Ask a question about your AI spend.
Ask about spend, ownership, outcomes, anomalies or savings. Valistry calculates the answer from your ledger and shows the evidence behind it.
Why did spend spike on the 21st?
2026-08-21 cost $514 more than a usual day. Draft reply agent $467 and No workflow yet $63 carried it.
- Draft reply agent$467
- No workflow yet$63
- Ticket triage$9
- Code review agent$3
The ledger calculates. Code writes.
Your question becomes a checked query, and code writes the answer from recorded data. A model may only help read the question.
Understand: code first
The question becomes a query
What is being measured, how it is grouped and over what period. Code does this first. If it cannot, a model may propose a plan, using only names in your records. Checked before anything runs.
Calculate: no model involved
The query becomes figures
It runs over your own reconciled records. Every figure, row and piece of evidence in the answer comes from this step.
Explain: code writes it
The figures become a sentence
Code writes the answer from the figures. No model sees the result or writes a word of it.
Every answer shows its math.
Code writes every answer from the computed figures. A model never writes one.
Ask in the product, or through our MCP server.
One set of figures, read-only everywhere.
From any screen.
Press ⌘K on any screen and ask. The answer comes with the figures it was worked out from, and a link to the rows behind it.
From Claude Code and Cursor.
Add our MCP server with one read-only key. An agent can read every figure and change nothing. The key does not expire; replace or delete it.
From Claude, Cowork and the desktop app.
Add our MCP server as a connector and sign in. The connection is yours, read-only, for one organization. It ends after 30 days without use.
Ask, understand, act.
Connect one source and ask the first question.
Every answer comes from the ledger your teams use, and opens the screen where you approve or reject.
Next: the connectors it reads →