Documentation
Build on one API. Reach every model.
Neural Router is an OpenAI-compatible gateway that routes each request to the best model across 400+ LLMs — by cost, latency, or quality — with governance, observability, and failover built in.
What is Neural Router?
Neural Router sits in front of 60+ providers and 400+ models behind a single, OpenAI-compatible endpoint. You send a chat completion request; the router scores eligible models against your objective and policies and dispatches to the best one, automatically failing over if a provider degrades.
Because the API mirrors the OpenAI Chat Completions schema, you can point an existing OpenAI SDK at Neural Router by changing only the base URL and API key — then layer routing, budgets, caching, and residency on top.
Start here
Quickstart
Make your first routed request in minutes.
Provider routing
Load balancing, fallbacks, and provider filters.
Model routing
Auto-select models by objective.
Models
Browse the catalog and pin or restrict models.
Structured outputs
Type-safe JSON responses.
API reference
Endpoints, parameters, and schemas.
Core concepts
- One key, one schema, every major model.
- Route by quality-per-dollar, lowest-cost, lowest-latency, or highest-quality.
- Ordered fallback chains and automatic provider failover.
- Budgets, guardrails, semantic caching, data residency, and an audit log per workspace.