LLM Routing Platform
LLM infrastructure·2025·Solo build
A gateway that sits between an application and several LLM providers, so calling code targets one interface instead of learning each vendor's SDK, auth model, and failure modes. Routing, key handling, and usage accounting live in the gateway rather than being reimplemented per service.
The constraint
Applications that talk to more than one model provider end up with provider logic scattered across the codebase — different auth, different request shapes, different error semantics. Swapping or adding a model becomes a refactor instead of a config change.
How it was built
- –One request shape normalised across providers, so model choice becomes configuration rather than code
- –Request routing and model selection at the gateway, keeping provider decisions out of application logic
- –Centralised API key management so credentials live in one place
- –Usage tracking per request, making cost attributable by model and caller
- –Turborepo monorepo on Bun for fast builds across the gateway and its clients
Stack
Turborepo
TypeScript
Bun.js
LLMs