API spend limit: can you set a monthly budget on Sume?

No: the Sume docs describe no monthly or account-wide spend limit setting. The limits they describe work at other levels: the workspace's spendable balance, a spend cap on each run, and plan-based concurrency. A monthly budget is something you enforce in your own code by reading usage and balance.

This comes from Format API errors, Create a Format run, Agent Completions and Generation admission, read 2026-09-29.

Which spend controls does Sume document?

Four, and they bound different things. None of them is a calendar-month ceiling.

How does a per-run spend cap work?

On a Format run, generation_spend_cap_usd names that run's ceiling, up to the $500 platform maximum. Send nothing and the run inherits the Format's cap; send null and it takes the $500 maximum. 0 or a value above 500 is a 400.

On an Agent Completion the cap is required. It has no default, and omitting it returns 400 invalid_request. The docs call the cap the substitute for the interactive spend-approval prompt that a backend caller does not get. The best-practices page says to treat a missing cap as a client bug. For the cap mechanics, see the spend-cap post.

How do I enforce a monthly budget myself?

Read GET /v1/balance before you submit a batch, and stop when the balance is under your floor. The response is USD-denominated, and available_amount_usd_cents is the spendable amount. For actual spend, add up the debited_usd of the runs you started, or read the usage ledger as in the per-run cost post.

BAL=$(curl -sS https://api.sume.com/v1/balance \
  -H "Authorization: Bearer $SUME_API_KEY" \
  | jq -r '.data.balance.available_amount_usd_cents')

if [ "$BAL" -lt 5000 ]; then
  echo "balance low: $BAL cents" >&2
  exit 1
fi

Does topping up the wallet raise other limits?

No. Generation concurrency is plan-only, and prepaid top-ups do not raise it. Top-ups are dashboard operations: the public API has balance and usage reads, not a top-up call. A manual top-up is Stripe-backed, and the usage ledger can show it afterwards.

What does the wallet alone not do?

It stops a submit only when the estimated cost cannot be reserved. It does not stop a single run from using a large share of a balance, which is why the per-run cap exists. It also does not warn you as it drains; the pages checked describe no low-balance alert, so poll the balance yourself.