Every model. One door.
Each call goes to the right model, with your keys and the budget you set.

Summarise 40 support tickets
- Anthropic
- OpenAI
- Gemini
The right model, every call.
One place for every provider, every key and every budget.
Learned routing.
Picks the best model for quality, cost or speed, call by call.
20+ providers.
And any compatible API.
Budgets on every key.
Virtual keys with their own budgets and limits.
Stays up.
Retries, circuit breakers and hedged requests when a provider falters.
The details.
- Drop-in APIsSpeaks the OpenAI, Anthropic and Gemini APIs.
- CachingAn exact cache, and prompt-cache affinity.
- ResidencyResidency profiles, and a dedicated instance pinned to your region.
- Your own keysBring them, held in your KMS or Vault.
- Batch at half priceWork that can wait costs half as much.
- Savings you can proveWith OpenTelemetry on every call.
Where Router sits.
One stack, from the agent to the ground it runs on.
Runs where you run.
From the public cloud to networks sealed from the world.
01CloudLive in minutes.
02EdgeBeside the work.
03On-premYour servers. Your walls.
04Air-gappedNothing gets out.
- Keys stay in your vault
- Region-pinned instances
- Budgets you set
- OpenTelemetry built in
Questions.
Do I need to change my code?
Usually not. Router speaks the OpenAI, Anthropic and Gemini APIs, so your client points at Router and keeps working.
Can I use my own provider keys?
Yes. Bring your own keys, held in your KMS or Vault.
Where does my data go?
Where you say. Residency profiles keep calls in region, and each customer can have a dedicated instance pinned to a region.
Which providers can it reach?
More than 20, among them OpenAI, Anthropic, Azure OpenAI, Bedrock, Gemini, Vertex, Mistral, Groq and DeepSeek, and any compatible API.



