The BYOK LLM gateway for every model
Bring your own provider keys to one OpenAI-compatible endpoint for every model. Automatic failover, per-key spend limits, PII redaction. 0% markup.
294+ models from 16+ providers, on your own provider keys at the provider's list price.
Bring your own key: each request runs on your own provider account, the provider bills you at its list price, and Autonomous Relay adds 0% on top. Your keys are encrypted with AES-256-GCM under a data key wrapped by Google Cloud KMS.
Why a gateway instead of calling providers
One client for every provider
One OpenAI-compatible SDK and one base URL, instead of an SDK and an integration per provider.
Failover you don’t write
Name a fallback chain in the request. When a provider errors or times out, the next healthy one takes over before any byte reaches your client.
One ledger for spend
A ceiling on every key and every request’s cost recorded, instead of a separate bill from each provider at the end of the month.
One OpenAI-compatible endpoint for every provider
Text, images, audio and tool calls behind a single OpenAI-compatible endpoint. Keep your model strings; change the base URL. Browse all
Automatic fallback across providers
Health-aware routing with automatic fallback. When a provider degrades, the next hop is already chosen — your users never see the retry. How fallback works
0% markup and per-key spend limits
0% markup on inference. Spend ceilings per key, holds settled against a ledger you can reconstruct to the cent. View pricing
Guardrails before a prompt leaves
Guardrails decide which providers may see a prompt: allowlists, data regions and PII redaction before anything leaves. View guardrails
Two lines of diff.
Any OpenAI-compatible client. Declare a fallback chain and the router picks the next healthy hop before your users notice. Quickstart
Questions
- Who bills me for inference?
- Your provider, directly, at its list price, on your own key. Autonomous Relay adds 0% markup. Solo is free; Pro, Team and Business are a flat $29, $99 and $399 a month per organisation.
- How are my provider keys stored?
- Encrypted with AES-256-GCM under a data key generated for that key, which is itself wrapped by a master key held in Google Cloud KMS. A stored key is never returned by the API.
- Do you store my prompts?
- No. Prompts and responses pass through the gateway in memory and are not stored or logged. We keep each request’s metadata (model, provider, tokens, cost) for 24 months. What a provider keeps is set by your own agreement with it.
- What happens when a provider degrades?
- Routing is health-aware. When a provider errors or times out, the request moves to the next candidate in your fallback chain, before any byte has reached your client.
- How do spend ceilings work?
- Each API key can carry a ceiling. A request is held against it before it runs and settled at its real cost afterwards, so a key cannot spend past its limit, and every hold and settlement is recorded in a ledger.
- What does Team add over Solo?
- Solo includes unlimited keys, failover, spend ceilings per key, and built-in PII redaction and prompt-injection detection. Team adds guardrails with custom patterns, block mode and budgets, client workspaces and reports, usage alerts, ledger export, and presets and classifiers.