Usage & token quota

Each plan carries an invisible token budget — an internal cost guardrail, not a commercial dimension. You never buy tokens; the budget simply smooths spend and blocks runaway loops.

How the budget is sliced

The monthly budget for your tier is split into weekly then short rolling windows, and is non-cumulative:

Consumption control

Before spending any tokens, the middleware applies three checks:

1. Access gate — is the plan active and the budget not exhausted? If not, the review is short-circuited and the PR status shows upgrade / credits.
2. Anti-loop — more than 3 reviews in 15 minutes on the same PR is blocked (protects against CI ↔ AI loops).
3. Hard ceiling — a PR changing more than the max line count (e.g. 1000) is rejected immediately, 0 tokens spent (generated/vendored files).

Degraded mode

When a window is tight, non-critical work (summaries, style) is deferred and only critical findings are produced, so security and correctness reviews still run.

Related

• Billing
• Roles & permissions