One endpoint. Three LLM protocols. Any upstream.
LLM Gateway normalizes OpenAI Chat, Anthropic Messages and OpenAI Responses traffic onto whichever upstream you configure — with hot-reload routing, failover, context trimming and a live usage console. It performs no inference itself: you bring your own upstream keys.
- Protocol conversion — codex, zcode and any OpenAI-compatible client on one gateway
- Streaming synthesis — non-streaming upstreams become protocol-correct SSE
- Zero marginal cost — the gateway routes; it never calls a model on its own
- Usage console — per-protocol counters, upstream health probes
Managed hosting is in preparation. Meanwhile the full gateway ships as a self-host package:
git clone && cp config/routing.example.json config/routing.json && ./run.sh
Part of the PZTools network · Runs anywhere Python 3.11+ runs · No telemetry