Specialist ERPsERP Platform

Dolibarr + private AI

AI for Dolibarr ERP, Without Sending Small-Business Data to a Cloud AI Vendor

Short answer

AI for Dolibarr connects a private LLM to Dolibarr's REST API and MySQL/MariaDB tables so an owner or small operations team can ask plain-English questions about stock, invoices, and manufacturing orders and get grounded answers, without paying for or trusting a third-party cloud AI service. It runs on the same self-hosted LAMP-style server most Dolibarr instances already use.

ERP: Dolibarr ERP CRM
Industries: Manufacturing, Distribution
Written for: Owner

Discuss your Dolibarr ERP CRM environment See use cases

Dolibarr is the ERP most small manufacturers and distributors reach for when a heavier platform is overkill and a spreadsheet has stopped being enough. It is open source, modular, and light enough to run on a modest self-hosted server, which is exactly why it has spread widely among owner-operated businesses in Europe and beyond since it originated in France.

The trade-off for that simplicity is a manufacturing module (MRP) that covers basic bill-of-materials and manufacturing order needs but genuinely limits companies with more complex routing, work center, or multi-level BOM requirements. Most Dolibarr shops work around this with a mix of the native module, spreadsheets, and manual coordination, which is manageable at a small scale but leaves the owner as the person who has to remember where each piece of information lives.

Reporting is similarly minimal: Dolibarr's built-in list views and basic exports answer routine questions, but anything more specific, like "which customers haven't ordered in ninety days but used to order every month," usually means exporting to a spreadsheet and building a pivot table by hand, a task that falls to whoever in the business is comfortable with Excel.

This page covers how a private AI layer sits on top of Dolibarr's REST API and database for a small business that wants better answers without the cost or data-exposure risk of a large enterprise AI platform, and what to weigh before connecting any AI tool to a live Dolibarr instance.

What usually gets in the way

The problems we hear most from owner teams running Dolibarr ERP CRM.

The manufacturing module has real limits

Dolibarr's MRP module handles basic bills of materials and manufacturing orders well but struggles with complex multi-level BOMs, detailed routing, or work-center capacity planning that a growing manufacturer eventually needs.

Reporting stops at list views and basic exports

Beyond Dolibarr's native list filters and CSV exports, any real analysis requires manually building a spreadsheet, a task that usually falls to the one person in the business comfortable doing it.

Module marketplace quality is inconsistent

The Dolibarr module marketplace adds real functionality, but quality and maintenance vary widely between modules, and vetting a community module before installing it on a production system takes real diligence.

Any real customization needs a PHP developer

Configuration through the admin interface covers a lot, but anything beyond that (a new hook, a custom trigger, a tailored workflow) requires a developer comfortable with Dolibarr's PHP codebase, which small businesses rarely have on staff.

Owners are the default analysts

In a business too small for a dedicated operations or finance analyst, the owner ends up being the one who has to interpret stock levels, overdue invoices, and manufacturing status personally, on top of everything else they do.

Where AI earns its place in Dolibarr ERP CRM

Each use case names the ERP objects it reads or writes, so your ERP team can judge the integration effort before anyone commits budget.

Natural-language queries over sales, stock, and invoices

An owner asks a plain question like "which invoices are more than sixty days overdue" and gets an answer built from live Dolibarr data instead of exporting a list and filtering it by hand.

Touches: llx_facture, llx_societe, llx_stock

Outcome: Replaces a recurring manual export-and-filter task with a direct answer in seconds.

Low-stock and reorder alert agent

The agent monitors stock levels against configured minimums and drafts a reorder list for the owner to review, rather than requiring someone to check stock levels manually each week.

Touches: llx_product_stock, llx_product, llx_commande_fournisseur

Outcome: Reduces the chance of a stockout on a fast-moving item that nobody happened to check that week.

Invoice and payment matching assistant

The assistant reviews incoming payments against open invoices and flags mismatches or partial payments needing attention.

Touches: llx_facture, llx_paiement, llx_bank

Outcome: Cuts the time spent manually reconciling payments against invoices at month end.

Manufacturing order status query

Someone on the floor asks about the status of a manufacturing order (MO) and linked bill of materials without opening the full Dolibarr interface.

Touches: llx_mrp_mo, llx_bom_bom, llx_stock

Outcome: Gives a quick status answer without interrupting whoever is managing the order in the system.

CRM opportunity follow-up drafting

The assistant drafts a follow-up note for a stalled CRM opportunity based on the last recorded interaction, for the owner or salesperson to review and send.

