Skip to main content
The limits you are most likely to reach. Plan-dependent limits are listed for Pay as you go, Pro, and Enterprise; see plans. For limits you set yourself, see spend limits.

API rate limits

Requests to the REST API, including those the CLI and SDK make, are limited to 1,000 per minute per member, shared by all of their API keys, with bursts of up to 500. A refused request returns 429 with a Retry-After header; wait that long and retry. The web app and the Default key that hosted runs carry are not limited. When the service is busy, a request can also return 503 with Retry-After. Retry it the same way.

Model calls

Model calls through HUD’s gateway are limited per team: Only a hosted run’s calls to the model it is evaluating are exempt. Everything else counts: scripts, agents that run on your own machine, even against a hosted environment, and calls a hosted run makes to other models, such as an LLM grader’s. Calls with your own provider key count too. A refused call returns 429.

Runs

Plans don’t set a fixed number of runs a team can have at once.

Traces

A trace stores the steps a run records and the screenshots, files, and video attached to them: Data past a limit is refused with 413 and is missing from the trace. A computer-use agent that takes one screenshot per step reaches the image limit after 500 steps.

Files and imports