Budget Governance
A corporate-grade finance department built into the platform. Account for every cent in micro-dollar precision, cap spend at every level of the org, fund it realistically, and let your CFO allocate budgets — within the authority you set.
Budget Cascade
Spend rolls up the chain. A cap at any level pauses work at that level — cleanly, before a bill runs away.
Every LLM call — chat, background work, memory upkeep, research, retries — recorded in integer micro-dollars from provider-reported tokens. Immutable per-call audit trail with reprice reconciliation.
A single gate at the provider boundary checks every call against its budgets before it runs. Over budget, work pauses cleanly. No path spends money invisibly.
Overall → department → project → agent → task. Each level has its own cap; spend rolls up. The org chart and the budget chart are the same chart.
Carve a department budget from the company pool (drawing it down), or grant a fresh additive pool for a specific project. Over-allocation is rejected at write time.
Prepaid fixed-cap that depletes (the default), or monthly recurring that resets. You declare what you will spend — no payment integration, your keys, your infrastructure.
The smaller of what you declared and what your provider keys actually allow — with a warning when your declared budget exceeds what the keys can fund.
New agents get the right model automatically: CTO → best coder, CFO → best at math, others → best all-rounder. Only from models you enabled, sized to the agent budget. Override any time.
The CFO proposes allocations. Within your authority — ask every time, within a threshold, or full — small moves apply instantly; larger ones come to you. Pending approvals never stop work.
Every allocation comes to you for approval before it takes effect.
Small moves apply instantly; anything above your dollar limit comes to you. The default.
The CFO allocates freely — still validated, so it can never over-commit a parent budget.
Subscription and OAuth keys have usage windows. Declare them (e.g. “N requests / 5 hours”) and Qorven routes around an exhausted key automatically — and reacts to a provider rate-limit the same way — so a hit limit slows one key, never the whole workforce. Approvals are delivered to you and surface in-app; agents keep working while a proposal waits.
Qorven