Rate limits are evidence budgets

Unlimited APIs encourage clients to scrape. Helix would rather you ask a precise question.

Each plan is an evidence budget:

• Explorer: 100 API calls / day, 10 / minute, 5 Ask Helix answers, 10 Helix Reviews / day
• Team: 2,000 / day, 60 / minute, unlimited Ask, 200 reviews / day
• Scale: 20,000 / day, 300 / minute, unlimited Ask, 2,000 reviews / day

Over quota, the API returns 429 with X-RateLimit-* headers. The workspace Ask limit is counted separately so a person exploring the UI does not starve an agent in CI.

if (used >= plan.limits.apiDaily) {
  return new Response("budget exceeded", { status: 429 });
}

Why this is product, not billing trivia

An agent that fires a thousand “who knows X” questions per build is not using the graph. It is brute-forcing it. Rate limits force the same discipline Helix asks of humans: retrieve, cite, stop.

Explorer is for trying the sample and one public repo. Team is for a working group wiring the API into review. Scale is for org-wide agents that already know which questions matter.

Create a key in Settings. Send it as Bearer hx_live_…. Spend the budget on questions you would ask a senior engineer in Slack.

Helix Review shares the same idea: a daily budget of verdicts, not an unbounded firehose of comments. Webhooks still run. They stop being useful when they become noise.

Evidence is expensive to reconstruct and cheap to waste. The budget exists so the graph stays a place you ask, not a place you scrape.

See the limits on Pricing, then open Helix and create a key when you are ready to spend them on real questions.