Cost control¶
This is the page to read first if your job is: "Is our AI spend under control, and can I show where it went?"
Like Compliance & governance, this is a story front-door: the arc, then links to the screen-level detail.
The story in one line¶
The control plane doesn't just report AI spend; it changes it at request time (cheaper model, capped budget, switched vendor) and then attributes every dollar saved or spent back to the rule, key, and team responsible. Control, then proof.
1. The obligation: what "cost control" actually requires¶
"Watching a spend graph" is not cost control. Real control is three capabilities:
| Capability | The question behind it |
|---|---|
| Act before the spend | "Can we route to a cheaper model or cap a runaway budget as the request happens, not in a monthly review?" |
| Attribute the spend | "Which key / team / rule caused this, precisely enough to charge it back?" |
| Trust the number | "If a price changed mid-month, does the historical figure stay stable, or silently re-write?" |
2. How the control plane enforces it: at request time¶
Spend discipline is enforced by routing-family rules, evaluated on every request:
- Cost-saver / vendor-switch rules send eligible traffic to a cheaper-but-acceptable model or provider.
- Spend-circuit rules trip when a budget threshold is breached; the control plane caps or redirects rather than letting spend run.
The mechanics (creating these rules, intents, priority) are on Routing. The pricing the decisions are made against (rate cards, what-if, healing) is the Cost Engine.
3. How you prove it: attribution + a stable ledger¶
- Per-rule / per-key / per-team attribution: every saved or redirected dollar is traced to the rule and principal responsible, precise enough for chargeback. The chargeback and external-BI surfaces: BI Tables.
- A replay-stable ledger: costs read from values frozen at write time. A rate card changing today does not silently re-write last quarter's figure; if you want a recompute you run Cost Healing explicitly. The mechanics: Cost Engine.
That stability is the cost-side equivalent of the compliance story's immutable record: the number you showed finance last month is the number that's still there next month.
4. Where the whole story is true at once: the dashboard¶
It reads end-to-end on the Dashboard:
- VIDAI Impact → Cost Saved: the headline dollar figure the control plane changed, click-through to the contributing rules.
- Cost Insights panel: the spend slicing and burn-rate detail.
That's the payoff. The screens above are why each figure means what it means.
Where to go next¶
- Cost Engine: rate cards, pricing, what-if, ledger replay, and Cost Healing. The screen-level detail.
- Routing: cost-saver / vendor-switch / spend-circuit rules. The enforcement engine.
- BI Tables: chargeback + external BI projections.
- Dashboard: the synthesis (VIDAI Impact + Cost Insights).
- Compliance & governance: the other flagship story, structured the same way.