Running it yourself

How many credentials do I need?

None. The application starts with zero, /api/health names every variable that would enable something, and the router routes around whatever is absent. A website audit runs entirely on sources that need no account.

What does adding a key change?

It brings a provider into rotation, and nothing else. There is no code path a credential unlocks — adding one and restarting is the whole enablement story.

Which database?

SQLite by default, so history survives a restart on a machine with nothing installed. PostgreSQL in production is a DATABASE_URL change with no code to touch. /api/health reports which one is actually in use, and whether it is ephemeral — a container writing to a SQLite file forgets everything on the next deploy.

Why is monitoring off by default?

A monitor that started itself would begin spending every existing key's quota the moment a new version deployed. That is not a decision a version bump should make on an operator's behalf.

Can I run more than one replica?

Yes, with one caveat: the monitor loop has no cross-process lock, so two replicas with monitoring enabled run every scheduled check twice and bill the owner twice. The rate limiter is also per-process — it protects one instance from a burst rather than enforcing a global limit. Spend is capped in the database, which is the part that actually guards money.