Rate limits

Requests per minute per key, set by plan, and an IP limit on Terminal.

Govern / rate-limits

Each key has a per-minute request limit. Your plan sets the ceiling, and you can set a lower limit on any key.

Over the limit, the gateway returns 429 rate_limited.

json

{ "error": { "type": "rate_limited", "message": "..." } }

Retries in the OpenAI SDKs

Both OpenAI SDKs retry 429 responses with backoff, twice by default. Raise the count if your agent sends bursts.

const client = new OpenAI({
  baseURL: "https://YOUR-PROVN-HOST/v1",
  apiKey: process.env.PROVN_API_KEY,
  maxRetries: 5,
});

python

client = OpenAI(
    base_url="https://YOUR-PROVN-HOST/v1",
    api_key=os.environ["PROVN_API_KEY"],
    max_retries=5,
)

Terminal runs have no key, so Provn limits them by IP: 12 runs per 10 minutes.