Beta docs: the shipped API is POST /api/v1/generate for gemini and sarvam. Some pages below describe planned features. See the shipped quick start.
Accounts and keys
| Question | Answer |
|---|
| Do I need my own provider accounts? | No. With a managed key OpenLink pays providers and itemizes the cost. BYOK is available per provider. |
| How do I rotate a key? | Create a new key, deploy it, then revoke the old one. Revocation is instant and logged. |
| Is a key shared across providers? | Yes — one key authorizes all enabled providers. |
Language and coverage
| Question | Answer |
|---|
| Which languages do the models cover? | Claude and Gemini cover general languages; Sarvam adds Hindi and 22 more Indian languages plus speech. |
| Can I force a specific model? | Yes, per request — the provider and model fields, or the x-openlink-provider header. |
| What happens if a provider is down? | Automatic routing moves to a healthy provider. Forced providers return 503 instead. |
Data and billing
| Question | Answer |
|---|
| Do you store my prompts? | No — request bodies are not persisted by default. Log retention is configurable. |
| How is billing itemized? | One invoice; each provider appears as its own line with token counts. |
| Where are requests processed? | In the in-1 region today. More regions are on the roadmap. |
Compatibility
| Question | Answer |
|---|
| Is your API OpenAI-compatible? | The response schema deliberately follows OpenAI's Chat Completion shape, so current clients parse it. |
| Can I migrate an existing codebase? | Usually: point base_url and auth at OpenLink. The request/response contract covers most chat flows. |