Budgets and rate limits

BYOK doesn't mean unlimited. Budgets cap how much a runaway conversation can accidentally spend; rate limits prevent a buggy or abusive caller from draining your provider quota.

Budgets

Configure per-provider in Setyenv → Agent → Settings → Budgets:

• Daily token budget — a hard cap on input + output tokens for that provider per UTC day.
• Monthly cost cap (USD) — an approximate cap based on the provider's published per-token pricing.

When a budget is hit, the agent refuses new turns against that provider with error: budget_exceeded and the admin sees a notice in Setyenv → Agent → Health. Budgets reset on the configured cadence.

Rate limits

Applied per-user, per-provider on a sliding window:

• Turns per minute (chat turns started by this user).
• Tool calls per minute (aggregated across all turns).

Hitting a rate limit returns error: rate_limited with a Retry-After hint. The chat UI surfaces this as a soft inline warning rather than a crash.

Provider back-off

Independently of your budgets and rate limits, the agent respects the provider's own Retry-After and 429 responses with exponential back-off. You will not double-charge yourself by hammering a throttled endpoint.

Where to start

The defaults are conservative on purpose: high enough to be unobtrusive for normal use, low enough to keep a runaway agent loop from doing financial damage. Tighten them in production once you have a baseline for your actual usage.