Touches: llx_actioncomm, llx_societe, llx_opportunity

Outcome: Reduces the number of opportunities that go cold simply because nobody remembered to follow up.

Multi-entity consolidated view

For businesses using Dolibarr's multicompany module, a single question resolves across every entity the requesting user has access to.

Touches: llx_entity, llx_facture, llx_stock

Outcome: Gives an owner running multiple related businesses a combined view without exporting and merging spreadsheets from each one.

Custom module and hook documentation

The assistant reads installed module code, hooks, and triggers to explain what a past customization actually does, useful when the developer who built it is no longer available.

Touches: htdocs custom module files, llx_hookmanager entries

Outcome: Reduces the risk of an undocumented customization becoming a mystery when it eventually needs to change.

Reference architecture

The AI layer connects through Dolibarr's own REST API and, where needed, a scoped read-only database role, so answers stay grounded in live data on the same self-hosted server most Dolibarr instances already run on.

1
Dolibarr connector
Connects via the REST API (api/index.php) using a dedicated API key, or through a scoped read-only MySQL/MariaDB role for heavier queries the REST layer is not built for.
2
Semantic layer
Maps Dolibarr's llx_ table structure and module-specific fields to plain business terms an owner or small operations team actually uses.
3
Model serving
A smaller open-weight model served with Ollama on modest, right-sized hardware, since Dolibarr's typical transaction volumes do not require enterprise-scale GPU infrastructure.
4
Retrieval and agents
Retrieval-augmented generation grounds answers in live queries; any agent that drafts a reorder list or follow-up note requires the owner's review before anything is sent or submitted.
5
Governance and audit
Queries and answers are logged against the requesting Dolibarr user, matching the access already configured in Dolibarr's permission system.

Integration notes for your ERP team

Connects via Dolibarr's REST API using a dedicated API key (DOLAPIKEY) scoped to the modules actually needed, not a full administrator account.
For queries the REST API is not built to handle efficiently, a scoped read-only MySQL/MariaDB role provides direct access to llx_ tables.
Module marketplace additions are reviewed during discovery so the semantic layer accounts for any non-standard tables they introduce.
Write-back actions (reorder suggestions, follow-up drafts) are queued for the owner's review, never submitted automatically.
Custom hooks and triggers are read as source, not executed, when generating documentation, avoiding any risk of triggering unintended automation.
Works with the typical self-hosted LAMP or LEMP stack most Dolibarr installs use, without requiring a migration to a different hosting model.

Deployment options

Air-gapped on-prem

The typical fit for a small manufacturer running Dolibarr on a self-hosted LAMP-style server already, often the same server the ERP itself runs on.

The model and connector run on the same server or a modest additional machine on the local network, with no dependency on an outbound API.

Private or sovereign cloud

Businesses running Dolibarr on a rented VPS who prefer not to add hardware on-site.

Deployed in a small dedicated cloud instance under the customer's own account, separate from any shared multi-tenant AI service, sized to match a small business budget.

Hybrid

Rare for a business this size, but relevant for an owner running a few related entities on the multicompany module across different locations.

A lightweight connector reaches each Dolibarr instance separately, keeping data for each entity where it already lives.

Compliance and data control

How the architecture supports your obligations. Certification and accountability stay with your organisation; the design keeps the evidence straightforward.

GDPR

Given Dolibarr's strong presence among French and European small businesses, keeping the AI layer on customer-controlled infrastructure avoids sending customer or supplier personal data to a third-party AI API, simplifying GDPR compliance for a business without a dedicated compliance function.

PCI DSS (where a payment module is active)

The AI layer is scoped to read-only access outside cardholder-data tables, and payment integration modules are excluded from retrieval by default.

Basic customer assurance for a small supplier

For a small manufacturer supplying larger customers, being able to describe where data is processed and stored in plain terms is often all that is needed to satisfy a customer's vendor questionnaire.

Data continuity for a single-admin business

Documentation use cases reduce the risk that critical business knowledge exists only in one owner's head or one developer's memory.

Where Netray fits

DataRay

Dolibarr's straightforward REST API and database structure fit DataRay's approach of chatting directly with an existing data source, right-sized for a small business budget.

Custom build

Instances with community marketplace modules or bespoke PHP customization often need a connector tailored to those specific additions rather than a generic template.

