Annos

The food diary that lives in your AI chat. Tell your assistant what you ate; Annos resolves the foods, logs them, and keeps the day's numbers. Free, open-source, and pseudonymous by design — it never asks for your real name.

→ annos.app

Two ways in, one behaviour

Annos is usable from a web app and from any AI client that speaks MCP. The two surfaces are kept at parity: anything you can do over MCP you can also do in the web UI. Logic lives once in a shared domain layer, with the MCP tools and the REST API as thin adapters over it, so the two can't drift on the things that matter — day boundaries, macro snapshots, goal targets.

• Web — a day sheet you read like a menu: meals, totals ruled off against your target, weight and goal phases.
• MCP — connect an AI client and log by chatting. Food resolution, portions in grams, and the day's totals come back on every call.

Registration and account deletion are web-only on purpose, so a hallucinating client can't create or destroy an account.

Privacy stance

What you track here is health data, so Annos is built to know as little as it can. You are a generated nickname, never a real name. Your email lives only in the sign-in system and is walled off from everything you log by a database permission — not just a promise. There is no analytics, tracking, or advertising anywhere in the app. See the in-app privacy policy for the full account.

Running it

You need Docker. From the repo root:

cp .env.example .env      # dev defaults work out of the box; no secrets needed locally
docker compose up -d      # brings up Postgres, the API, and the web app

The web app is then at http://localhost:3000 and the API at http://localhost:8000. The dev stack hot-reloads the web source; .env.example documents the handful of variables (email, Google sign-in, production secrets) you'd set for a real deployment.

Backend tests run against real Postgres in Docker:

docker compose up -d db
cd api && uv run pytest

Stack

FastAPI + FastMCP (Python 3.12), SQLAlchemy 2.0 async on Postgres, and a Next.js + Tailwind web app. Identity and OAuth are handled by Better Auth inside the web app; the Python service is only a resource server.

Data and attribution

• Food and nutrient data is from Fineli, the Finnish Institute for Health and Welfare, licensed under CC BY 4.0.
• Activity energy costs are from the 2024 Adult Compendium of Physical Activities (Herrmann et al.), pacompendium.com.

License

MIT.