Skip to main content
Peel meters two things separately. Builds (new APIs, revisions and rebuilds) count against a build allowance. Runs (calls to published endpoints) count against a run allowance. Per-actor rate limits cap how quickly you can send either.

Plans

To upgrade to Pro, go to Settings → Billing. To talk about Business, contact kaan@peel.sh.

What each operation costs

The credit_cost field on an endpoint, when present, is in runs and overrides the default weight above.

Failure semantics

Failed builds, revisions and rebuilds are not counted, including builds canceled before publication and matched builds (matched: true). A build that times out or runs out of its turn budget is still counted.

Build budget

Each build spends money on AI models, so Peel also caps the model spend on builds, revisions and rebuilds each UTC month. The spend counts whether the build succeeds or fails. When the month’s spend reaches the budget, new builds, revisions and rebuilds answer 429 quota_exceeded with details.quota set to build_budget until the first day of the next month (UTC). Builds already running finish, and endpoint calls are never affected. A build that is already queued waits while your running builds could use the rest of the budget, and fails with quota_exceeded if the budget runs out before it starts. Automatic repairs that Peel starts use Peel’s own budget, not yours. If model spend across all of Peel spikes, Peel can pause new builds for everyone until the next UTC day. Those requests answer 429 builds_paused with details.resumes_at. Builds already queued stay queued and start when the pause ends. Running builds finish and calls keep working. A run is counted only when the call returns a result with status: "success". Execution errors, timeouts, 5xx, capacity errors and 4xx errors are not charged. An upstream 4xx that the compiled program surfaces as an error is reversed. An upstream 4xx that the program deliberately returns as a successful result stays charged. Each call response includes credits_charged and credits_remaining reflecting runs used and remaining.

Run packs

Buy additional runs when the plan allowance is not enough. Purchased runs stack on top of plan runs and do not expire at the next period. Buy run packs in Settings → Billing. Auto-recharge refills runs automatically when the balance drops below a threshold. The default pack is 15,000 runs, the minimum threshold is 500 runs, the default threshold is 1,000 runs, and it buys at most 4 packs a month unless you change that under Settings, Billing, Limits (up to 12). With the daily cap below, this bounds what one month can cost.

Rate limits

Limits are per account, in a one-minute window. They scale with your plan. The values below are the defaults in the current source; deployed values can differ, and a 429 response is authoritative. The multipliers per plan (dispatch / execution / MCP): Each limited response carries the IETF RateLimit and RateLimit-Policy headers. The policy names the group, its quota q and its window w in seconds; RateLimit gives the requests left r and the seconds t until the window resets:
Every 429 carries Retry-After and a RateLimit header with r=0, including refusals from the build queue, quotas and per-address limits. RateLimit-Policy comes with it when the refusal has a fixed quota. When a refusal names no reset time, such as file_limit_reached, request_budget_exceeded or execution_capacity_exceeded, Retry-After is 30 seconds: a hint for when to try again, not when a limit resets. Browsers can read these headers and X-Request-Id on cross-origin answers. Peel does not send the older X-RateLimit-* headers. Concurrent builds are limited by plan: Free 1, Pro 4, Business 12. A global build queue also has a hard cap, so a build can return 429 with Retry-After: 30 when the queue is full even if your own limit is not reached. Sign-in, OAuth and marketplace routes have their own per-IP limits. Abusive load can be throttled or blocked under the Acceptable Use Policy.

When a limit is reached

Every limit answers 429. The code says which kind it is.

Check your quota

GET /v1/quota returns what is left this period and today. It needs a key with billing:read. The MCP tool is get_quota.
A null limit means no limit. Business accounts have no run or build limit, and no build budget unless one is agreed. build_budget resets at the start of each UTC month, even when your billing period starts on another day.

Daily cap

A daily cap stops runs after a set number each UTC day, so a loop in your code cannot spend the month in an hour. It counts runs the same way the monthly quota does, and a refunded run gives its place back. Set it in Settings under Billing, or with a full-access key:
Send {"daily_run_cap": null} to remove it. The answer is the same shape as GET /v1/quota.

Usage emails

Peel emails the account owner when this period’s runs reach 80% and then 100% of the plan, once at each level per period. At the end of each billing period it sends a summary of the runs and builds used. Change either in Settings under Billing, or with the same PUT /v1/quota:
usage_alerts takes up to five whole percentages from 1 to 100. An empty list turns alerts off. Accounts with no run limit get no alerts.

Check your balance

Expected response shape:
The peel credits CLI command and the MCP tool get_credits return the same state. The response keeps all existing fields and adds plan_id, builds_allowance, builds_used, builds_remaining, runs_allowance, runs_remaining and period_start.