Architecture

TinyFat runs your AI agent in an isolated cloud container with persistent storage. Here’s the high-level picture.

How it works

1. You message your agent — via Email, Telegram, Slack, or web chat
2. Platform wakes your container — with your storage attached
3. Agent processes your message — reads context, thinks, uses tools
4. Agent responds — on the same channel you messaged
5. Container stays warm — for follow-up messages, then sleeps

All channels share the same agent — same memory, same files, same brain.

Key properties

• Isolated containers — Each agent runs in its own sandbox. Agents can’t see each other’s data.
• Persistent storage — Your agent’s files survive between conversations and restarts.
• Multi-channel — Email, Telegram, Slack, and web chat all reach the same agent.
• Scheduled tasks — Your agent can set up recurring or one-time events. The platform wakes it at the right time.
• Extensible — Skills and tools can be added to your agent’s workspace.

Open source

The agent runtime is fully open source:

• crawdad: github.com/tinyfatco/crawdad — Agent runtime
• fat-skills: github.com/tinyfatco/fat-skills — Platform skills (browser, PDF, etc.)

You can self-host the runtime if you want full control.

Next steps

• Memory & Persistence — How data survives restarts
• Agent Workspace — Your agent’s filesystem
• Skills & Tools — What your agent can do