Skip to content
Gbolagade Ishola
Project

Multi-Provider LLM Proxy

One endpoint for every LLM provider, with automatic failover and cost control built in.

The problem

Wiring a product directly to one model provider is a standing liability. When that provider has an outage, raises prices, or changes a model, the whole product feels it, and nobody can say what a single request actually costs.

What I built

A FastAPI gateway that puts one messages-first endpoint in front of every provider. It routes each call, fails over automatically, normalises responses, estimates cost per request, and logs everything, with new providers added through a small registry. It ships in Docker with test coverage.

My approach

I modelled every provider behind one messages-first contract, so the application never knows which model answered. Fallback, normalisation, and per-request cost accounting live in the gateway, with a pluggable registry and pytest coverage keeping new providers cheap to add.

The result

Switching or combining providers becomes a config change instead of a rewrite, and every call is observable and costed.

Stack

PythonFastAPIOpenAIAnthropicDockerpytest

Connect

Building something, or thinking about it? Book a 1:1 and let's dig in.

Gbolagade Ishola