Skip to main content
Two kinds of key matter. Your FlexInference key says who you are. Your provider keys pay for the tokens.

​
Your FlexInference key

Create it in the dashboard. It starts with flex_live_ and you see it once, so copy it then. Send it as a bearer token.
A malformed or unknown key returns 401 invalid_api_key. A revoked key returns 401 revoked_api_key, so you can tell the two apart.

​
What a key can carry

A plain key holds your identity and nothing else, so each request sends start_within itself. An agent key also holds a duration, plus a retry policy when you ask for one. We fill in whichever field the request left out, and a value in the body always beats the key. The response header x-flexinference-defaults-applied lists what the key filled in. Edit those defaults in the dashboard and live traffic picks them up within seconds. You keep the same key. Revoking a key clears its defaults with it.

​
Your provider keys

We don’t resell inference. Add each provider key once in the dashboard, and your provider bills you for the tokens.
  • openai takes an OpenAI key, sk- prefix.
  • google takes a Gemini API key from Google AI Studio.
  • anthropic takes an Anthropic key, sk-ant- prefix.
  • vertex takes a Vertex express key, AQ. prefix.
  • bedrock takes a Bedrock inference key, ABSK prefix, plus a region.
  • foundry takes an Azure AI Foundry key, plus its resource name.
  • cloudflare takes a Workers AI token, plus your account id.
We store every provider key encrypted, never log one, and never return one. You never send one on a request. Every route you name in a backup chain needs its own key. We refuse a missing key before any upstream call, and we never fall through to the next route. Rather not paste provider keys at all? Managed Keys serves on ours and your balance pays.