OpenClaw - Open-Source AI Personal Assistant

OpenClaw

Open-source personal AI assistant with multi-channel messaging, persistent memory, and reliable task execution

See https://github.com/openclaw/openclaw

Features

• Multi-Channel Access: WhatsApp, Telegram, Discord, Slack, iMessage, web UI, CLI, macOS app - one AI assistant accessible everywhere
• Hub-and-Spoke Architecture: Central Gateway coordinates all messages and maintains state
• Persistent Memory: Remembers context across conversations and channels
• Reliable Execution: Lane Queue prevents race conditions - serial by default, parallel only for safe tasks
• Multi-Assistant Routing: Run different assistants for different channels/teams with isolated configurations
• Memory System: JSONL transcripts (audit trail) + Markdown memory files (knowledge base)
• Browser Control: Chrome automation for web interaction, screenshots, data extraction
• Task Execution: Shell commands, file operations, API calls, web automation
• Session Management: Approval system for new contacts, permissions per channel
• Rich Media: Images, audio, documents, full attachments support
• Dashboard UI: Web interface to browse history, manage settings, view active sessions
• Scheduled Tasks: Cron jobs for recurring automation
• Extensibility: Community skill marketplace (ClawHub)

Superpowers

One AI Everywhere: Access the same assistant from WhatsApp, Slack, Discord, iMessage - single conversation state across all platforms.

Production-Ready: 46,000+ GitHub stars, proven in thousands of deployments, 150,000+ active instances worldwide.

Reliable Operation: Lane Queue philosophy ensures complex workflows don’t have race conditions or state corruption. Safe by default.

Full Transparency: JSONL transcripts of everything - complete audit trail to see exactly what happened.

Extensible: Community skills marketplace + ability to modify code directly.

Team-Ready: Different teams/departments can route to specialized assistants with different personalities and tools.

Architecture

Six-Stage Processing Pipeline:

1. Channel Adapter - Standardizes inputs from different platforms
2. Gateway Server - Routes messages to appropriate sessions
3. Lane Queue - Enforces serial execution by default (prevents conflicts)
4. Agent Runner - Selects AI model, assembles context, manages tokens
5. Agentic Loop - AI proposes actions, system executes, results fed back
6. Response Path - Streams responses back to user and archives to JSONL

Configurable Personality:

• AGENTS.md - Core instructions and boundaries
• SOUL.md - Personality and communication style
• TOOLS.md - Available capabilities
• Pi Agent Core - Base system instructions

Multi-Assistant Support:
Different channels route to different assistants (Discord with Claude Sonnet + friendly personality, Telegram with GPT-4 + formal tone, each with own tool access).

Security

• Command-level allowlists - Blocks dangerous command patterns
• Pattern blocking - Prevents shell redirections, command substitution, subshells
• Session isolation - Unknown contacts are sandboxed
• Approval system - New users must be approved
• Note: Application-level security (see NanoClaw for OS-level container isolation)

Deployment

Requirements:

• Docker or native Node.js setup
• Messaging platform credentials
• AI model API key

Setup Time: 5-15 minutes with Docker

Hardware: Mac mini-class (~$500)orVPS($10-50/month)

Memory: >1GB RAM

Model Support

• OpenAI (GPT-4, GPT-5)
• Anthropic (Claude Sonnet, Opus)
• Google (Gemini)
• Custom/local models
• Multi-model per assistant (different models for different channels)

Use Cases

• Personal Assistant: Handles emails, calendar, tasks, reminders
• Team Coordination: Different departments with specialized assistants
• Business Automation: Customer support, lead management, workflows
• Research & Writing: Access to web, can draft with real-time information
• Home Automation: Control and coordinate smart home devices

Pricing

Software: Free and open-source

Operating Costs:

• VPS (if not self-hosting): $5-50/month
• AI API usage: Varies by model choice

Comparisons

• vs NanoClaw: More features + convenience vs stronger security; choose OpenClaw for full capabilities, NanoClaw if security is critical
• vs PicoClaw: Full-featured personal assistant vs minimal embedded system
• vs Kimi Claw: Full control + self-hosted vs one-click simplicity
• vs II-Agent: Multi-channel focused vs multi-model focused

Strengths

✅ Production-ready with large community
✅ Extensive platform support (15+ channels)
✅ Full browser control and automation
✅ Reliable execution (Lane Queue pattern)
✅ Multi-assistant orchestration
✅ Transparent audit trail (JSONL)
✅ Established ecosystem (ClawHub)

Limitations

⚠️ Large codebase (430K+ lines) hard to fully audit
⚠️ Complex dependency tree (52 modules, 45 dependencies)
⚠️ Application-level security (not OS-isolated)
⚠️ High resource requirements
⚠️ Single process means one vulnerability could affect entire system

Getting Started

1. Clone: git clone https://github.com/openclaw/openclaw.git
2. Docker setup: ./docker-setup.sh
3. Connect messaging platform (WhatsApp, Telegram, Discord, etc.)
4. Select AI model (OpenAI, Claude, Gemini)
5. Start using via your preferred messaging app

Related

• NanoClaw - Security-hardened version
• PicoClaw - Ultra-minimal embedded version
• Comparison: OpenClaw vs NanoClaw vs PicoClaw
• Personal AI Systems
• Multi-Channel AI Assistants