How an engagement runs

Phase 1 . 1-2 weeks

Discovery

-Inventory of installed modules and any custom fields or hooks
-Review of current Dolibarr user roles and permissions
-Priority use cases ranked by owner and staff pain
-Right-sized deployment recommendation for the business's scale

Phase 2 . 3-5 weeks

Pilot

-Working connector against a staging or backup Dolibarr instance
-2-4 use cases live for the owner and key staff
-Accuracy review against known correct answers
-Governance validated against existing Dolibarr user roles

Phase 3 . 2-3 weeks

Production

-Cutover to the production instance with a scoped API key
-Basic audit logging in place
-Simple runbook for adding a use case without needing a developer each time
-Short handoff session for the owner or whoever administers the system

Phase 4 . Ongoing

Scale

-Additional use cases added as the business grows
-Expansion to a second entity if the multicompany module is adopted
-Model refresh as open-weight options improve
-Periodic check-in to confirm the deployment still matches business size

Questions to ask any vendor, including us

A short list that separates real Dolibarr ERP CRM AI work from a chatbot demo.

Does the AI layer use a scoped API key, or does it need broader access than a normal Dolibarr user would have?
Where does the model actually run, and can that be confirmed in plain terms for our own customer questionnaires?
How are community marketplace modules we have installed accounted for in the setup?
What happens if we later add or remove a module, does the AI layer need to be rebuilt?
Is every question and answer logged against the specific Dolibarr user who asked it?
What is the actual hardware requirement, and is it proportionate to a business our size?
Can we start with a small number of use cases and add more later without a large upfront cost?
If we ever stop using this, do we keep the connector setup and any documentation it produced?

Frequently asked questions

Is this practical for a very small Dolibarr installation?

Yes. Dolibarr's typical user base is small businesses, and a right-sized deployment, a smaller open-weight model on modest hardware, focused on two or three high-value use cases, is built to match that scale and budget rather than assuming an enterprise-sized project.

Does this replace Dolibarr's manufacturing module?

No. It sits alongside the existing MRP module and other Dolibarr modules, answering questions about the data already there. If the manufacturing module itself is outgrown, that is a separate ERP question, not something an AI layer can solve.

How does this handle Dolibarr's module marketplace additions?

Installed modules are reviewed during discovery, since Dolibarr's marketplace introduces a wide variety of custom tables and structures. The semantic layer is built to account for whatever modules a specific instance actually has installed.

Can the owner run this without an IT department?

Yes, that is the intended fit. Discovery, setup, and handoff are scoped for a business without a dedicated IT staff, with a simple runbook so routine use does not require ongoing developer involvement.

Is data sent to a cloud AI service like ChatGPT?

No. The model runs on infrastructure the business controls, either the same server Dolibarr runs on or a small dedicated instance, so customer, supplier, and financial data does not pass through a third-party cloud AI API.

What does this cost for a small business?

Costs scale with the size of the deployment and number of use cases; a small Dolibarr installation typically needs modest hardware and a narrower scope than a larger ERP project, keeping the investment proportionate to the business.

Can the AI create invoices or purchase orders automatically?

It can draft suggestions, like a reorder list or a follow-up note, but the owner or a staff member reviews and submits anything before it becomes a real record in Dolibarr. Nothing is created automatically without that review.

Related guides

RAG + SQL + permissions

A Private LLM Grounded on Your ERP Data

How a private LLM answers questions on your ERP data: RAG plus text-to-SQL, role-based permissions inherited from the ERP, and where each fits.

Ask your ERP anything

Natural Language Query for ERP Data: Ask SAP, Infor, or Oracle a Question in Plain English

See how natural language query over SAP, Infor, Oracle, and NetSuite data works: grounded text-to-SQL, role-based permissions, and a visible audit trail.

Agents + approval gates

AI Agents for ERP, Running On-Prem

A practical guide to on-prem AI agents for ERP: what they can safely automate, where human approval belongs, and how to design the guardrails.

ERP AI Cost Guide

What ERP AI Actually Costs: A CFO's Guide

A CFO's guide to what ERP AI actually costs: GPU hardware, model licensing, integration and connector work, and realistic ongoing run-rate ranges.

ERP AI Buyer Guide

How to Choose an ERP AI Implementation Partner

A CIO checklist for picking an ERP AI implementation partner: the architecture questions to ask, red flags, pricing models, and what to demand in the SOW.

QAD + on-prem AI

AI for QAD Adaptive ERP in automotive and industrial manufacturing

Add AI to QAD Adaptive ERP or Enterprise Edition for automotive and industrial manufacturing, grounded on QXtend and QAD's API layer, on-prem or private cloud.

Plan it with numbers

Free Tool

Talk it through with an engineer who knows Dolibarr ERP CRM

Bring one real question your team cannot answer from the ERP today. We will map the data path, the model, and where it runs, and tell you honestly if AI is the wrong tool for it.

Start the conversation See ERPray

All AI for ERP guides