Inference keys

Creating a key

Under Inference API → Keys, choose a project and create a key. The full secret is shown once at creation — copy it immediately into your secret manager. Afterwards only a masked prefix is visible.

curl -X POST https://api.upgreat.ai/v1/projects/{projectId}/inference/keys \
  -H "Authorization: Bearer <access-token>" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "production-backend",
    "rpm": 600,
    "tpm": 200000,
    "token_limit": 5000000,
    "token_limit_period": "month",
    "allowed_models": ["qwen3.6-27b", "bge-m3"]
  }'

Scoping a key

Each key can be constrained so a leak or a runaway job is contained:

• rpm — maximum requests per minute.
• tpm — maximum tokens per minute.
• token_limit with token_limit_period — a hard token budget per minute, hour, day, week or month.
• allowed_models — restrict the key to specific model IDs. Omit to allow every model the project can access.
• expires_at — an optional expiry timestamp, after which the key stops working.

Using a key

Send the key to the LLM API as a bearer token. See the LLM quickstart for full examples.

curl https://llm.upgreat.ai/v1/chat/completions \
  -H "Authorization: Bearer <inference-key>" \
  -H "Content-Type: application/json" \
  -d '{ "model": "qwen3.6-27b", "messages": [{"role": "user", "content": "Hi"}] }'

Rotating and revoking

To rotate, create a new key, deploy it, then revoke the old one with DELETE /v1/inference/keys/{keyId}. Revocation is immediate. A revoked key can be archived to hide it from the default list, and unarchived later if you need its history back.