Getting Started with SafeAgent
SafeAgent is a self-hosted AI agent framework that runs entirely on your machine. It provides a chat interface where you can give natural language commands, and your AI agent executes real actions — reading email, scraping websites, posting to Slack, managing GitHub repos.
What You Can Do
- Chat with AI — natural language interface, like ChatGPT but connected to your real services
- Read & send email — works with Gmail, Outlook, Yahoo, or any IMAP/SMTP server
- Scrape any website — extract text, links, prices, emails from any URL
- Search the web — DuckDuckGo integration, no API key needed
- Manage GitHub — list repos, create issues, track PRs
- Post to Slack — read channels, send messages
- Notion pages — search and read your workspace
- Control via Telegram/Discord — send tasks from your phone
Key Principles
- Self-hosted — everything runs on your machine, nothing goes to the cloud
- Zero config to start — ships with local AI (Ollama + Mistral 7B), no API keys needed
- Encrypted — all credentials encrypted with AES-256-GCM at rest
- Judgment engine — every action is risk-scored; dangerous actions require approval
- UI-configured — all services connected through the dashboard, no .env editing for integrations
Requirements
| Requirement | Details |
|---|---|
| Docker | Docker Desktop (Mac/Windows) or Docker Engine (Linux) |
| RAM | 8GB minimum (4GB for Ollama + Mistral) |
| Disk | ~6GB (Docker images + AI model) |
| OS | macOS, Linux, Windows (WSL2) |
SafeAgent works on Apple Silicon (M1/M2/M3/M4) and Intel/AMD machines.