EN
language & localization

AI · ERP · CRM

AI agents in your ERP that know your processes

You ask a question in plain English, the agent reads your ERP, cross-checks orders, invoices and your approval rules, then answers with the evidence. We build it around your configuration, your extensions and your data, with the model you choose: hosted in your country, installed on your servers, or one of the large market models.

  • Reply within 2 business days
  • Your data stays under your control
  • The agent proposes, your teams decide
  • Since 2010e-commerce and systems integration
  • Odoo, Dynamics AX, CegidERPs already connected in our projects
  • Lyon and HanoiEurope and Asia, two time zones
  • Your choice of modelin your country, self-hosted or market

Models and tools we work with

  • Mistral AI
  • OVHcloud
  • Anthropic
  • Google Gemini
  • OpenAI
  • n8n
  • Odoo

Demo

What an agent connected to your ERP looks like

A sample conversation, replayed in front of you. The agent reads the ERP, reasons with your rules, cites the records it consulted and proposes the next step. It writes nothing without your approval.

Example: an AI agent connected to the ERP, asked in plain English

ERP agentConnected to the ERP · reads sales, invoicing and approval rules
  1. You

    Review this month’s order-to-cash cycle and give me the two most important issues.

  2. Agent
    1. Reading draft orders
    2. Checking balances and overdue invoices
    3. Checking approval rules

    1. 14 draft orders for Orvalis Distribution, worth €186,400. This customer already has €42,300 of invoices more than 60 days overdue. Confirming them raises the exposure without a credit decision.
    2. Order SO-10482 (€12,800) exceeds the €10,000 limit with no approval recorded. The approval rule was not applied.
    I recommend a credit decision today for Orvalis Distribution, then holding SO-10482 until it is approved.

  3. You

    Prepare a 5-line briefing for the team on the first issue, with links to the records.

  4. Agent
    1. Reading the customer record
    2. Matching invoices and payments

    Team briefing: Orvalis Distribution
    1. 14 draft orders, €186,400, 9 of them due for delivery before the 20th.
    2. 3 overdue invoices, €42,300, the oldest for 74 days.
    3. Last payment received 31 days ago.
    4. Risk: delivering without a credit decision takes the exposure to €228,700.
    5. Proposal: credit decision today, delivery of the 9 urgent orders once agreed.

    Not checked: payment history before 2025.

  5. You

    Show me the overdue balances per customer.

  6. Agent
    1. Reading due invoices
    Overdue invoices per customer
    • Orvalis Distribution42,300 €
    • Hôtel Belverane11,900 €
    • Maison Tessandre6,200 €
    • Others (7 customers)4,100 €

    Orvalis Distribution accounts for 66% of the overdue balance. I can turn this chart into a dashboard refreshed every day.

  7. You

    Give me the contact for the account.

  8. Agent
    1. Reading the customer’s contacts

    Camille Ravelot, purchasing manager at Orvalis Distribution. Her phone and email are on the contact record. Do you want me to draft a message listing the 3 overdue invoices?

Illustrative conversation: company, people and figures are fictional.
  • It reads, it does not change

    In this example the agent consults sales, invoicing and approval rules. Any write to the ERP goes through a human approval.

  • Every answer cites its sources

    The records consulted are linked in the answer: you check in one click, and you forward it to the team without rewriting.

  • Tasks that take minutes

    Cross-checking orders, balances and rules used to take an hour across three departments. The agent does it while you read.

Use cases

What an agent can do in your ERP

The same requests you would make to an analyst, written in plain English, run on your data.

  • Analyses

    Order-to-cash, stock, margins

    The agent reviews a whole process, spots the anomalies (piling drafts, exceeded limits, delays) and ranks them by importance.

  • Briefings and reports

    A brief ready to send

    A five-line briefing for the team, with the amounts, the risks, what was not checked and the links to the records.

  • Dashboards

    Created in one sentence, changed in another

    A one-off chart or a dashboard refreshed every day, without waiting for a development: you ask for a view, you adjust it by talking.

  • Talk to your ERP

    Without clicking through twenty screens

    Orders above €5,000 that need action, the contact of an account, the status of a delivery: you ask, the agent answers with the record.

  • Data entry and control

    Documents read, entries proposed

    Supplier invoices, purchase orders or product sheets received as PDF are read, matched and prepared in the ERP. Your teams approve before anything is written.

  • Alerts and follow-up

    Deviations surface on their own

    A balance above a threshold, a bypassed approval rule, negative stock: the agent watches and warns the right person.

ERP integration with MagentoETL and data syncAI in Magento

Custom-built

Built around your ERP, not around a demo

A generic agent knows neither your approval rules, nor your custom fields, nor your exceptions. Ours is engineered from your configuration, your extensions and your data: it learns your processes before it answers.

Odoo, Microsoft Dynamics, Cegid or an in-house ERP: we go through the API or the database, with the rights of the user who asks. Your CRM, PIM and OMS are added the same way.

