We gave our AI agent fleet a credit limit, and it hit it the same day

Description and use

The system automatically capped autonomous agent spending by replaying event logs into ledgers and applying a rolling budget threshold, preventing overspending.

Problem

Uncontrolled AI agent fleet spending exceeded budget, risking runaway costs.

Outcome

The system automatically prevented overspending by throttling new work once a rolling 5‑hour budget was exceeded, providing audit trails, real‑time alerts, and a fail‑open design that avoided paralysis.

Workflow

• 1. Replay board events into three hledger journals (money, promises, labour).
• 2. Validate ledgers with hledger check and independent replay for parity.
• 3. Compute rolling 5‑hour labour spend per mind using mesh‑labor.
• 4. Price spend via token‑to‑USD rate table (mesh‑ledger).
• 5. Monitor spend against configured cap with mesh‑labor‑alert (edge‑triggered).
• 6. When spend exceeds cap, mesh‑pace throttles new task dispatch until spend falls below threshold.
• 7. Operators can manually raise cap or bypass throttle via dedicated channels.
• 8. Ensure environment variables for cap are sourced in all entry points to avoid stale values.

Evidence

The board gets replayed into three separate double-entry hledger journals, each tracking a different commodity: money, promises, labour.

Source

mesh‑pace reads that rolling spend and, once it crosses the cap, holds every dispatch of new board work.

Source