AI Spend Controls
Default: mock, $0
Section titled “Default: mock, $0”AI runs in mock mode by default (AI_LIVE_CALLS_ENABLED=False). Mock output
is deterministic and clearly a sample; workflows stay fully exercisable with
no model cost.
Live calls need three gates
Section titled “Live calls need three gates”AI_ENABLED— global kill switch.PAID_EXTERNAL_CALLS_ENABLED— human approval gate for paid calls.AI_LIVE_CALLS_ENABLED— live-vs-mock switch, default off.
The cap
Section titled “The cap”A hard monthly spend limit (default $60) is enforced in a shared ledger, so the ceiling holds across processes and survives restarts. Per-user token and prompt quotas, per-source sync caps, and an output-token clamp bound usage inside the cap.
Cost arithmetic is published
Section titled “Cost arithmetic is published”Per-workflow cost estimates (bid workflow, full proposal, auto-sync ceiling) are computed from real code paths and list prices, not estimates. The numbers live with the platform’s planning documents and are updated whenever call volume changes.