- Per-transaction classification at journal-post time, not in batch
- Rules + embedding match: deterministic overrides beat ML, not the other way around
- Counterparty normalization maps raw merchant strings to vendor master records
- Tag-driven override: a #research tag reclassifies a Stripe charge as R&D instantly
- Confidence scores and a review queue for ambiguous charges
- Per-agent categorization policies so different fleets behave differently
- Full audit trail: every reclassification is an immutable ledger event
Agent Expense Categorization: Classify Every Spend at the Moment of Ingest
A Stripe charge from "OpenAI" could be COGS, R&D, or Marketing depending on which sub-agent incurred it. Agent expense categorization inspects the transaction context — counterparty, agent ID, task tags, memo — at the moment the journal entry is posted, and stamps the correct chart-of-accounts code on it before it ever hits the ledger. No nightly ML batch, no end-of-month cleanup script.
Categorize agent spend automatically
AgentBooks is a double-entry ledger API built for autonomous systems. Join the waitlist for early access.
You're on the list.