AI prepares. FID checks.
We use AI for the routine work in bookkeeping and reviews, with tools we choose and configure ourselves, connected to your own Odoo. A person at FID always oversees the work.
What we work with
Langdock
Our AI workspace, with all the language models Langdock offers. All hosted in the EU, GDPR-compliant and certified to ISO 27001 and SOC 2 Type II.
Claude, by Anthropic
Via a Claude Team business account. We use Claude Code to build our own applications and integrations. Under Anthropic’s commercial terms, data is not used to train AI models.
Our own MCP server
The secure connection between AI and Odoo. It runs in our own Railway environment in the EU (Amsterdam), in partnership with Hemisphere AI. This lets AI work directly in your Odoo, without exports or separate files.
From vendor bill to journal entry
The AI bookkeeper reads the bill, fills in the vendor, amount, VAT and general ledger account, and prepares a proposal in Odoo. Nothing is posted until FID has checked it.
Quick answers. Always checked by a person.
Ask your question in Odoo or by email. Our AI colleague knows your file, looks up the answer and prepares a draft. A colleague at FID checks it and sends it. So you get an answer sooner than from a typical firm, and nothing goes out unchecked.
Beta We are now running this with our first clients, with promising results.
- QuestionVia the chatter on an invoice, a message in Odoo, or email.
- AI colleague looks it upIn your Odoo: invoices, entries, payments and attachments, plus our knowledge base with your agreements.
- Draft with sourcesThe draft answer appears in our review queue, with sources attached.
- FID approvesA colleague checks it and amends it where needed, or handles the question personally.
- AnswerIn the same place as the question, under FID’s name.
“Has this invoice been paid yet? Our customer says they paid last week.”
Yes, the payment came in on 14 October and has been reconciled with this invoice. Nothing is outstanding.
INV/2026/0412Bank line 14-10-2026Reconciliation line
Every step is recorded
We want to be able to show how every answer was reached, in line with the EU AI Act. So for each question, the system records:
- what the AI read, with the sources;
- the draft the AI proposed;
- who checked, adjusted and approved it, and when;
- what was finally sent to you.
The AI only reads what it needs for that question. Nothing is sent until a colleague has approved it.
Looking beyond the invoice
Odoo can read an invoice. Our AI looks further: at previous invoices, entries and agreements in the client file, so anything unusual stands out sooner.
A client asked about the VAT on an HGV toll invoice. The AI answered the question, but also noticed something else: some of the lines had already appeared on another invoice a month earlier. None of us had spotted that.
- Trip 1012 · A2€48.20
- Trip 1019 · A12€63.75
- Trip 1023 · A15€39.10
- Trip 1019 · A12€63.75
- Trip 1023 · A15€39.10
- Trip 1107 · A1€52.40
Sample data
A person stays in control
AI makes the work faster, not less careful. That is why we have firm ground rules.
- Human in the loop: AI prepares and makes proposals, a person at FID approves. Only then is anything posted.
- We work through a direct connection to your own Odoo, not with exports or separate files.
- Under Anthropic’s commercial terms, data is not used to train AI models.
- A small number of fixed automations, set up and managed through GitHub, run independently on our own DigitalOcean server in the EU. What runs there is defined in advance.
How the AI learns
The AI learns in two ways. First, from our knowledge base in Odoo, which records each client’s agreements, configuration and way of working.
Second, from the corrections our team makes. Whatever we adjust, the AI takes into account next time.
Towards our own models
In 2027, we plan to move to hosting our own model, or several. The language model itself will then also run in an environment we manage.
A move like that needs careful preparation. We are working on it now, step by step and without disrupting the day-to-day work.
Want to get started with AI yourself?
With existing clients, we currently share our knowledge case by case: what works, what does not, and how to set up proper oversight. We also help set up the AI features in Odoo itself.
We are considering knowledge sessions for a wider group. Interested? Let us know.
Frequently asked questions
How quickly do we get an answer to a question?
Sooner than from a typical firm. Our AI colleague knows your file and prepares a draft answer with sources. A colleague at FID checks it and sends it. This way of working is now running in beta with our first clients.
Could we ever receive an answer that no one has checked?
No. Every answer to a client question is first checked and approved by a colleague at FID. For each question, we record what the AI read, what it proposed and who approved it.
Which AI do you use?
Langdock as our AI workspace, with all the language models available there, hosted in the EU. Claude by Anthropic via a Claude Team business account, and Claude Code to build our own applications. Our own MCP server, in partnership with Hemisphere AI, connects AI to Odoo.
Does the AI post entries on its own?
The principle is human in the loop: AI prepares, and a person at FID approves before anything is posted. Only a few automations, defined in advance, run independently, from our own server in the EU.
Where is our data stored?
In your own Odoo. The models in Langdock are hosted in the EU, and our automation server is at DigitalOcean in the EU. AI requests to Anthropic may be processed outside the EU, with the safeguards the GDPR requires. You can read more about this in our privacy statement (Dutch).
What is an MCP server?
MCP (Model Context Protocol) is an open standard that lets AI work securely with other software. Our MCP server is the connection between AI and Odoo, and determines what AI may read and do there.
How does the AI learn?
From our knowledge base in Odoo, with each client’s agreements and configuration, and from the corrections our team makes.
Do you use Odoo’s own AI?
Not at the moment. Odoo’s own AI features have improved considerably in Odoo 20, but for a service provider like us, our own set-up works better and is more closely integrated. Would you like to use the AI in Odoo itself? We have been setting up AI automations for some time, and are happy to help you configure it properly.
Are you going to host your own AI models?
That is the plan for 2027: our own model, or several, in an environment we manage. We are now carefully preparing for that move.
Can we use this in our own organisation too?
We are happy to think it through with you. For now, we share our knowledge with clients case by case. We are considering knowledge sessions for a wider group.
Curious what AI could do for you?
In a half-hour introductory call, we look at where your bookkeeping involves the most routine work, and what AI could prepare for you.
Book a call