Last updated July 13, 2026.
When you use AgentBooks, we process only what you submit to run a request: an account identifier (to reply to you) and the content you provide to be priced — a bookkeeping quote (text or an uploaded file) and an approximate transaction volume.
That content is sent to our pricing engine solely to parse the quote and compute the comparison you asked for, and the result is returned to you. We do not use it to train models.
We do not sell or share your data with third parties, and we do not use it to train models. We do not retain personally identifying information — customer and provider names and other identifying details are stripped and not stored — nor do we retain the raw uploaded document.
With your authorization, AgentBooks accesses read-only account, balance, and transaction data from services you connect via OAuth (such as your bank). It is used solely to compute the cash, burn, runway, and books you request. AgentBooks does not move money, and does not retain the raw account data — only the de-identified economic outcomes described above. Access is revocable at any time by disconnecting the service.
The effectiveness of agents and human-in-the-loop work — the outcomes that contribute to ROI data and comparables — is retained, because it is essential to how the Service functions: it powers the benchmarks, retrieval (RAG), and projections that make an agent's books meaningful. This data is de-identified by design: it records what worked and by how much (rates, proposed pricing, ROI outcomes, and agent cost, margin, tier, and metered usage), never who — the name of the company having its books done is not stored. We retain this outcome data for two reasons: defensibility (an auditable basis for the ROI figures we report) and quality of service (improving the accuracy of future benchmarks). In short: we store what worked, not who it worked for.
Data is hosted on Google Cloud Platform (Cloud Run and Firestore) and MongoDB Atlas. Language models (Google Gemini and Anthropic) may assist in extracting figures from an uploaded document; no submitted data is used for model training.
Retained data is de-identified and purpose-limited to the benchmarking and projection functions above. Disconnecting a connected service or ceasing to use AgentBooks revokes its access; to request removal of records associated with your account, contact us below.