CoWork-OS is a self-hosted operating system for running personal AI agents across multiple messaging platforms. Security-first architecture. Multi-provider support. Full control over your data.
Core Philosophy
Most AI agent platforms optimize for convenience. CoWork-OS optimizes for control.
You run it on your infrastructure. You choose which models to use. You own the conversation history. You set the security policies.
Features
Multi-channel support:
- Telegram
- Discord
- Slack
- iMessage
One agent, accessible from all your communication channels. Conversations sync across platforms.
Multi-provider:
- Claude (Anthropic)
- GPT (OpenAI)
- Gemini (Google)
- Ollama (local models)
Switch providers per-task or per-channel. Use local models for sensitive data, cloud models for complex tasks.
Security-first:
- End-to-end encryption for stored conversations
- No external logging or telemetry
- Sandboxed execution environment
- Granular permission controls
Architecture
CoWork-OS runs as a containerized service. You deploy it on your server (or run it locally). It connects to your messaging platforms via official APIs.
The core components:
- Router: handles incoming messages from all channels
- Agent Runtime: executes tasks using your chosen model
- Memory Layer: stores context and conversation history
- Scheduler: runs background tasks (cron-style)
Everything runs in Docker containers. Configuration is code (YAML files). Updates are git pull + restart.
Use Cases
Personal assistant:
Manage your calendar, tasks, and reminders from any messaging app.
Knowledge base:
Query your notes, documents, and research from Telegram or WhatsApp.
Automation:
Schedule recurring tasks. The agent checks your email, summarizes updates, posts to your social accounts.
Private AI:
Run entirely local models. No data leaves your machine.
Comparison to Alternatives
vs OpenClaw:
OpenClaw has more features and a larger ecosystem. CoWork-OS is simpler, more opinionated, and security-focused.
vs Gaia:
Gaia is proactive (Jarvis-style). CoWork-OS is reactive (you initiate, it responds).
vs Custom setup:
CoWork-OS gives you the infrastructure. You don’t build the message router, the scheduler, the memory layer. You configure and deploy.
Deployment
Requirements:
- Docker and Docker Compose
- 2GB RAM minimum (more for local models)
- API keys for messaging platforms
- API keys for AI providers (or Ollama for local)
Setup is ~15 minutes. The repo includes a deployment guide and example configs.
The Bigger Picture
The personal AI OS space is fragmenting. Every framework makes different trade-offs:
- OpenClaw: feature-rich, complex
- Gaia: proactive, resource-heavy
- CoWork-OS: security-first, multi-channel
- zeroclaw: minimal, local-only
No dominant platform yet. Pick based on your threat model and workflow.
CoWork-OS is for people who want control > convenience. If you’re comfortable running Docker and managing your own infrastructure, it’s a solid foundation.
GitHub
CoWork-OS is open source and actively maintained.
Repository: https://github.com/CoWork-OS/CoWork-OS
Stars: 102 (as of 2026-02-23)