Limits
Every enforced number in one place: rate limits, request sizes, lifetimes, workspace and host limits.
Values marked workspace are defaults that a workspace owner can change under Usage > Limits; the console shows your workspace's current values there. Values marked host are set per worker host by the operator; the ones shown are the current pilot host's.
Requests
| Limit | Value |
|---|---|
| Requests per API key | 20 per second, burst 60 (429 rate_limited) |
| Failed authentications per source IP | 1 per second, burst 20 |
| Request body | 4 MiB |
| pyautogui code per action | 16384 characters |
sleep action | at most 60 s |
| Gateway request timeout | 950 s (the SDK's default timeout is 900 s) |
GET /v1/slots page size | 1 to 5000 (500 by default in the paged form) |
| Task id prefix | at least 8 characters |
Computers
| Limit | Value |
|---|---|
| Live computers | 10 (workspace) |
| Desktops on the host | 10 (host) |
| Desktop size | 4 vCPU, 4 GiB memory, no swap (host) |
Default lifetime when ttl is absent or <= 0 | 3600 s (workspace) |
| Maximum lifetime | 14400 s (workspace); the console's lifetime presets stop there |
| Minimum lifetime through the API | 60 s |
| Idle timeout | 1800 s by default, never longer than the lifetime |
| Guest reply timeout per action | about 2 minutes, then 502 (GuestError) |
Saved states and forks
| Limit | Value |
|---|---|
| Saved states on the host | 5 (host), about 0.6 GB of host memory each |
| Copies per fork | 1 to 64 in the API, at most 8 on the host (host) |
| Saved-state idle lease | 1800 s by default (idle_timeout on save, state_idle_timeout on fork) |
| Quiet-tree wait before a capture | 30 s by default (quiesce_timeout, at most 300 s), with a 1 s quiet window |
Runs (console)
| Limit | Value |
|---|---|
| Parallel runs | 2 (workspace) |
| Copies per task, fork copies | 1 to 8 (and at most the workspace's run parallelism, 8 by default) |
| Tasks per run | 1 to 366 |
| Steps per attempt | 1 to 50 |
| Rounds | 1 to 50 |
| Prefix steps before a fork | 0 to 20 |
| Time cap | 1 to 1440 minutes |
API keys
| Limit | Value |
|---|---|
| Expiry | 1 to 365 days, or none; 90 by default in the console |
| Rotation grace | 1 hour in the console (the old key then expires) |
| Name | 1 to 80 characters |
Spend
| Limit | Value |
|---|---|
| Spend cap | per period (month or day, UTC), on measured compute cost (not billed); none by default (workspace) |
| At the cap | creates and forks are refused (403 budget_exceeded), live computers are closed and active runs cancelled |