Skip to content

Limits and usage

Limits apply across the account and its keys. Creating another key does not increase capacity or reset usage.

ResourceLimitScope
Active API keys5Account
New keys5 per hourAccount, rolling hour
Key lifetime30, 90, or 365 days; 90 defaultKey
Pro code searches1,000 per dayAccount, UI and all keys together
API search starts10 per minuteAccount, fixed minute windows
Active API searches1Account
Workspace-list requests60 per minuteAccount, fixed minute windows
API ingress requests120 per minuteClient IP, fixed minute windows
Query length1–2,048 trimmed charactersSearch
JSON body4,096 bytes, known lengthRequest
Search hits100 maximumStream
Search runtime45 seconds maximumStream
Event frame65,536 bytes maximumStream
Event output1,048,576 bytes maximumStream

Search allowance resets at 00:00 UTC. API execution requires current eligible Pro access even though a Free account can create and manage keys. Consult your account’s usage display for its current entitlement and consumption.

Fixed minute windows mean requests either side of a minute boundary can occur close together. Do not rely on that behavior to sustain a burst. Shared service capacity is separate from these account limits and may be exhausted even when your account has remaining allowance.

Read the JSON error code and Retry-After response header. Wait at least the indicated number of seconds, then use a bounded retry policy. For search_quota_exceeded, the daily reset or a change to account allowance is required; repeatedly retrying after a short delay will not restore the budget.

Prefer a queue with one active search per account. Share that queue across application processes rather than creating independent high-concurrency clients for each API key.

The UI and API share the daily search ledger. A request is reserved before execution. Failed requests before stream headers are handled with a refund; after streaming begins, partial results and client cancellation still count. A retry is a separate search. Listing workspaces does not consume this search allowance, but does consume metadata and ingress request budgets.

Limits are operational bounds, not a latency or corpus-coverage guarantee. Read the start event’s advertised limits and retain terminal completeness metadata.