Odoo 19 · the honest map

AI agents for Odoo: what’s real in 2026.

Short answer: three kinds of AI agent work in your Odoo today, and all three are real. Odoo 19 ships its own native agents. Any general assistant, ChatGPT or Claude, connects over MCP in minutes. And specialized agents are built for one job, already fluent in the screen they serve. What separates them is not intelligence. It is what the agent knows about your Odoo’s meaning before it answers, and what happens on the write path after it acts.

Written by nanti.ai. We build specialized agents and the layers under them, so read us knowing that. Odoo quotes are verbatim from Odoo’s documentation; source findings are our own reading of the Odoo 19 Enterprise source. Corrections welcome at hello@nanti.ai.

First, be fair

Three kinds of agent, all real.

The market talk makes them sound interchangeable. They are not, and none of them is fake. Here is what each kind actually is, before any argument about which to use.

Native · Odoo 19

Odoo’s own agents

Odoo 19 ships real AI agents, not a chatbot with an ERP logo on it. We read the Odoo 19 Enterprise source: a real tool-calling agent loop, up to twenty rounds of up to twenty parallel calls, with retrieval over your documents and scheduled, unattended runs. The default Ask AI agent cannot change the database, which is a good default, and for reading and drafting the whole thing is genuinely useful.

Generous engineering. The open question is the write path, below.

ChatGPT · Claude · MCP

General assistants, connected

Any capable assistant can be pointed at Odoo through an MCP server, and the connection takes minutes. That is real too: for pulling records, drafting text, and answering questions, a connected ChatGPT or Claude earns its keep on day one. But a general model knows Odoo’s schema, not your Odoo’s meaning, and a connector moves data without proving what happened.

The connection is the easy half. Meaning and proof are the hard half.

One job · in the screen

Specialized agents

An agent built for one job, living in the screen where that job happens: coaching the rep on the quotation, planning the plant, watching the writes. It arrives already fluent in the models it reads, so it does not guess what a field means, and it acts inside Odoo instead of in another tab. This is the kind nanti.ai builds.

Fluency is the point: the agent knows the screen before it speaks.

“The standard Ask AI agent cannot make changes to the database. As such, it can open views and display reports, but it cannot create leads or alter data.” Odoo documentation

The write path

What none of them should skip.

Reading and drafting are the safe half, and every kind above does them usefully. The moment an agent can write to your database, three properties stop being optional, whoever built the agent. Judge ours by them too.

Meaning before the answer

An agent that reads your Odoo must know which records count and which fields mislead. In Odoo, account.move is invoices, bills, refunds, and journal entries at once, and amount_total is not revenue. A schema tells an agent which fields exist; it takes a verified meaning layer to tell it what they mean.

Why AI gets Odoo revenue wrong →

Rules where prompts cannot reach

The rules an agent must obey belong on the server, not in a prompt. A prompt is advice; a precondition that runs before the write, under the user’s own identity, is a rule. If an action is destructive, a named human should approve exactly that action before it runs.

Can AI write to Odoo safely? →

Proof after the write

A success message is the tool’s own claim. The only way to know what an agent actually wrote is to re-read the committed row from the database and compare it, field by field, with what the agent was asked to do. An agent that cannot show you that read-back is asking to be trusted, not verified.

How to verify what AI wrote →

The nanti.ai line-up

The agents we build, by job.

One agent per job, each native to Odoo’s data model, plus the foundation that makes any AI correct and safe on Odoo. Every entry is honestly labelled by status; nothing here is a mockup.

Agents for the people who run Odoo

One agent per job, each already fluent in your Odoo.

Quotations & sales orders Available now

Nanti Sales Coach

Coach your team to a better order, right in Odoo Sales.

Try it today
Manufacturing & planning In development

Nanti Manufacturing Planner

See what runs short, when, and why, before it bites.

See it
Every Odoo app In development

AI Agents for Odoo

Agents that pull reports, draft records, and chase exceptions.

See it

Make AI correct and safe on Odoo

The layer on top of any AI connection to Odoo, including Odoo’s own, so answers are correct and actions are provable.

AI answers · meaning Early access

Context Layer for Odoo

So AI reads your Odoo correctly.

Try it today
AI writes · rules & proof Available now

Execution Layer for Odoo MCP

So AI acts in your Odoo safely, with proof.

Try it today
Your own servers In development

Secure On-Prem AI

Run the whole stack on your own hardware.

See it
Guides

Plain answers

Asked, answered.

What is an AI agent for Odoo?

