oll.am — the platform, at a glance

What it is

oll.am is an umbrella platform for shipping small AI products on shared infrastructure. It is one monorepo with a frozen Core — auth, billing, and email, deployed and live at core.oll.am — plus a set of small product services that own only their product logic and call Core (and the oll-model gateway) over HTTP for everything else. Products carry zero auth / Stripe / email code, so Core can be built once, deployed once, and never touched again, while each product stays tiny and disposable.

Positioning: any AI model, pay once — no subscription.

The exit strategy is modular: each service is its own deploy unit, so you can sell one product or the whole platform.

Architecture in one picture

• Core is frozen infrastructure — build once, deploy once, never touch.
• The gateway (oll-model) is a frozen model-call relay — one OpenAI-compatible surface in front of swappable providers ("any model, pay once").
• Product services are product logic only — they delegate auth/billing/email to Core and (where they need an LLM) the model call to the gateway, over HTTP.

Service map

# cd core — or services/oll-model · services/write · services/foto
cd core
pip install -r requirements.txt -r requirements-dev.txt
pytest

Repo conventions

Monorepo, one Dockerfile per service, independent deploys

One repo, unified version control. Each service folder has its own Dockerfile, requirements*.txt, openapi.yaml, tests/, and pyproject.toml. There is no shared Python package — services integrate only over HTTP, never by importing each other or touching another service's database (database-per-service on Neon; stateless services have none).

HTML is the interface

Live project status is not tracked in prose. The single source of truth is state/state.json, rendered by scripts/state.py into the Control Room at site/oll-am-status.html. site/ is the deployed build-in-public landing + docs (flat static HTML, served by the root Dockerfile → nginx). specs/ holds the API contract; CLAUDE.md carries the full living context.

CI/CD — the "dark factory" (merge ≠ deploy)

• CI = GitHub Actions, path-filtered. A change only runs the jobs it touches — Core gates (core/**), the services matrix (services/**), and the site design-conformance gate (site/**).
• CD = Coolify on the Hetzner VPS. Each service is a separate Coolify app keyed on its Base Directory + Watch Paths, so only the changed service redeploys.
• main = production, auto-deploy OFF = frozen. A stable service on main is never disturbed. New work rides a feat/<service> branch → PR preview → merge graduates it to production. Merging a PR does not deploy it — deploy is a separate Coolify action.

Run a service locally

cd services/foto            # any service folder
pip install -r requirements.txt -r requirements-dev.txt
pytest                      # run its test suite
python app.py               # or: flask --app app run — boots the Flask app

Product services expect Core reachable over HTTP (locally via the shared ollam Docker network → http://oll-core:5000, in production → https://core.oll.am); write additionally calls the gateway (OLL_MODEL_BASE_URL). For the full local stack, see docker-compose.local.yml and docker-compose.core.yml.

Where the live status lives

• Control Room — site/oll-am-status.html (generated from state/state.json): what's running, branch, DBs, what's awaiting a decision, and the current single next action.
• CLAUDE.md — the full living project context + operating procedure.
• core.oll.am — the live, frozen Core.