API reference
Rate limits and quotas
Limits keep one caller from filling the queue. The backend analyses one video at a time and the others wait.
| Limit | Value | When you pass it |
|---|---|---|
| Requests per API key | 10 a minute on Free (fixed one-minute windows); see the plan table below | 429 rate_limited with Retry-After |
| Playground calls per account | 6 a minute | 429 rate_limited |
| Free video minutes | 1,000 a month per account, reset on the 1st (UTC) | 429 monthly_quota_exceeded |
| A cap you set on one key | Any number of video minutes a month | 429 key_cap_exceeded |
| Video size | 25 MB | 413 file_too_large |
| Video analysed per call | 120 s (use start_s to walk a longer file) | Analysis stops at 120 s |
| Call duration | 5 minutes (about 20 s of video per call is safe on the CPU backend) | 504 inference_timeout |
| Sign-in attempts | 10 per 15 minutes per network and per email | 429 rate_limited |
Model calls carry X-RateLimit-Limit, X-RateLimit-Remaining and X-RateLimit-Reset (Unix seconds), and on a capped plan X-Quota-Limit-Minutes and X-Quota-Remaining-Minutes. GET /v1/me returns the same for the calling key. Sample calls (?sample=) count as requests but never as minutes.
By plan
| Plan | Requests a minute per key | Video minutes a month | Active keys | Logs kept |
|---|---|---|---|---|
| Free | 10 | 1,000, then calls stop | 20 | 30 days |
| Pay as you go | 60 | No cap | 20 | 90 days |
| Always on | 120 | No cap | 50 | 90 days |
| Enterprise | 600 | No cap | 200 | 90 days |
Billing is not switched on, so every account is on Free today. The paid rows are what the API will enforce when it is.
Versions
The path carries the major version (/v1). Every response also carries a dated Vitals-Version header, currently 2026-11-21. Changes that could break a client get a new date and a line in the changelog.