AI usage & cost
Partner / Manager view of token, call, and per-user AI usage.
The AI usage dashboard tracks how much the firm leans on Kiravy AI — by feature, by user, over time. Useful for capacity planning, spotting one teammate who's gone all-in on AI for everything, or just understanding usage patterns.
Open from Firm settings (avatar menu) → AI usage.
Partner or Manager only. This page is for internal firm visibility; Staff and Viewer roles redirect to the dashboard.
What the page shows
Four sections, all scoped to your firm:
1. This month — top metrics
- Calls — total AI API calls so far this calendar month (UTC)
- Tokens — total input + output + cache tokens
A call is one round-trip to Kiravy AI — answering a chat question, classifying a transaction's deductibility, extracting a receipt, suggesting a CSV column mapping, etc.
A token is the unit AI models use to count input + output text — roughly 1 token = ¾ of an English word. Why we surface it: it's a stable proxy for the size of work being done.
2. Last 30 days sparkline
One small bar per calendar day for the last 30 days. Quiet days show as zero bars (no skipping) so trends read at a glance.
Useful for spotting:
- The day someone batch-classified a backlog of transactions
- Whether usage is growing month-over-month
- Anomalies (a spike that's worth asking about)
3. By feature
A table grouping AI calls by their feature — what part of the product made the call:
Feature (prompt_ref) | What triggers it |
|---|---|
chat-firm | Conversations in the firm-level AI tab |
chat-client | Conversations in a client's AI tab |
receipt-extraction | OCR pass on uploaded receipts |
tx-deductibility-classifier | The tax verdict pass from engagements |
classify-rows | CSV import column-mapping suggestion |
| (others) | Smaller features as they ship |
For each, you see calls + tokens this month. Helps answer "are we spending most of our AI budget on chat or on classification?" — the answer affects which feature we should make more efficient first.
4. By user
Same shape as the By feature table but grouped by team member. Identifies who's the heaviest user, which might be perfectly fine (they're doing all the month-end closes) or might mean a teammate is using AI as a search engine and would benefit from quick training.
What the page doesn't show
You won't see a dollar amount on this page. By design.
Why no cost?
Kiravy's billing is fixed by tier — your firm pays a known monthly price (per your contract / firm settings), and we absorb the AI provider cost. Exposing Kiravy's wholesale AI spend on this page would let you reverse-engineer our margin — that's neither here nor there for your operations, and conflates two different concepts (what we pay our supplier vs what you pay us).
If you want to know your monthly bill, that's in Firm settings → Billing (planned).
The internal cost-per-call is not the same as what your firm is charged. Don't divide the Tokens this month number by 1,000 and multiply by some online "GPT pricing" — that's not how your bill works.
Time window
- This month: calendar month, UTC, 1st to end-of-month
- Last 30 days: rolling 30-day window with daily granularity
No date-range picker yet — these two windows cover what 99% of firms actually want to look at.
Empty state
If your firm hasn't made any AI calls this month, the top section shows "No AI calls this month yet." and the sparkline shows 30 empty days. That's normal in the first day or two of a new month.
Permissions
| Role | Access |
|---|---|
| Partner | Full view |
| Manager | Full view |
| Staff / Viewer | Redirected to dashboard |
What's coming
- Per-client breakdown — currently usage is aggregated firm-wide; some firms will want to know which client engagement burned the most
- Quota threshold alerts — banner at the top when you approach your tier's limit
- Date-range picker — for arbitrary windows
- Billing tier summary card showing your current plan + included AI usage
What's next
- AI assistant — the chat surface that generates most of the usage
- Tax deductibility — the engagement classifier that contributes batch calls
- Receipts (OCR) — the upload flow that contributes per-receipt calls