---
title: "Odoo AI 19.4: Chatbots and Task Automation"
description: "Learn how Odoo AI agents, skills, tools, sources, live chat and server actions work, including a safe pattern for company-specific chatbots."
canonical_url: "https://websitefuze.com/blog/odoo-ai-19-4-chatbots-task-automation-company-specific-agents"
last_updated: "2026-09-17T01:00:27.322Z"
keywords: ["Odoo AI", "Odoo 19.4 AI", "Odoo AI chatbot", "Odoo AI agents", "Odoo task automation", "Odoo multi-company chatbot"]
---

# Odoo AI in 19.4: Chatbots, Task Automation and Company-Specific Agents

Published: 2026-09-17

Category: Odoo Updates

Learn how Odoo AI agents, skills, tools, sources, live chat and server actions work, including a safe pattern for company-specific chatbots.

Odoo 19 introduced an AI layer that reaches beyond text generation. It can help users find information, interpret records, open views, classify documents, support helpdesk work, answer website visitors and select controlled actions. Odoo 19.4 extends that direction with automatic model selection, voice interaction, stored conversations, richer live-chat record cards and MCP connectivity.

The useful question is not whether Odoo has AI. It is which agent is allowed to use which information and tools in a particular circumstance. Odoo separates those concerns through agents, skills, sources and server actions.

## Start with the standard Ask AI assistant

The standard Ask AI assistant is available from the command palette and the AI button throughout the Odoo interface. According to the official[Odoo AI documentation](https://www.odoo.com/documentation/19.0/applications/productivity/ai.html), it can answer natural-language questions, open views, improve content, summarise chatter and suggest next steps based on context.

The standard Ask AI agent cannot change database records. It can help a user understand and navigate information, but Odoo deliberately separates that assistance from tools that perform writes. This is an important default because a fluent answer should not silently become an accounting, sales or operational transaction.

## Custom agents define a particular job

A custom Odoo AI agent has a system prompt, response style, one or more skills, and optional sources. The official[AI agents documentation](https://www.odoo.com/documentation/19.0/applications/productivity/ai/agents.html)describes an agent as a natural-language assistant with a defined purpose and access to Odoo tools.

In the earlier Odoo 19 documentation, collections of instructions and tools are called topics. Odoo 19.4 renames them to skills. The purpose remains the same: each skill explains the context, rules and workflow for a task, while tools are the actual functions the agent may call.

This structure supports agents designed for different circumstances instead of one assistant with broad and ambiguous authority. Examples include:

- A sales assistant that finds opportunities, explains pipeline data and drafts follow-up messages.
- A support agent that answers from approved Knowledge articles and escalates uncertain cases.
- A document agent that classifies an upload and selects an approved folder or tag.
- A field-service assistant that summarises task history and prepares a customer update.
- A live-chat agent that answers common questions and creates a CRM lead after collecting the required details.

## Sources determine what an agent can refer to

Odoo agents can use PDFs, web links, Documents files and Knowledge articles as indexed sources. The Restrict to Sources option limits the agent to active uploaded sources. This is useful for customer-facing answers because the organisation can define the material the agent is expected to use instead of allowing broad, unsupported responses.

Source restriction improves control, but source quality still matters. Documents should have owners, review dates and a process for removing obsolete pricing, policies or instructions. Conflicting sources can create inconsistent answers even when every document is technically approved.

## How Odoo AI Live Chat works

Odoo can assign an AI agent to a Live Chat channel rule. The official[AI Live Chat documentation](https://www.odoo.com/documentation/19.0/applications/productivity/ai/live-chat.html)states that an agent can answer common questions, request more information, qualify conversations and escalate to a human operator.

The supplied live-chat pattern does not simply create a lead after the first message. Its instructions define when escalation is appropriate, collect missing contact details, confirm known details and call the lead-creation tool once. This shows why workflow instructions are as important as the language model.

Odoo 19.4 can also display rich, clickable cards when an agent mentions supported Odoo records in live chat. This can make a product, appointment or other supported record easier for the visitor to understand and open.

## A company-specific chatbot is a design pattern, not one checkbox

A multi-company organisation may want each company website to have a chatbot that answers only about that company. Odoo has the building blocks for this design, but the boundary must be configured explicitly.

