Connect a model provider key
EvalGate is bring your own key (BYOK). It does not bundle model inference
credits. To use LLM judges, the Playground, online monitors, or the Copilot, you
connect a provider key (OpenAI, Anthropic, a gateway, etc.). That provider bills
your usage directly.
Step 1 — Open settings
In the sidebar, click Settings, then choose Provider Keys.
Step 2 — Add a key
- Click Add provider key.
- Pick your provider (for example, OpenAI or Anthropic).
- Paste the API key.
- Give it a label you will recognize later, like
prod-openai-2026-07.
- Click Save.
EvalGate stores the key encrypted. Never paste a key into a screenshot, a
commit, or a chat. If a key leaks, revoke it at the provider and add a new
one.
Step 3 — Check the Model Gateway
Model-backed features route through the Model Gateway, not the provider
directly. In Settings, open Model Gateway and confirm:
- Your provider key is listed and healthy.
- The models you want to use are synchronized (visible in the model list).
- The organization budget is not zero if you want calls to run.
Step 4 — Test a call
Still in Model Gateway, click Test call, pick a model, and send a trivial
prompt. If you get a response and a cost record, the key works end to end.
Troubleshooting
provider_unhealthy — the key is wrong, expired, or lacks the model. Fix
it at the provider, then re-save.budget_exhausted — raise the organization budget in Settings, or wait for
the budget window to reset.model_not_synchronized — click Sync models in Model Gateway, then
retry.- Calls work locally but not in CI — CI uses a different API key. Add a key
for the CI environment, or set the key as a CI secret.
Up next
Now you can create an evaluation that uses
an LLM judge.