Why we built oc8, and why it is open

We have spent years running Odoo projects: rollouts, migrations, integrations, custom development.

Somewhere in the middle of all that, AI showed up, and it did not fit anywhere. Here is why we built oc8, what it does, where it came from, why it is open, and what you can do with it today.

The problem we kept running into

This did not start from a thesis about AI. It started from what we kept seeing in those projects.

Odoo does a lot of the job well. Sales, inventory, accounting and projects land in one place, and for most of the companies we work with it is the system the business genuinely runs on.

It is also never the only one. Email and files sit in Microsoft 365 or Google Workspace. Sales works in HubSpot or Salesforce. Support has its own helpdesk, the shop has its own platform, and somewhere there is an industry tool nobody is going to replace. That is not a flaw in anyone's software. It is simply what a grown company looks like, and it is not going to change.

So the work spreads out across all of it, and a person is the thing that carries it from one system to the next.

AI arrived in the same shape

It sits in a browser tab next to the systems, or inside one of them, doing that one system's work. It does not reach across. Someone still reads the number, judges it, and types the next step by hand.

What was missing sat above them: a layer that leaves those systems where they are, uses what each one is good at, and lets AI work across all of them at once.

What is oc8?

oc8 is an open platform that orchestrates AI agents across every system a company already runs. Instead of one assistant sitting inside one tool, you get agents that work across all of them, while a person keeps permissions, approvals and a full audit trail.

You set agents up roughly the way you would set up a team. They sit in departments, they have a job, and they have permissions. In a single run an agent reads data out of your ERP, acts in your CRM and drafts in your inbox, carrying out the steps a person used to carry between those systems by hand.

Two connections make that possible, and they are deliberately different. Systems connect through MCP, the open standard for giving AI models access to tools and data. AI models connect through their own APIs, which means you choose the model per agent and swap it whenever a better one appears.

The human does not disappear from this. Agents do the work; you decide what they are allowed to touch, and what has to wait for you.

That is what the name is about. Stop operating. Start orchestrating.

Where does oc8 come from?

oc8 is an open-source project by the BOP Alliance, a global network of around 70 leading Odoo Gold Partners and specialised Solution Partners across 50 countries. The network works together on international rollouts, on shared extensions, and more recently on AI. oc8 came out of that last strand.

It started with a question rather than a product plan. When Odoo 19 brought AI into the product, we wanted to know what customers would actually do with it. So in March 2026 we ran the Odoo AI Challenge and asked our partners and the wider Odoo community the same thing: where could AI genuinely help in your customers' daily work?

The response was huge, and the solutions that came back were strong. One submission sketched an AI layer above Odoo that connected every communication channel, tied them to real business data, logged each interaction in the CRM on its own, and kept the AI grounded in a proper knowledge base. Small in scope, but pointing somewhere, and that is roughly where oc8 began.

Whether it would be open was never really a question. Most of us have been open-source people for far longer than we have been anything else.

It started inside an Odoo network, and it was never meant to stay there. Whatever systems a company runs on, oc8 is built to work across them. You can see the whole picture on the product page.

Why is oc8 open?

Because software that acts on your behalf should not be a black box. oc8 has an open-source core you can run on your own infrastructure, the AI model is chosen per agent, the systems connect through an open standard, and every action is logged. Nothing about it requires you to stay.

When software only stores your data, a closed vendor is an inconvenience. When it starts doing the work, reading your contracts, answering your customers, moving your money, it becomes something else.

A closed platform decides for you

Most AI platforms hand you one model, one place to run it, and a list of integrations somebody else controls. When the model gets worse, when the price changes, when the connector you need is not on the list, you wait. And leaving is not really an option, because by then the system holds the way your company works.

Open means you decide

oc8 has an open-source core and runs where you put it: your own cluster, on-prem, or hosted in Europe. Your data stays there. You choose the AI model per agent, from OpenAI, Anthropic, Google or Mistral to an open model on your own hardware, and you swap it whenever you like. You connect the systems you already run, because the standard underneath is open too. Your models, your data, your rules, and the freedom to leave with all of it.

Anyone can build on it

Around oc8 there is a community and a partner network instead of a single vendor, and both are open to anyone. What they build are Capas: a connector to a system you run, a single skill, a complete agent, a whole department template. Everything published lands in the ecosystem, where any company installs it in a few clicks. Build your own instead if you would rather, or have a partner do it with you.

A closed vendor has one roadmap and a fixed number of engineers. Every connector you need has to earn a place on that plan, and until it does, you wait.

An open platform has no such ceiling. Every company, every partner and every developer working with oc8 can extend it, and whatever they publish is there for everyone else. One team solves an integration on a Tuesday and the rest can install it on Wednesday. The more people build, the faster the platform moves.

So the list of work a company can hand to AI keeps growing, with no limit in sight.

You stay in control, and you can prove it

Access rights run down to the individual record. Guardrails limit what an agent can do at all. The steps that carry risk wait for a person to approve them. Every action, including every model call, is written to an audit log you can read.

That is not a compliance feature bolted on the side. It is what makes an agent safe to hand real work to. It also happens to be what the EU AI Act asks for, since human oversight and record-keeping are becoming legal duties for a growing set of systems.

How to start today

Four ways in, depending on how far you want to go.

Run it yourself. Clone the repository and follow the quickstart.

Build it with us. The code, the roadmap and the issues are public. Write an agent, a skill or a connector, or come and argue about the architecture, on GitHub and Discord.

Build a business on it. Implement oc8 for your customers, build services around it, or publish your own Capas. Join the partner program.

Let us run it for you. A hosted edition and an enterprise edition are coming, and the open core stays open when they arrive. Join the waitlist.