OpenClaw

repo openclaw Apr 9, 2026

Personal AI assistant platform you run on your own devices. The gateway is a single long-lived TypeScript daemon listening on WebSocket (ws://127.0.0.1:18789) that acts as the control plane for 24+ messaging channels (WhatsApp/Baileys, Telegram/grammY, Discord/discord.js, Slack/Bolt, Signal, iMessage, Matrix, IRC, Microsoft Teams, Google Chat, LINE, Feishu, Mattermost, Nextcloud Talk, Nostr, Twitch, Zalo, and more), an embedded Pi agent runtime with model failover, native companion apps (macOS/iOS/Android) connecting as capability “nodes”, and a skills platform.

Typed RPC protocol with 100+ methods over WebSocket, role+scope authorization (operator vs node), rate limiting on write methods, and event broadcasting for presence, health, and agent streaming. Channels implemented as typed adapter plugins with 20+ optional interfaces (gateway, outbound, pairing, security, groups, threading, heartbeat, agentTools) — composition over inheritance. Plugin SDK boundary (openclaw/plugin-sdk/*) strictly separates core from extensions; 55+ bundled plugins in the extensions workspace.

Agent runtime embeds inference in-process: session history, auth profile rotation, model fallback, context compaction, tool dispatch, and sandbox policy enforcement. Skills are markdown SKILL.md files with frontmatter injected into the agent prompt as XML. Configuration via JSON5 with env substitution, includes, zod validation, legacy migrations, and backup rotation.

v2026.4.10, MIT license, TypeScript ESM, Node 24 runtime, ~13,000 files. Sponsored by OpenAI, GitHub, NVIDIA, Vercel, Blacksmith, Convex.

Backlinks

Agent Exec Policy A three-axis policy model for human-in-the-loop approval of agent tool execution. The three axes are independent and composable: what commands are structurally permitted, when to prompt the human, and what to do when the human doesn't respond. →

Agent Learning Loop A closed learning loop in an AI agent that enables self-improvement across sessions. Three components working together: →

Channel Adapter Pattern Design pattern for implementing messaging channel integrations in AI agent systems. Each channel implements a typed set of optional interfaces (adapters) that the core system calls at well-defined points, rather than inheriting from a monolithic base class. →

Context Window Compression A technique for maintaining long-running agent conversations within fixed context windows by selectively summarizing older turns while preserving recent context and key decisions. →

Credential Pool Pattern A pattern for managing multiple API credentials for the same provider with automatic failover, enabling higher throughput and resilience for AI agent infrastructure. →

Multi-Channel AI Gateway Architecture pattern where a single daemon process acts as the control plane for an AI agent, connecting it to multiple messaging platforms simultaneously while keeping agent logic unified. →

OpenClaw Source summary of OpenClaw, a personal AI assistant platform that runs on your own devices and answers on channels you already use. →

Sleep-Phase Memory Consolidation A background memory consolidation process for AI agents modeled on human sleep phases. Short-term signals accumulate over multiple sessions and days; only high-confidence, multi-evidence memories are promoted to durable long-term storage. The system runs offline (typically cron-triggered at 3 AM), not during conversations. →

Tool Loop Detection A content-aware system for detecting when an AI agent is stuck in a tool-calling loop, distinguishing between legitimate polling (where results change) and stuck loops (where the agent repeats the same action with no progress). →