Resources¶
Everything around the product rather than in it: how to get help, how to contribute, and how to extend the platform.
Help AgenticOS¶
The project is young and the fastest way to help it is to use it and say what broke.
-
What breaks, the sequence that triggers it, and how you would know it was fixed. That third one is the part most issues are missing.
-
Check the issue map first — it says what is already planned and what has no scope yet.
-
It is the cheapest signal that this is worth continuing.
Development — contributing¶
Read the page that matches what you are touching. They are the repository's own engineering notes, published rather than rewritten, so what you read is what contributors read.
| Working on | Read |
|---|---|
| Anything at all, first | Architecture · The console's code — routes → services → repositories, and the request's transaction |
| A shape you have seen before | Patterns |
| A feature, end to end | Adding features |
| A test, or a red coverage gate | Testing |
| A pull request | Code review and Branching |
Three things that will bite you
A repository commits with db.flush(), never db.commit(). Background work
that reads a row the request just wrote is handed over with
spawn_after_commit, never spawn. And the platform layer is held at 100%
coverage, enforced in CI. All three are in
Architecture and Testing, and all
three have been got wrong here at least once.
Extending the platform¶
Adding to what AgenticOS itself can do, in Python. These are contributor recipes — you need the repository checked out.
- Add a capability — a new tool the model can call
- Add a server to the MCP catalog
- Add an API endpoint
- Add a background task
- Add a sync connector
Before you write a capability
Ask whether it is really an MCP connection. A capability that would be an API client for one SaaS product is a server somebody has already written, and taking it into this repository means maintaining it against that product's API forever.
Built on¶
AgenticOS is generated from the Full-Stack AI Agent Template, and it stands on:
- Pydantic AI — the agent runtime
- FastAPI and Pydantic — the backend
- pgvector — retrieval
- Prefect — background work
- Next.js — the console
- Model Context Protocol — every integration
Licence¶
Apache 2.0. See
LICENSE.