Skip to content

Our work · Read-only AI

Ask your accounts your questions,in plain language

A clinic that keeps its accounts in Horus, across several companies, and a team that wanted to query its figures without opening the software.

Illustration of an assistant linked to the accounts in read-only mode

The need

“Which supplier invoices are still open?”, “Where does this account stand?”: for every question, someone had to open the accounting software, search and export. The team wanted to ask its AI assistant directly.

Before, after

After

  • Horus accountsinvoices, trial balance, general ledger
  • User accountsone each, read-only
DeepFlow
  • Answer with its sourcedocument number, date, amount
  • Question logwho asked what, and when

What we did

01

A secure connector between the assistant and the accounts

We set up a connector, hosted by DeepIT, between the clinic's AI assistant and its accounting software. The assistant can consult the chart of accounts, the journals, third parties, the trial balance, the general ledger, open invoices and documents.

02

Read-only, twice over

The connector only accepts reads: any write request is refused before it even reaches the accounts. On the accounting software side, the access granted is also limited to reading. All our rules on your data: your data under control.

03

Everyone connects with their own account

No shared password and no key handed to the client: each user connects with their own accounting account, and only sees the files they already have rights to.

04

Tested on real data

Before handover, every tool was checked against the clinic's real data, and write attempts were duly blocked. The clinic's team is now connecting, one user at a time.

What it changes

  • An accounting question is asked in one sentence, instead of ten clicks.
  • The data doesn't leave the agreed framework: read-only, named access.
  • The same method applies to any AI connected to a practice's data.

What about your data?

Connecting an AI to a practice's data calls for a framework: who sees what, what is read, what never is. We set it with you before we start, and check it afterwards. Your data under control

An AI in your practice, with a clear framework

We look with you at what the AI can read, what it will never read, and where to start.