How to connect ChatGPT to Odoo

Short answer

There are three ways to use ChatGPT with Odoo: paste or export data into it, give a custom GPT an action that calls a small proxy in front of the Odoo API, or use Odoo's own optional ChatGPT-powered features. For live manufacturing data, put a read-only proxy in the middle and decide whether the data may leave your network.

Get this built on your Odoo

Most Odoo users who ask about ChatGPT want one of two things: a faster way to write and summarise (emails, product descriptions, chatter), or a way to ask questions about their own data. The first is easy and low risk. The second needs real integration work.

ChatGPT cannot talk to your Odoo database on its own. It needs an endpoint it can call. That endpoint is where you control what is exposed, how users authenticate and what gets logged.

If your Odoo holds drawings, customer pricing or export-controlled part data, the first question is not how to connect, but whether that data may go to a third-party cloud at all.

What teams use it for

Drafting customer and vendor emails

Ask ChatGPT to turn a sale.order or purchase.order summary into a polite email. Low risk if you paste only the fields you choose.

Custom GPT for order status

A custom GPT with an action that calls your proxy to fetch sale order and delivery status for an authenticated salesperson.

Explaining BOM and routing changes

Export two versions of a bill of materials as CSV and ask for a plain-language diff for the engineering change note.

Data cleanup helper

Use ChatGPT to propose normalised product names or categories from an exported product.template list, then import the reviewed result back with Odoo's import tool.

Formula and report help

Ask for help with domain filters, Studio automations or SQL for a read-only report, then test it yourself before using it.

Odoo integration surface

Odoo's list views can export to CSV or XLSX, and the Import tool accepts the same formats. This is the simplest and most controlled way to move data to and from ChatGPT.
External API (XML-RPC or JSON-RPC) with API keys. ChatGPT custom GPT actions need an HTTPS REST endpoint described by an OpenAPI schema, so you normally build a small proxy that translates.
Odoo has optional features that call OpenAI from within the product in recent versions, such as text assistance in chatter or composition. Availability and naming differ by version and edition.
Webhooks and server actions can send record data out when something changes, but sending to a third party should be a deliberate choice.
Per-user access rights and record rules still apply to any call made as that user, which is what keeps a proxy honest.

How it works

1
Decide what may leave your network
Classify fields: public product text is fine, customer pricing and drawings may not be. Write that down before building.
2
Pick the integration pattern
Manual export for ad-hoc work, a custom GPT action for repeated lookups, or a private model if the data should not leave.
3
Build a read-only proxy
A small web service with endpoints such as GET /order-status and GET /stock. It calls Odoo with a dedicated API key and returns only whitelisted fields.
4
Describe it with OpenAPI
Write the schema for the proxy and add it as an action in the custom GPT. Keep descriptions short and precise so the model picks the right call.
5
Authenticate users properly
Use OAuth in front of the proxy so each user acts as themselves, not through one shared key that sees everything.
6
Review enterprise data controls
Check your OpenAI plan's data retention and training terms, and confirm with your security lead before connecting production data.

Example flows

Sales rep asks about a delivery

Trigger: A salesperson asks the custom GPT where order S00412 is.

The GPT calls the proxy with the order reference.
The proxy queries sale.order and the linked stock.picking records.
It returns status, planned date and carrier tracking if present.
The GPT writes a short reply the rep can paste into an email.

Result: The rep answers the customer without opening several Odoo screens.

Export, analyse, import

Trigger: A buyer wants to clean up vendor product codes.

Buyer exports product.supplierinfo to CSV with only the needed columns.
ChatGPT proposes normalised codes and flags duplicates.
Buyer reviews the result in a spreadsheet.
Buyer imports the corrected file using Odoo's import with a test run first.

Result: A tedious cleanup is done in an afternoon with a human check at each step.

Keeping data on-prem

ChatGPT is a hosted service, so a self-hosted Odoo that must keep data inside the network cannot use it for live data without sending that data out. The alternative is to keep the same pattern with an open-weight model served on your own GPUs, exposed to users through a private chat interface that calls the same read-only proxy. You keep the OpenAPI-style tool design and the approval model, but the model and the logs stay in your environment. Many teams use ChatGPT for non-sensitive writing tasks and a private model for anything that touches drawings, pricing or controlled data.

Security checklist

Use a dedicated, read-only Odoo user for the proxy, never an administrator.
Whitelist fields in the proxy; do not forward arbitrary domain filters from the model.
Check retention, training and regional terms of your ChatGPT plan before sending production data.
Keep export files off personal devices and delete them after use.
Log who asked what, and which records were returned.

Ways to get it done

Manual export and paste

Effort: None to set up; discipline needed about what is pasted.

Ad-hoc writing and cleanup with non-sensitive data.

Custom GPT with an action and a proxy

Effort: A few days to a couple of weeks including OAuth and testing.

Repeated read-only lookups for a defined group of users.

Private model with Netray or in-house

Effort: Longer set-up for GPUs and models, but the same tool pattern applies.

Regulated or sensitive data that cannot go to a hosted service.

Pitfalls

!Sharing one API key across all GPT users, which gives everyone administrator-level visibility.
!Letting the model build raw search domains, which can cause heavy queries or leak unintended fields.
!Pasting full customer lists into a consumer account with unclear data terms.
!Importing model-generated data without a test import and human review.
!Assuming the Odoo version in your head matches the model's training; field names change between releases.

Frequently asked questions

Does Odoo have a native ChatGPT integration?

Recent Odoo versions include optional AI-assisted features that can use OpenAI models in places like text composition. What exists and how it is configured varies by version and edition, so check your release notes. It does not replace a purpose-built data integration.

Can ChatGPT read my Odoo database directly?

No. It needs an endpoint to call. You provide that endpoint, usually a small proxy over the Odoo external API, and decide which fields and records it can return.

Is it safe to send Odoo data to ChatGPT?

It depends on the data and your plan's terms. Public product text is usually fine. Customer pricing, drawings and controlled technical data often are not. Ask your security lead and read the retention and training terms of your plan.

Can ChatGPT write to Odoo?

Technically yes through a proxy, but it is better to keep it read-only and have humans import or confirm changes. If you add writes, use draft records and an explicit approval step.

What if we cannot use a cloud model?

Use the same tool design with an open-weight model hosted in your own environment. You lose some model capability on hard tasks but gain control of data and logging.