The control plane between your product and every AI model.
Neural Router combines a high-performance data plane with a governed control plane, so your team can route inference traffic by cost, quality, latency, policy, provider health, and workload intent.
Six layers that turn a call into a proven decision.
OpenAI-compatible data plane
Use standard OpenAI-compatible clients for chat, streaming, embeddings, and model discovery. Keep your app code simple while routing across a governed model and provider pool.
QuickstartRequest profiler
Before scoring, the router reads modality, context size, expected output, required capabilities, session state, latency class, and governance constraints.
How routing worksDecision engine
Apply policy, filter ineligible providers, weigh cost, latency, quality, and cache affinity, then select the cheapest route that satisfies every constraint.
See the receiptsObservability and receipts
Every route leaves evidence: route reason, candidate scores, policy version, cache status, tokens, cost, latency, and overhead. Export traces to your existing tools.
Trust layerProvider quality layer
Provider claims are measured, not assumed. Endpoints are scored on conformance, availability, latency, reliability, and quality drift, and can be quarantined or auditioned.
Provider standardsBilling and accounting
Usage events, token accounting, provider costs, credits, and statements run through a micro-USD ledger, so cost reports reconcile back to individual requests.
See pricingDrop in one base URL.
Point your existing OpenAI-compatible client at Neural Router. Your application code stays the same while requests route across a governed model and provider pool.
- Chat completions, streaming, embeddings, and model discovery
- API key auth, workspace policy, rate limits, and budget checks
- Provider fallback constrained by capabilities, allowlists, and quality floors
- Route metadata returned through response headers and request logs
import OpenAI from "openai";const client = new OpenAI({baseURL: "https://api.neuralrouter.ai/v1",apiKey: process.env.NR_API_KEY,});# every other call stays the sameThe route starts by understanding the request.
Neural Router profiles modality, context size, expected output, required capabilities, session state, latency class, and governance constraints before it scores a single candidate.
Optimize for the cheapest correct path.
The router applies policy, filters ineligible providers, weighs cost, latency, quality, and cache affinity, then selects the route that satisfies every configured constraint. Nothing about that decision is hidden from you.
See a routing receiptEverything you need for production inference.
One control plane for every model.
Join the teams that measure inference like infrastructure.