An AI agent for Odoo is software that reads your Odoo records, reasons about them with a language model, and can act on them through tools: opening views, drafting records, answering questions, or writing changes, under a user’s permissions. Three kinds exist today: Odoo 19’s own native agents, general assistants such as ChatGPT or Claude connected over MCP, and specialized agents built for one job in one screen. All three are real; they differ in what they know about your data’s meaning and in what happens when they write.

Does Odoo have its own AI agents?

Yes, and they are real. Odoo 19 ships native AI agents with a genuine tool-calling loop: our reading of the Odoo 19 Enterprise source found up to twenty rounds of up to twenty parallel tool calls per turn, retrieval over uploaded documents, and scheduled unattended runs. By default the Ask AI agent cannot make changes to the database, which Odoo’s documentation states plainly and which is a good default. The serious questions begin when an agent is given tools that write; we keep a plain decision checklist at nanti.ai/should-you-turn-on-odoo-19-ai.

Can I connect ChatGPT or Claude to Odoo?

Yes. Both connect to Odoo through an MCP server, and the connection is genuinely quick: Odoo Online has a native path, open-source gateways cover self-hosted installs, and app-store modules exist. We maintain honest step-by-step guides at nanti.ai/connect-claude-to-odoo and nanti.ai/connect-chatgpt-to-odoo, and a comparison of the MCP server options at nanti.ai/odoo-mcp-servers. Connecting is the easy half; a connected general model still needs to be told what your fields mean and still reports its own success unverified.

Is there a copilot for Odoo?

Nothing official carries that name, but the role people mean by it is filled three ways today: Odoo 19’s built-in Ask AI sidebar answers questions inside Odoo; a connected assistant such as ChatGPT or Claude works alongside Odoo over MCP; and specialized in-screen agents coach a specific job while it happens, the way Nanti Sales Coach coaches the quotation screen. Which one fits depends on whether you want general answers or one job done well.

What should an AI agent be allowed to write in Odoo?

Start with nothing. Read-only agents are genuinely useful and carry little risk, which is why Odoo’s own default is a good one. When you do allow writes, insist on three properties: the rules run on the server where a prompt cannot argue with them, destructive actions wait for a named human’s approval of exactly the previewed action, and every write is verified by re-reading the committed record rather than trusting the tool’s success message. Any agent, ours included, should be judged against those three.

Which AI agents does nanti.ai build?

One agent per job, each native to Odoo’s data model. Nanti Sales Coach, the AI Sales Coach for Odoo, is available now on Odoo 19 and coaches quotations and sales orders as they are typed. Nanti Manufacturing Planner is in development with its deterministic planning core already built and running. General-purpose AI Agents for Odoo and Secure On-Prem AI are in development. Under them sit two foundation layers: the Context Layer for Odoo, in early access, which gives any AI the verified meaning of Odoo data, and the Execution Layer for Odoo, available now, which makes every AI write verified, governed, and audited.

Do nanti.ai’s agents work on Odoo Community?

Yes. Nanti Sales Coach is available now on Odoo 19, Community and Enterprise, on-premise or hosted. The Execution Layer for Odoo installs on Odoo 19 Community and Enterprise, self-hosted and Odoo.sh. The Context Layer’s entries are verified against Odoo 19 anchors covering both editions, with edition and module applicability enforced per entry.

How is a specialized agent different from ChatGPT with an Odoo connection?

A connected general model is a brilliant generalist meeting your database for the first time. It does not know that the salesperson on an invoice is invoice_user_id, that amount_total is a document total rather than revenue, or which of your records actually count. A specialized agent arrives knowing the screen it serves, reads verified meaning instead of guessing, works inside Odoo rather than in another tab, and its wins are measured in the record: Nanti Sales Coach, for example, tracks every accepted suggestion to revenue on confirmed orders.

The mission

An agent for every part of Odoo.

Sales is live today. Manufacturing’s deterministic core is built and running. The rest open product by product, honestly labelled the whole way. Leave your email and hear when the next one does.

Try Sales Coach See all products

Evaluating an AI tool that writes to your Odoo?

Name it. We read its public docs and your description and tell you, in writing within five working days, whether it checks the proposal before it writes, logs the result, or re-reads the committed row after. Sourced line by line, using the same pre-registered method as our public write-verification census. Free, no pitch, and we tell you the truth even if the tool is not ours.

Get the Write-Check

Our own write-verification, the Execution Layer for Odoo, is available now. The verdict stays honest either way.

Not ready to talk? Follow what we ship.

An email when something real ships. No newsletters, no spam.

Living this problem every day and want to shape what comes next? We work closely with a small number of Odoo teams on the Execution Layer for Odoo, the ones who cannot fully trust what their AI writes to production today. Tell us about your Odoo setup.