  • 01ConfigurationRoles, rights, workflows, companies and currencies: the agent respects what your ERP already knows.
  • 02ExtensionsYour modules and custom fields are part of its vocabulary.
  • 03DataHistory, master data, documents: the agent reads at the source, never a rough copy.
  • 04ProcessesYour approval rules, thresholds and exceptions: written with you, tested on your real cases.

Models

You choose the model

The agent's code does not depend on the language model. We choose it with you, based on data sensitivity, expected performance and cost, and you can switch later.

  • Hosted in your country

    An open model served by a cloud provider in your country, so your data is processed under your own law: Mistral AI or OVHcloud in Europe, a local provider in Singapore, Australia or wherever you operate.

  • On your servers

    An open model installed on your infrastructure, on a dedicated server in a data centre or on premises. Nothing leaves your network.

  • The large market models

    Anthropic Claude or Google Gemini, through their APIs and under a data processing agreement, when data sensitivity allows it and quality justifies it.

Interchangeable: you switch models without rewriting the agent. The code and the data belong to you.

Talk about my project

Governance

Governance: rights, traces, approval, costs

  • Rights

    The agent has the rights of the user asking, nothing more. Your ERP roles apply as they are.

  • Traceability

    Every read and every write is logged, with the question asked and the records consulted.

  • Approval

    By default the agent reads freely and writes only after your approval. You decide, action by action, what it may do on its own.

  • Costs

    A budget per agent and per month, alerts before it is exceeded. An agent never loops at your expense.

Case studies

Projects where AI already works inside the systems

Grid of product cards, some highlighted: a catalogue enriched by AI

Magento 2 B2B · Pimcore · Odoo · LLM

Hospitality supplier

472,663 product sheets categorised and enriched by AI from the PIM and manufacturer data, then pushed to Magento and Odoo.

472,663products enriched
2,000sheets in 45 min
496suppliers
Carré Blanc gift ideas page with Balata collection close-up

ETL · Cegid, Octipas, Quable

Carré Blanc

More than 110 stores connected to the website: our ETL orchestrates the Cegid ERP, the OMS and the PIM. The integration base an agent relies on.

110+connected stores
3systems: ERP, OMS, PIM
M1 → M2migration and Hyvä
Bébé9 product page on Magento 2 and Hyvä, with split payment and the mobile store locator

Magento 2 · ETL · OMS, ERP, CRM

Bébé9

Stock and prices of more than 100 stores synced in real time across three systems.

100+Connected stores
12Months, fixed timeline
60Ten-day sprints

Method

From the diagnostic to an agent your teams keep alive

  1. 01

    Diagnostic

    45 minutes on your processes and your ERP. You leave with 3 agent use cases ranked by gain and feasibility.

    45-min video call, free · Book my diagnostic →

  2. 02

    Proof of concept

    One agent on one process, connected to a copy of your data, with success criteria set from the start. At the end, go or no-go.

  3. 03

    Production

    Rights, logging, model choice, cost monitoring. The agent lives inside your tools, not in a separate app.

  4. 04

    Handover

    Documentation, training and code handed to your teams: they extend the agent, with or without us.

FAQ

Your questions about AI agents and ERPs

Which ERPs do you work with?

Any ERP that exposes its data through an API or a database. In our projects we have already connected Odoo, Microsoft Dynamics AX and Cegid, as well as ERPs built in-house by our clients. The CRM, PIM and OMS connect the same way.

Can the agent change data in the ERP?

By default, no: it reads, analyses and proposes. Writes (creating an order, holding a delivery, sending a message) go through a human approval. You can then allow some actions, one by one, once trust is established.

Where does our data go?

Where you decide. With a model hosted by a provider in your country or installed on your servers, it never leaves the perimeter you set. With a market model, a data processing agreement frames the use. In every case the agent only reads what the user is allowed to see.

Which language model should we choose?

The one that suits your data and your use cases: an open model served by a provider in your country when the data must stay under your law (Mistral AI or OVHcloud in Europe), an open model on your servers for the most sensitive data, Anthropic Claude or Google Gemini when their quality makes the difference. The agent's code stays the same: you can change your mind.

How much does it cost?

Two items: building the agent (scoping, integration, tests, handover) and running it (calls to the model, billed on usage or included in your server). We quote both at the diagnostic, and we set a monthly budget per agent with alerts.

Do we need a technical team on our side?

To build the agent, no: we take care of it. To keep it alive, one person who knows the process well is enough at first; we train your teams and hand them the code and the documentation.

What if the POC does not deliver?

The success criteria are set before we start. If they are not met, you stop there, with no commitment for what follows, and a clear picture of what the agent can and cannot do.

Contact

An agent in your ERP?

Enter your work email and your company: we reply within 2 business days to set up the diagnostic. You can describe your ERP right after.

  • Reply within 2 business days
  • No commitment
  • Your data stays confidential


    Your information is only used to answer your request. Legal notice and privacy

    Talk about my project