Rate limits

Rate limits protect Caedral's infrastructure and ensure fair access. Limits apply per API key. Exceeding a limit returns HTTP 429 with type rate_limit_exceeded.

#Current limits (API gateway)

These limits are enforced in the API gateway with sliding-window counters. Requests billed against a team seat also draw down a shared per-team bucket, so one seat (or one compromised key) cannot consume the team's entire capacity. Throughput is separate from subscription pools: rate limits control request frequency; included pools control how much paid inference you can run each billing period.

#Handling 429 responses

1import { Caedral, CaedralAPIError } from "caedral";2 3async function callWithBackoff(caedral: Caedral, prompt: string) {4  for (let attempt = 0; attempt < 5; attempt++) {5    try {6      return await caedral.chat.completions.create({7        model: "deepseek/deepseek-v4-flash",8        messages: [{ role: "user", content: prompt }],9      });10    } catch (err) {11      if (12        !(err instanceof CaedralAPIError) ||13        err.type !== "rate_limit_exceeded" ||14        attempt === 415      ) {16        throw err;17      }18      await new Promise((r) => setTimeout(r, 2 ** attempt * 1000));19    }20  }21}

#Best practices

• Implement exponential backoff with jitter on 429 responses
• Queue or batch requests rather than unbounded parallel calls
• Use a fast model such as deepseek/deepseek-v4-flash for high-QPS paths
• Keep concurrent SSE streams under the per-key cap — finish or cancel an active stream before opening more
• Monitor usage via GET /v1/usage or the dashboard billing page