Platform
Limits
This page collects the limits that are documented elsewhere on this site, so you can find one without knowing which page holds it. It is for developers planning traffic and for anyone sizing a job. It adds no limit of its own: each row links to the page that explains it.
Model calls
| Limit | Value | Page |
|---|---|---|
| Rate limit per key | None of Mubit's own. Your provider's limits apply. | Streaming, timeouts and limits |
| Connect to the provider | 10 s | Time limits |
| First byte from the provider | 610 s | Time limits |
| Silence in an answer | 15 minutes | Time limits |
| Total time of a call | 60 minutes | Time limits |
Retries on 503 or 529 | Up to 2 more, for a request body of 4 MiB or less | Retries |
| Size of a request | No limit of Mubit's own | Sizes |
| Stored of a request | Up to 32 MiB | What Mubit records |
| Stored of a reply | The first 4 MiB and the last 256 KiB | What Mubit records |
Management API
| Limit | Value | Page |
|---|---|---|
| Requests | 600 a minute for each key | Rate limits |
| List task training examples | Also 120 a minute for the whole organization | Rate limits |
| Page size of a list | 1 to 200, default 50 | Pagination |
| Lifetime of a cursor | 24 hours | Pagination |
| Idempotency window | 24 hours, with a key of 1 to 255 characters | Retries |
| Usage history | 61 days before today | Usage, billing and activity |
| SDK retries | Up to 2, on a 429 or a 5xx answer | Manage Mubit |
| Rating comment | Up to 500 characters; a correction up to 8,000 | Manage Mubit |
Every answer from a rate-limited route has RateLimit-Limit, RateLimit-Remaining and RateLimit-Reset headers. Past a limit, the answer is 429 rate_limited with Retry-After in seconds.
Runtime calibration
| Limit | Value | Page |
|---|---|---|
| Rate limit | 60 POST calls and 600 GET calls a minute for each key | Runtime calibration |
| Wait for the store | Up to 60 seconds | Keys, records and limits |
| Session ID | 1 to 512 characters | API reference |
Adapters
| Limit | Value | Page |
|---|---|---|
| Size of a call an adapter answers | At most 256 KiB, within a context of 8,192 tokens | What to expect |
| First training | At 25 kept examples, with at least 12 more kept back for the exam | What to expect |
| Examples in one training | At most 250 | What to expect |
| Automatic rating | Up to 200 calls of each task a day | What to expect |
| Your rating comment | Up to 500 characters | What to expect |
| Spot-check copies | About 2% of the calls a version answers, at least 10 and at most 500 a day | Costs and data |
| Routing decision | The first 2,000 characters of the instructions and of the last user message | Costs and data |
Console
| Limit | Value | Page |
|---|---|---|
| Sign-out after inactivity | 15 minutes | Members and roles |
| Import window | 7, 14 or 30 days | Import past calls |
| Requests range | Last 24 hours, 7 days or 30 days | Find calls in Requests |