Basic — the assistant on its own
Who this is for: anyone who wants to try MitDesk end to end.
You need: Edition: Basic (module minddesk). See Editions.
Basic is one module, minddesk. It gives every user a chat that reads Odoo with
their own rights, sends every change to a person for approval, and records what
happened. This page walks through it from installation to the audit log.
About the screenshots
All screenshots come from one database with every Business module installed. So the top menu shows entries that Basic does not have, such as Workforce, Governance and Telephony. The screens on this page are all part of Basic.
1. Install MitDesk
Open Apps and search for MitDesk. MitDesk is a single application; every other MitDesk module is an add-on that installs next to it.
2. Connect a model
MitDesk does not ship a model. You connect the provider you want under Configuration ▸ Settings ▸ AI keys: Anthropic, OpenAI, Gemini, Mistral, a local Ollama server, or any self-hosted, OpenAI-compatible server. Every entry has a Test button.
A self-hosted server is an AI endpoint (Configuration ▸ Engine ▸ AI endpoints). You give it a name and an address; MitDesk works out from the address where data sent to it ends up. The office server below is on a private address, so it is classified On-premises. You cannot type that classification yourself.
Configuration ▸ Engine ▸ AI Models lists every model with its provider and where it sends data — Non-EU cloud, EU cloud or Local / on-premises.
3. Ask your first question
Open Chat. A new chat offers starter questions; you can also type your own. The header shows which assistant you talk to and how much it may do alone. It also says runs as you: it sees only what your Odoo account sees.
An answer lists the records it is based on under Sources, and the right-hand panel repeats them. What this agent may do explains its limits in one line each.
4. Decide what the AI wants to change
Nothing that changes data happens without a person unless a rule allows it. Every request lands on Decisions: how many need your approval, how many need an answer, how long the oldest has been waiting.
Four kinds of request always go to an approver, not to you:
- anything that moves money,
- anything that changes MitDesk's own settings or permissions,
- a tool that picks its own target,
- any action that would run as another account. These rows show a lock and "Approvers" instead of an Approve button.
The same requests as a list, with filters, are under Oversight ▸ Approvals.
Oversight ▸ How Approvals Work explains this inside the product. Every action has one of three outcomes: it runs, it waits for a person, or it is refused. The page also shows who controls what.
5. Set the rules
Configuration ▸ Rules ▸ Autonomy Rules says, per tool and model, whether an action runs or asks a human.
Before you change a rule, Test a decision answers what MitDesk would do for a person, an operation and a model. Nothing is run and nothing is recorded.
Each assistant (Configuration ▸ Who does the work ▸ Agents) has a model and a default autonomy — Read only, Proposes or Asks first. The tools they can use are listed by area under AI Tools.
6. See what happened
Oversight ▸ Audit keeps one row per event: who, which tool, whether it touched money, and the data class involved.
Configuration ▸ What has run ▸ Jobs shows every background run and, when one stopped, why.
7. The Cockpit
The Cockpit is the start page: what waits for you, what ran, what failed. The sections for digital employees and automations appear when those modules are installed.