AI features are now standard in modern ERP systems. Odoo is no exception. But how AI is integrated (at the architecture level) determines what you can actually do with it, who controls it, and what it costs over time.
Odoo SA and the OCA (Odoo Community Association) have taken two distinct paths. This article documents what each one looks like, and where they diverge.
1. Odoo SA: AI as a Utility
Odoo SA targets the broadest possible user base. The design goal is simplicity: once an API key is configured, AI features work across the platform with no further setup required.
Odoo 19 is the first version to introduce native AI as a system-wide featureAn individual measurable property or characteristic of the data used as input to a model. Feature engineering — selecting, transforming, and creating features — is a critical step in the ML pipeline., available on Enterprise edition only. AI capabilities are integrated across every major module: agents, natural language search, email drafting, field completionThe text output generated by an LLM in response to a prompt. Also called a response or generation. LLMs generate completions by predicting the most likely next token given the context., voice transcription, document-based Q&A, and more. Two providers are supported: OpenAI (ChatGPT 5.0) and Google Gemini. All deployment types require a direct API key from either provider.
For agents specifically, users have meaningful configuration options: system promptAn instruction provided to an LLM before the user's message, typically used to set the model's behavior, persona, tone, and constraints. System prompts are invisible to end users and are a key tool…, LLMA neural network trained on vast amounts of text data to understand and generate human language. LLMs use the Transformer architecture and can perform a wide range of tasks — summarization,… modelA mathematical function trained on data that maps inputs to outputs. In ML, a model is the artifact produced after training — it encapsulates learned patterns and is used to make predictions or…, response style (Analytical / Balanced / Creative), topics, and knowledge sources. Beyond agents, built-in AI features work as configured by Odoo SA.
This is a deliberate design choice, not a limitation. For the intended audience, it works well.
The trade-off is provider lock-in and transparency. ModelA mathematical function trained on data that maps inputs to outputs. In ML, a model is the artifact produced after training — it encapsulates learned patterns and is used to make predictions or… selection is limited to OpenAI and Google Gemini; there is no path to a self-hosted modelA mathematical function trained on data that maps inputs to outputs. In ML, a model is the artifact produced after training — it encapsulates learned patterns and is used to make predictions or… or a third-party provider. Beyond that, because requests are processed through Odoo’s infrastructure, the exact payload sent to the modelA mathematical function trained on data that maps inputs to outputs. In ML, a model is the artifact produced after training — it encapsulates learned patterns and is used to make predictions or… is not visible. Whether Odoo appends additional instructions to your configured system promptAn instruction provided to an LLM before the user's message, typically used to set the model's behavior, persona, tone, and constraints. System prompts are invisible to end users and are a key tool…, or modifies the response before returning it, is not something you can inspect or verify.
2. The OCA: AI as Infrastructure
The OCA’s oca/ai repository approaches the problem from a different premise: the AI provider should not be the ERP vendor’s decision.
Their published roadmap outlines three integration paths. The Bridge is the current choice; Native and MCPAn open protocol developed by Anthropic that standardizes how AI models connect to external tools, data sources, and services. MCP allows LLMs to call tools (file systems, APIs, databases) in a… are under consideration for the future:
- Bridge (chosen): external systems like n8n handle workflows. Odoo is a trigger and receiver, independent of specific platforms.
- Native (future): workflow logic inside Odoo, external LLMA neural network trained on vast amounts of text data to understand and generate human language. LLMs use the Transformer architecture and can perform a wide range of tasks — summarization,… calls. Similar to Odoo Enterprise, but harder to implement complex workflows.
- MCPAn open protocol developed by Anthropic that standardizes how AI models connect to external tools, data sources, and services. MCP allows LLMs to call tools (file systems, APIs, databases) in a… (future): AI agents connect to Odoo via Model Context ProtocolAn open protocol developed by Anthropic that standardizes how AI models connect to external tools, data sources, and services. MCP allows LLMs to call tools (file systems, APIs, databases) in a…. Follows current standards but raises security concerns.
What distinguishes the bridge and MCPAn open protocol developed by Anthropic that standardizes how AI models connect to external tools, data sources, and services. MCP allows LLMs to call tools (file systems, APIs, databases) in a… paths from Odoo SA’s approach is that the workflow logic itself lives outside the ERP.
The current implementation is the bridge. The core module is ai_oca_bridge, decoupling Odoo’s business logic from the AI provider so that switching models or vendors requires no changes to business code.
