# Rate limits and quotas The API is free; its limits act as a safeguard. The plan is assigned by Kelki and can be raised case by case. ## Limits per plan | Plan | Requests / min (per key) | Requests / day (per account) | Real requests / day | Agent messages / day | Webhooks | | `decouverte` (sandbox only) | 30 | 2,000 | 0 | 50 | 2 | | `standard` | 120 | 20,000 | 50 | 500 | 5 | | `partenaire` | 600 | 200,000 | 500 | 5,000 | 10 | Your plan, limits and usage: `GET /account` and `GET /account/usage` (92 days at most). ## Shared limits - Per IP address, before authentication: 120 requests per minute. - Anti brute force: beyond 20 refused keys in 10 minutes, the IP address is blocked for a while. - Per contact: at most 3 requests per 24 hours for the same phone number or e-mail address, across all Kelki channels (`contact_limit_reached`). - Global cap: 200 API requests awaiting confirmation per hour, all accounts together. Do not work around these limits (several accounts, multiplied keys, rotating addresses): the [acceptable use policy](https://kelki.ai/developers/en/terms/acceptable-use/) forbids it. ## Headers ``` RateLimit-Policy: "minute";q=120;w=60 RateLimit: "minute";r=118;t=42 ``` `RateLimit-Policy` describes the policy applied to the key, `RateLimit` the remaining quota (IETF draft `draft-ietf-httpapi-ratelimit-headers`). ## On a 429 response - `rate_limited`: per-minute rate exceeded; wait for the `Retry-After` duration, in seconds, before retrying. - `quota_exceeded`: daily quota reached; no point retrying before the next day. - Space out retries (increasing waits) and never resend a creation without the same `Idempotency-Key` ([idempotency](https://kelki.ai/developers/en/idempotency/)). ## About this page - URL : https://kelki.ai/developers/en/rate-limits/ - Language : en - fr : https://kelki.ai/developers/fr/rate-limits/ - en : https://kelki.ai/developers/en/rate-limits/ - x-default : https://kelki.ai/developers/ - Portal : https://kelki.ai/developers/llms.txt