Cost transparency
LiveA usage view for the current month, broken down by feature and by model, with a 30-day daily series.
- ✓Per-request log — input, output, cache-read and cache-creation tokens, the model that actually answered, estimated cost
- ✓Cost in US dollars on every row, from one price list, never converted at read time
- ✓Anthropic, OpenAI and Google calls on the per-token paths all write a row
- ✓User, matter and artefact recorded on each request
What it is.
Every model call that is priced per token — Anthropic on every path, OpenAI and Google on the multi-agent path — is written to a per-request log with input, output, cache-read and cache-creation tokens, the model that actually answered, and the user, matter and artefact it belongs to. Each row carries its cost computed from one US-dollar price list at the time of the call and never converted afterwards; the usage view shows the figure in your billing currency (US dollars on this product), and if any request in the period has no figure in that currency it says so rather than showing a partial sum. The usage endpoint reads the current calendar month and returns totals broken down by request type and by model, a 30-day daily series, storage, seats, and where you stand against your token allowance.
Two things are outside the token meter, by their nature rather than by omission. Embeddings — every chunk of every uploaded contract — write no cost row. Meeting transcription is priced per minute of audio, not per token, so it is not in this log either; its record is the meeting itself, which carries the recorded duration and the transcription model used. And the log is richer than the view over it: user, matter and artefact are recorded per request but the endpoint does not yet group by them, so there is no per-user chargeback view.
Three steps.
End to end.
Every per-token model call is logged automatically with its tokens, the model that answered, and its cost.
Totals by request type and by model, a 30-day daily series, storage and seats, and progress against your token allowance — in your billing currency.
Embeddings and meeting audio are outside this log, and there is no per-user breakdown yet.
What you get.
- ✓Per-request log — input, output, cache-read and cache-creation tokens, the model that actually answered, estimated cost
- ✓Cost in US dollars on every row, from one price list, never converted at read time
- ✓Anthropic, OpenAI and Google calls on the per-token paths all write a row
- ✓User, matter and artefact recorded on each request
- ✓Current-month totals by request type and by model, plus a 30-day daily series
- ✓Storage, seats and token-allowance position in the same view
- ✓A period with any unpriced request shows no total and says how many, rather than a partial sum
- ✓No date, user or feature filters on the endpoint — it returns the current month
- ✓Daily spend cap and AI audit history in AI Administration
Quick answers.
No. There is no CSV export on the usage endpoint and no outbound webhook carrying usage events.
Not as a separate product. Usage is read in the app, by a signed-in user.
No. It covers every call priced per token. Embeddings are not logged, and meeting audio is priced per minute and recorded as duration and model on the meeting instead — so total model spend is higher than the figure shown, by those two amounts.
More in AI Controls & Guardrails.
Every citation comes back with a verdict — verified, ambiguous or unresolved — and a match on title alone is never verified.
Each legal claim we hold about a jurisdiction is checked against that government's own legislation service — and where it cannot be, the answer is "unverified", not silence.
We do not train models on customer content. There is no opt-in, because there is no training programme.
User-pasted text is sanitised before reaching the LLM.