What this enables:
Any LLMA neural network trained on vast amounts of text data to understand and generate human language. LLMs use the Transformer architecture and can perform a wide range of tasks — summarization,… provider
Swap between OpenAI, Anthropic, local LLMs, or any compatible provider without changing business code
Full data control
Control exactly what is sent to the modelA mathematical function trained on data that maps inputs to outputs. In ML, a model is the artifact produced after training — it encapsulates learned patterns and is used to make predictions or…: records, chatter, metadata, custom context. Keep sensitive data on-premise with a locally hosted modelA mathematical function trained on data that maps inputs to outputs. In ML, a model is the artifact produced after training — it encapsulates learned patterns and is used to make predictions or…
Full transparency
The bridge maintains a log of every interaction. Audit exactly what was sent to the modelA mathematical function trained on data that maps inputs to outputs. In ML, a model is the artifact produced after training — it encapsulates learned patterns and is used to make predictions or… and what was returned
Any Odoo version
Community-maintained branches from 16.0 to 19.0. As an open source project, it can be extended to other versions as needed
The table below compares both approaches across the criteria that matter most when evaluating AI integration for an Odoo project.
| Criteria | Odoo SA | OCA Bridge |
|---|---|---|
| Architecture | Managed through Odoo's infrastructure | External, platform-independent layer |
| Provider | OpenAI or Gemini only | Any compatible LLMA neural network trained on vast amounts of text data to understand and generate human language. LLMs use the Transformer architecture and can perform a wide range of tasks — summarization,…, including self-hosted |
| Agent configuration | System promptAn instruction provided to an LLM before the user's message, typically used to set the model's behavior, persona, tone, and constraints. System prompts are invisible to end users and are a key tool…, modelA mathematical function trained on data that maps inputs to outputs. In ML, a model is the artifact produced after training — it encapsulates learned patterns and is used to make predictions or…, response style configurable for agents | Full control over promptThe input text provided to an LLM to guide its response. Prompt design — choosing words, structure, and examples — significantly affects output quality. Also referred to as the user message or query., data, and workflow logic |
| Transparency | Request payload not visible; Odoo intermediates the call | Full request/response history, auditable |
| Cost modelA mathematical function trained on data that maps inputs to outputs. In ML, a model is the artifact produced after training — it encapsulates learned patterns and is used to make predictions or… | Own API key (OpenAI or Gemini) | Direct provider billing or self-hosted |
| Target user | Business users (one-time API key setup) | Developers, integrators |
| Odoo version | 19.0+ (Enterprise only) | 16.0 to 19.0, extensible |
3. Trobz and the OCA Bridge
Trobz uses the OCA bridge in production and has contributed extensions back to the OCA repository. Two of these directions have open PRs on the OCA/ai repository (16.0 branch).
Workflow integration with n8n
The bridge acts as a trigger and endpoint for automation flows. When an event occurs in Odoo, the bridge forwards it to n8n, which can gather additional context before passing everything to the LLMA neural network trained on vast amounts of text data to understand and generate human language. LLMs use the Transformer architecture and can perform a wide range of tasks — summarization,…. The LLMA neural network trained on vast amounts of text data to understand and generate human language. LLMs use the Transformer architecture and can perform a wide range of tasks — summarization,… response then drives an action back in Odoo.
One flow we have built with the bridge handles helpdesk ticket routing:
Customer creates a helpdesk.ticket in Odoo. The bridge forwards the ticket data to n8n.
n8n calls the Odoo API to collect additional context: team availability, current workload, cost data.
All data is passed to the LLMA neural network trained on vast amounts of text data to understand and generate human language. LLMs use the Transformer architecture and can perform a wide range of tasks — summarization,…, which assesses the ticket's feasibility and identifies the most suitable team member.
n8n calls the Odoo API to assign the ticket automatically based on the LLMA neural network trained on vast amounts of text data to understand and generate human language. LLMs use the Transformer architecture and can perform a wide range of tasks — summarization,…'s recommendation.
This moves AI from “answering questions” to “making operational decisions.”
Server Action automation
PR #52 adds server_action result type and field-based triggers. AI can evaluate business conditions (inventory levels, order status, supply chain signals) and trigger Odoo Server Actions directly. The result is a system that acts on its own analysis rather than waiting for a human to interpret a report.
Scheduled reporting
PR #62 introduces ai_oca_bridge_scheduler. Datasets from Odoo are sent through the bridge on a schedule, analyzed by the modelA mathematical function trained on data that maps inputs to outputs. In ML, a model is the artifact produced after training — it encapsulates learned patterns and is used to make predictions or…, and returned as structured reports. The AI is not reading a dashboard; it is receiving raw data and producing a summary with its own interpretation.
These contributions reflect both our use of the bridge in real implementations and our capacity to extend it. The underlying architecture is what makes this range of use cases possible.
4. Two Different Bets
Odoo SA is betting that most businesses want AI that just works, without configuration overhead. That is a reasonable bet for a large portion of the market.
The OCA is betting that AI providers will keep changing: new models, new costs, new capabilities. Tying those decisions to the ERP vendor creates a long-term constraint. Their approach treats AI the same way Odoo treats email servers or payment gateways: as external infrastructure that the platform connects to, not owns. That is also a reasonable bet, but it requires a different kind of investment.
Neither is wrong. The choice depends on what you are trying to do with the system.