Rate limits

Requests are rate-limited per agent key (fixed window). The default is set by the platform (AGENT_DEFAULT_RATE_LIMIT_PER_MIN); a key can carry its own rate_limit_per_min override set at creation.

Scope: agent_key:<key_id> (or the account for JWT callers).
Exceeding the window → 429 with error_code: RATE_LIMIT_ERROR.
The unauthenticated GET /health probe is exempt.

Handling 429

Back off exponentially and retry. Because mutations are idempotent, retrying a 429’d POST with the same Idempotency-Key is safe.

import time, httpx

def call_with_retry(client, *args, **kwargs):
    delay = 1.0
    for _ in range(6):
        r = client.request(*args, **kwargs)
        if r.status_code != 429:
            return r
        time.sleep(delay)
        delay = min(delay * 2, 30)
    r.raise_for_status()

Need more throughput for a specific agent? Mint its key with a higher rate_limit_per_min, or split work across role-specific keys.

Reference

Endpoints

Handling 429

​Handling 429

Handling 429