Bring your own keys (BYOK)

With BYOK, requests to a model are served with your own provider key against the provider's official endpoint. Inference charges appear on your own provider account, and request data stays within your existing provider agreements — useful where procurement or compliance requires a direct provider relationship.

Zero lock-in by design: the keys are yours, and your application talks to an OpenAI-compatible endpoint. You can point base_url back at the provider's official endpoint at any time.

• Your application keeps using a single BoostRail API key; models are selected with the model parameter of each request. BYOK does not add any new key to your code.
• Provider keys you store are backend inventory — your application never touches them. When a model you call is covered by one of your keys, BoostRail uses your key against the lab's official endpoint and counts one request against your quota.
• Keys are shared at the organization level: multiple BoostRail API keys (per app or environment) share the same provider keys, monthly quota, and plan.
• Order: keys marked prioritized (in your order, moving to the next on failure) → platform routes (on by default; needs a balance) → keys marked fallback → an error only when all of these fail.
• Both kinds of exhaustion switch over automatically: if your provider account runs out of credit, the request falls back to platform lines on the spot; if your monthly BYOK quota is used up, traffic runs on platform lines until the quota resets on the 1st of the month (UTC); models that run only on your own key stop until then. No application changes either way.

• Plan fees are deducted from the wallet balance — a one-off top-up is enough; no card subscription. Auto-renewal can be switched on or off at any time; when it is off the plan simply ends with the month.
• If the balance cannot cover renewal, the organization falls back to the Free plan automatically and can re-purchase after topping up.
• In-console notices are sent at 80% and 100% of the monthly quota. Over-quota requests switch to platform lines at the published platform prices — no hidden markup — and the quota resets on the 1st of the month (UTC).

• Prioritized keys come first, tried in the order you set; a failed key moves the request to the next one.
• Platform routes (on by default): when your prioritized keys fail or are rate-limited, requests continue on platform routes of the same model. A balance is all that is needed.
• Fallback keys: keys you mark as fallback are tried after platform routes.
• Use my keys only: per lab, you can turn platform routes off for strict data boundaries. When your keys are exhausted, requests fail instead of falling back.
• A small wallet balance works as fallback insurance — optional, and not required to use the free quota.

Anthropic, OpenAI, DeepSeek, Kimi (Moonshot), GLM (Zhipu), Qwen (Alibaba), Grok (xAI), Gemini (Google), MiniMax, Mistral, Cohere, Perplexity, AI21, Groq, and Aion Labs — fifteen labs, with the in-console list as the source of truth as more open up. Each provider can hold multiple keys with priority ordering and automatic rotation.

• Keys are stored encrypted (AES-256-GCM); interfaces and logs only ever show the last four characters, never a full key.
• Rotate or delete keys at any time — deletion takes effect immediately.
• Control stays with you: revoke or cap a key in the provider's own console at any moment, effective instantly, with no need to trust any promise of ours.
• Keys are isolated per organization and used only for that organization's requests. Every call and every key change is audited.

• Leading models are priced at or below their official list prices on platform routes. Live per-token prices are on the Models page.
• Models not covered by BYOK are available on platform lines with no setup.
• Why some models are not covered: BYOK needs the lab to be on the supported list and the model to be available on that lab's official API. Other models run on platform routes only; coverage grows with the catalog.