake, a live canvas, and sandboxed multi-user safety defaults.
Overview:
OpenClaw is a personal AI assistant designed to run locally on your own devices as a single-user gateway that connects to the messaging channels you already use. It acts as a control plane that routes inbound messages from supported channels — including WhatsApp, Telegram, Slack, Discord, Signal, iMessage, and others — to configurable AI agents. The assistant supports text, voice (with wake words on macOS/iOS and continuous voice on Android), and a live agent-driven canvas. Companion apps are available for Windows, macOS, iOS, and Android, though the gateway itself provides the core experience. Its security model defaults to DM pairing for untrusted senders on most channels and supports sandboxing non-main sessions via Docker, SSH, or OpenShell backends.
Key Decision Points:
Runs locally on your own devices: The gateway runs as a user service on macOS, Linux, or Windows, not as a cloud-hosted service.
Multi-channel, single inbox: Supports 20+ messaging channels including WhatsApp, Telegram, Slack, Discord, Signal, iMessage, and WeChat, accessible through one gateway.
Agent routing model: Channels, accounts, or peers can be routed to isolated agents using workspaces and per-agent sessions.
Security defaults for inbound DMs: Unknown senders receive a pairing code by default on Telegram, WhatsApp, Signal, iMessage, Discord, Slack, Google Chat, and Microsoft Teams; public DMs require explicit opt-in.
Sandboxed session execution: Non-main sessions can run inside Docker, SSH, or OpenShell sandboxes with tool allowlists, restricting access to host-level tools like the browser and cron.
Companion apps are optional: The gateway delivers the core experience; macOS, Windows Hub, iOS, and Android apps add menu bar control, voice forwarding, and node capabilities.
Core Features:
Multi-channel messaging gateway: Sends and receives messages across WhatsApp, Telegram, Slack, Discord, Google Chat, Signal, iMessage, IRC, Microsoft Teams, Matrix, Feishu, LINE, Mattermost, Nextcloud Talk, Nostr, Synology Chat, Tlon, Twitch, Zalo, Zalo Personal, WeChat, QQ, and WebChat.
Multi-agent routing: Routes inbound channels, accounts, or peers to isolated agents with separate workspaces and per-agent session storage.
Voice Wake + Talk Mode: Supports wake words on macOS and iOS, plus continuous voice interaction on Android using ElevenLabs with system TTS fallback.
Live Canvas: Provides an agent-driven visual workspace using A2UI that the assistant can render and control.
DM pairing security model: Unknown senders on supported channels receive a short pairing code; messages are only processed after explicit approval via CLI.
Sandboxed non-main sessions: Supports running non-main agent sessions inside Docker, SSH, or OpenShell sandboxes with configurable tool allowlists that deny browser, canvas, nodes, cron, discord, and gateway access by default.
Use Cases:
Single-user personal assistant: Developers or individuals running a local AI assistant that answers across the messaging apps they already use daily.
Multi-channel message routing: Users who want one assistant to handle incoming messages from different platforms and route them to separate agent workspaces.
Local-first voice assistant: macOS and iOS users who want a voice-triggered assistant with wake words; Android users who want continuous voice interaction.
Sandboxed agent experiments: Developers exposing the assistant to group chats or external contacts who want non-main sessions to run inside container sandboxes with restricted tool access.
Open-Source Alternative Value:
OpenClaw provides a local-first, self-hosted gateway for running a personal AI assistant across multiple messaging channels without depending on a cloud service. The gateway runs as a user service on your own devices, and all session routing, channel connections, and agent configuration stay under your control. Optional companion apps extend the experience to macOS menu bar control, Windows system tray, and mobile nodes without requiring a hosted backend. The sandboxing model lets you isolate non-main sessions in Docker or SSH environments with explicit tool restrictions, giving you a practical way to limit agent access when exposing the assistant beyond personal use.
