Bring your own key

In the app, go to Settings → Providers, paste your key, and it's tested against the provider before being saved. Keys are encrypted (AES-256-GCM) before they're written to the database, and the API never returns the original key again -- only a masked hint like sk-...ab12.

Once connected, call the model exactly like a hosted one -- just use its slug (e.g. openai/gpt-5.1, anthropic/claude-sonnet-5, google/gemini-2.5-pro, mistral/mistral-large-latest). If no key is connected for that provider, the request fails with 424 provider_not_configured instead of silently falling back to anything else.

BYOK usage is billed by the provider directly to your own account, at their rates. LLM Studio doesn't see or charge for token counts on BYOK traffic -- usage accounting still records request counts and latency for your own visibility, but not cost.

If the provider starts rejecting your key (revoked, expired, out of credit), the next request fails with 424 provider_invalid_credentials and the key's status in Settings → Providers flips to Invalid so you notice without having to check logs.