TinyAGI Architecture and Runtime
1. The workspace shape​
The README already reveals the core product shape:
| Area | Why it matters |
|---|---|
| agents | isolated workers with specialized roles |
| teams | collaboration layer for handoffs and fan-out |
| channels | external messaging surfaces |
| SQLite queue | atomic jobs, retries, dead-letter handling |
| TinyOffice | browser dashboard and visual control plane |
This is a bigger runtime than a normal single-assistant terminal app.
2. The runtime mental model​
At runtime, TinyAGI usually works like this:
- a message or task enters the system,
- the queue persists and schedules it,
- an agent or team processes it,
- results are streamed into TinyOffice and logs,
- responses can return through channels or stay in the portal.
3. Why the queue matters​
The SQLite queue is not a boring implementation detail. It is part of why TinyAGI can support:
- parallel work,
- retries,
- background operation,
- and more durable 24/7 behavior.
4. Why TinyOffice matters​
TinyOffice is not just a dashboard. It is the human control surface for:
- chat,
- agents and teams,
- tasks,
- logs,
- and settings.
That makes it central to the runtime, especially for non-terminal operators.
5. What to read first in code​
Start with:
- the README quick start,
- queue and daemon surfaces,
- team and agent code,
- TinyOffice-related areas,
- channel adapters last.