AI without a door
A model in a notebook, no auth, no limits. FastAPI becomes the door.
APIs & AI - FastAPI
Lean, typed APIs for integrations and AI orchestration. OpenAPI included - without a framework cathedral.
When you need a clear HTTP edge - in front of models, jobs or several clients.
A model in a notebook, no auth, no limits. FastAPI becomes the door.
Shop, CRM, internal tool. One contract, versioning, error codes.
Clean up incoming events, sign them, pass them on.
You do not want to migrate - just a fast edge in front of it or beside it.
Contract, security, operations.
Pydantic models, OpenAPI, auth (token/OIDC), validation.
RAG pipelines, prompt routing, fallbacks. See AI adoption.
Long-running work in queues, status endpoints, webhooks back.
Bitrix24, Shopware, Symfony/Laravel - as clients, not as copies.
Before we wire up models, the API contract is in place. Clients should build against a schema, not against our mood.
Spec, slice, hardening.
Endpoints, errors, auth. In writing.
One flow in staging against real data.
Rate limits, timeouts, observability, PII.
Deploy, metrics, optional maintenance.
APIs you can quote.
The spec does not lie. Breaking changes are real work, not just “some Wednesday”.
Costs, timeouts, human in the loop - part of the design.
FastAPI orchestrates, it does not replace your CRM.
API questions, briefly.
Close to Python and AI: FastAPI. A JS team and a BFF in front of an SPA: often Node.js. Sometimes both, clearly separated.
Django when the admin or CMS lives in Python. We know Flask. For APIs plus AI, FastAPI is usually the shorter route.
When data protection or costs call for it. Otherwise API models plus RAG.
URL or header, with visible deprecation. Clients get advance notice.
Token, OIDC or m2m - depending on the client. No open endpoints “because it is internal”.
One contract plus one flow. A fixed scope after the kickoff.
Related technologies we often use alongside.
Tell us the clients and the first flow - we outline the spec and the pilot.