1С/BAS ↔ OpenAI integration

An AI agent on OpenAI answers questions from 1С/BAS data, turns supplier PDFs and warehouse voice messages into draft documents, and the accountant confirms every one

1С/BAS holds the truth about debts, stock and sales, but getting data in and out of it is manual work. Supplier invoices arrive as PDFs and photos, the storekeeper reports a delivery by phone, the owner wants to know who is overdue and has to wait for the accountant. The 1С/BAS ↔ OpenAI pair adds an AI agent to that routine: it pulls the fields out of documents in a fixed form, turns a spoken message from the warehouse into text, prepares draft documents in 1С/BAS and answers questions about the data in plain language. It never posts anything itself: every draft waits for a person. Below: what exactly moves, which scenarios we cover and which capabilities of both systems we rely on.

Request an integration System page: 1С/BAS →System page: OpenAI →

What moves between the systems

1С/BAS → OpenAIA question in plain language («which clients are overdue by more than a month», «how much of item X is left») is answered from current 1С/BAS data: the model picks the report it needs, our code reads it and the answer names the source.
OpenAI → 1С/BASA supplier invoice in PDF or a photo of a delivery note is read into a fixed set of fields agreed with the accountant, such as registration code, items, quantities and amounts, and becomes a draft receipt document in 1С/BAS.
OpenAI → 1С/BASA voice message from the warehouse («twelve boxes of X arrived from Alfa») is turned into text and then into a draft document in 1С/BAS, with the original recording attached for checking.
1С/BAS → OpenAIA client's email is matched to the counterparty in 1С/BAS, and the manager gets a reply draft with the current order status and balance, which they read, edit and send.
1С/BAS → OpenAIEvery evening the agent writes a short summary from 1С/BAS for the owner: receipts of the day, new overdue debts and items below the minimum stock.
1С/BAS ↔ OpenAIEvery answer and every draft is logged with the source document or recording, the data it was built from and who confirmed it, so any figure can be traced back.

What we connect

We agree with you which questions the agent may answer and which documents it may draft; anything else it declines and passes to a person.
On the 1С/BAS side we use the open exchange interface the platform builds itself, under a separate user who may only read and create drafts.
You open your own OpenAI account and access key; the agent runs under it, the usage is billed to your account, and you can revoke the key at any time.
Together with the accountant we fix the set of fields for each document type; the model always answers in that shape, and our code checks the counterparty and items against the 1С/BAS directories before creating a draft.
We connect the channel people will use: a messenger bot for the warehouse and the owner, a mailbox for supplier documents, or a page in the browser.
For the first week the accountant checks every draft and every answer against 1С/BAS, and only then we widen the list of allowed questions and documents.

Automation scenarios

A pile of supplier PDFs

Before every supplier invoice is typed into 1С/BAS line by line, and a wrong quantity surfaces at stock-taking.

After the agent reads the PDF into the agreed fields and prepares the receipt draft, the accountant compares it with the original and posts it.

The storekeeper calls, the accountant writes

Before the storekeeper phones in every delivery, the accountant writes it down on paper and enters it later, and half the details are lost.

After the storekeeper leaves a voice message, the agent turns it into a draft document with the recording attached, and the accountant confirms it.

«Who is overdue?» waits until evening

Before the owner asks the accountant for the overdue list and gets it at the end of the day.

After the owner asks the agent and gets the current list from 1С/BAS with the report it came from.

Client emails without context

Before a client asks where the order is, and the manager opens 1С/BAS, looks up the order and the balance, then writes the answer.

After the manager opens a reply draft that already has the status and balance from 1С/BAS, checks it and sends it.

How it works technically

For those who want to check: each point links to the official documentation page it comes from.

The official BAS support portal documents its integration technologies under the title «Технології інтеграції 1С:Підприємства 8.3» — BAS is built on the 1C:Enterprise 8.3 platform.
Source: its.bas-soft.eu ↗ checked 2026-09-25
According to the same documentation, the platform automatically builds a REST interface using the OData protocol, through which outside systems can read, change, create and delete data over HTTP.
Source: its.bas-soft.eu ↗ checked 2026-09-25
Data sent through the OpenAI platform is not used to train or improve OpenAI models by default, unless the account owner explicitly opts in.
Source: developers.openai.com ↗ checked 2026-09-27
Structured Outputs makes the model always answer in a predefined structure, without missing a required field or inventing a value outside the allowed list.
Source: developers.openai.com ↗ checked 2026-09-27
Function calling lets the model pass data to the client's application, where the code reads data or performs actions the model suggested.
Source: developers.openai.com ↗ checked 2026-09-27
File transcription: upload a finished recording and get the final text, or receive the text as a stream while the model is processing the file.
Source: developers.openai.com ↗ checked 2026-09-27

Limits

The limits come from both systems. The model can be wrong: a fixed set of fields fixes the shape of the answer, not the correctness of every value, so the agent never posts documents itself, every draft is confirmed by a person and every answer names the 1С/BAS report it came from. Request limits depend on your OpenAI account's usage tier and grow with usage, and the usage is billed to your own account. By default OpenAI keeps abuse-monitoring logs of requests for up to 30 days; the zero-retention mode is available only with OpenAI's prior approval. A recording for transcription is limited to 25 MB, so long voice messages are split. Ukraine is on OpenAI's list of supported countries with exceptions the vendor does not spell out. 1С/BAS is on the official prohibited software list: we do not take projects for state bodies, critical infrastructure or systems with state data, and with private businesses we discuss the risk openly.

How to start

Three steps from a first message to work on your project.

01
Leave a request
Describe the task in the form — in text or by voice. Budget and timing are optional.
02
We review and estimate
A developer studies the brief, asks about anything unclear and sends an estimate.
03
We start
Once scope and price are agreed, the work begins.

Leave a request

or write to info@coreviaflow.space

Live project We built and run Gemalli — a B2B export platform with producer pages, a CRM and a deal chat. Read the case →

Questions

Will our accounting data be used to train OpenAI models?

By default, no: data sent through the OpenAI platform is not used for training unless the account owner opts in. OpenAI does keep abuse-monitoring logs for up to 30 days, and the agent only sees the reports and documents you allowed, not the whole database.

Can the agent post a document or a payment on its own?

No. Its 1С/BAS user may only read and create drafts. Posting stays with the accountant, and the journal shows who confirmed what.

OpenAI or Claude: which one should we choose for 1С/BAS?

Both can do this work, and the scheme with drafts and a person confirming is the same. We usually decide on your own documents: we run a sample of real invoices and questions through both and show the accountant the results. If you already have an account with one of them, that often settles it.

Does it understand voice messages in Ukrainian and Russian?

Yes, OpenAI models work in many languages. Quality depends on the recording: noise in the warehouse and unusual item names cause mistakes, which is why the recording stays attached to the draft and the accountant can listen to it.