Skip to main content
There are no API keys to copy, store or leak. Your agent signs in to your Summer account in the browser, once, and you choose Allow with a spend limit. The sign-in renews itself.

​
Sign in

The browser opens the Summer sign-in page. Sign in, check which app is asking, pick a monthly spend limit, and choose Allow. From then on every summer api command uses this sign-in:

​
Call the API over HTTP

summer token prints a one-hour access token from the same sign-in. Send it as a bearer token:
Run summer token again for a fresh token; the sign-in behind it stays valid. Reading models (GET /v1/models) and pricing a request (POST /v1/estimate) need no sign-in. Your own app. Summer’s sign-in is standard OAuth 2.1 with PKCE, the same one the Summer MCP uses. Discovery: https://mcp.summerengine.com/.well-known/oauth-protected-resource. Send the access token your app gets as Authorization: Bearer <token>.

​
Spend limits per app

Every app or agent you connect has its own spend limit, chosen on the Allow page: 5, 20 or 50 US dollars a month, your own amount, or no limit of its own. Change it at any time in Settings > Connected agents & apps, where each app shows its limit, what it spent, when it was last used and when its access ends. Revoke stops the app at once; it works again only after you connect it again.

​
How spending works

  • Paid credits. Every call is paid from your paid Summer credits: your plan’s credits and your top-ups. Top up at any time.
  • A failed job costs nothing. Most jobs are charged when they succeed. A job that is charged when it starts, such as a video, is refunded in full when it fails.
  • Hard limits. A call that would go over the app’s limit or your paid balance is refused with 402 before any work starts. Several calls at once never go past a limit.
  • Never twice. Send an Idempotency-Key header with each job. A retry with the same value returns the first answer and charges nothing more.

​
Read the app’s limit and your balance

Answers data for the app behind this token: name, client_id, scopes, limit_usd, limit_reset, spent_usd, remaining_usd, resets_at, total_spent_usd, held_usd (held for this app’s queued and running jobs until they are charged; it counts against remaining_usd), expires_at and created_at. The app can lower its own limit with PATCH /v1/key ({"limitUsd": 2}), never raise it, and sign itself out with DELETE /v1/key.
Answers data with balance_usd (your paid credits), held_usd (held for your queued and running jobs), key_held_usd (the part of it that is this app’s), spendable_usd (what this app can spend now, holds already taken off), key_remaining_usd, key_resets_at and top_up_url. All amounts are US dollars. null means no limit. Both calls are in the API reference: Read the app’s limit and Read your balance.

​
No rate limits on paid usage

Paid work has no request limits per minute, hour or day. Send as many jobs as your credits and the app’s limit cover.
  • Running slots. Each account runs 10 jobs at once at the start, and up to 40 as it buys more credits. A job past your slots answers 202 with state: "queued" and an id that starts with q:. It starts in order as a slot frees up. Queued work is accepted: follow it, do not send it again.
  • Held price. A queued or running job holds its price against your balance and the app’s limit until it is charged, so a queue never spends past either one.
  • Money is the only refusal. 402 no_credits (the paid balance) or 402 over_limit (the app’s own limit), at once and before any work starts.
  • Many at once. POST /v1/batches takes up to 500 jobs in one call, all or none. See batches.

​
Keep tokens safe

A token spends your credits for an hour. Keep it in the environment, never in game code, a build or a commit.