A practical pattern is:

- Use a separate website and Live Chat channel or channel rule for each company-facing experience.
- Create a separate AI agent for each company rather than one agent with a prompt that tries to infer the company.
- Give each agent only that company’s approved Knowledge articles, documents, policies and web links.
- Enable Restrict to Sources so answers stay within the company-specific source set.
- Assign only the skills and tools required for that channel, such as information retrieval or controlled lead creation.
- Ensure any created lead, ticket or appointment receives the intended company context and responsible team.
- Test cross-company questions, ambiguous brand names and attempts to retrieve another company’s information.

Odoo multi-company records can be shared or company-specific. Its[multi-company documentation](https://www.odoo.com/documentation/19.0/applications/general/companies/multi_company.html)explains that company-linked quotations, invoices and vendor bills are restricted to the relevant company context, while some records such as contacts and products may be shared by default.

That record model does not mean every AI source is automatically isolated. Do not treat a system prompt such as “only discuss Company A” as a confidentiality control. Separate source collections, channel assignments, access rights and tool-side checks provide the stronger boundary. If strict legal isolation is required, separate databases may be more appropriate than shared multi-company configuration.

## Agents can perform tasks only through assigned tools

If an agent has no skills, Odoo says it can provide information but cannot complete tasks or change the database. A tool extends the agent beyond conversation. Preconfigured examples include opening a view and creating a lead, while installed apps can make other tools available.

For automated record workflows, Odoo uses AI server actions. The[AI server actions documentation](https://www.odoo.com/documentation/19.0/applications/productivity/ai/server-actions.html)separates the AI manager from the worker tool. The AI reads the record and prompt, chooses a tool and supplies declared arguments. The standard server action then executes the real business logic.

This allows AI to make a contextual decision while deterministic code enforces the rules. For example, AI may classify a supplier document, but the selected tool should validate the destination, company, allowed record state and required fields before writing anything.

## Useful task-automation patterns

- Classify incoming documents and apply an approved tag or folder.
- Summarise a helpdesk ticket and populate a structured triage field.
- Select a follow-up path based on a record while leaving sending or approval to a user.
- Create a CRM lead from live chat after required contact details are confirmed.
- Draft an email from current record context for human review.
- Extract structured details from a document into a controlled workflow.

The weakest pattern is giving one broad agent a collection of powerful write tools. Narrow agents are easier to test, explain and audit.

## What Odoo 19.4 adds

The official[Odoo 19.4 release notes](https://www.odoo.com/odoo-19-4-release-notes)add several capabilities around this agent model:

- Automatic selection of the most appropriate configured AI model for a task.
- Voice interaction for dictating requests to an agent.
- Storage of agent conversations for up to 30 days after the last message.
- MCP connectivity for compatible external tools.
- Rich preview cards for supported records mentioned in live chat.
- The renaming of topics to skills.

MCP expands integration possibilities, but it also expands the security review. Credentials, exposed tools, user permissions, logging and data boundaries should be documented before connecting an external AI client.

## Provider and pricing considerations

Odoo supports OpenAI and Google Gemini providers. Odoo Online users do not need to add their own API key for the standard service, although they can do so for more control. Odoo.sh and on-premise databases need provider keys for AI features. The[Odoo AI API key documentation](https://www.odoo.com/documentation/19.0/applications/productivity/ai/apikeys.html)notes that provider charges may apply when customer-owned keys are used.

## A safe rollout checklist

- Choose one narrow use case with a measurable manual process.
- Define approved sources, prohibited topics and the human escalation path.
- Assign the minimum tools needed and enforce rules inside each tool.
- Test company, user and record access with ordinary accounts, not only administrators.
- Test incorrect, incomplete, adversarial and cross-company questions.
- Require review for financial, contractual, destructive or externally sent actions.
- Monitor answers, tool calls, provider usage and source freshness after launch.

## The practical conclusion

Odoo AI is most useful as a collection of focused assistants inside existing business workflows. Live chat, company-specific knowledge, lead creation, document classification and record-based automation are credible use cases because Odoo can combine natural-language decisions with structured records and controlled tools.

The design should never rely on confidence or a prompt alone. Separate agents by purpose, restrict their sources, minimize their tools and keep business rules in deterministic Odoo actions. That is what turns a chatbot demonstration into a supportable operational capability.