Agentic AI Platform: Build and Deploy AI Agents | Axelor
Axelor Agent Suite

Agentic AI platform: build, orchestrate and automate with AI

One platform to build applications and business processes with AI, and to deploy agents that act directly inside your operations.

French software vendor, sovereign by design ISO 27001 certified Open source
erp.axelor.com/agents
Procurement Agentrunning the request Running
"The valves for the Nantes site are running late. Find an alternative and place the order."
Delay confirmed on order PO-221411 days behind the promised date
3 alternative suppliers comparedlead time, price, certification level
Order PO-2231 createddelivery rescheduled for 12 September
Approval required: amount above thresholdwaiting on the procurement manager
3 actions completed, 1 decision escalated to a human
Who it is for

Built for complex organisations

Where processes cut across several departments, several tools and several levels of approval.

Large enterprises

Your processes run across departments, subsidiaries and applications that do not talk to each other. The platform orchestrates the whole chain and hands over to agents the coordination work that used to tie up entire teams.

  • Cross-system orchestration
  • Autonomous agents on high-volume processes
  • Incremental rollout, one department at a time
  • Full control over hosting

Public sector

Data sovereignty, an audit trail on every decision, and approval chains that are actually enforced. Axelor is a French software vendor, and you keep control of both the hosting and the source code.

  • Sovereign solution from a French software vendor
  • Hosting in a private cloud, on premises, or with a sovereign provider
  • Open, auditable source code
  • ISO 27001 certification
The turning point

An engine executes. An agent decides.

Both cut down on manual work, but not in the same way. Everything hinges on the cases that fall outside the expected scenario, and those are exactly the ones that eat up your teams' time.

Conventional automation

A rule written in advance

The outcome is predictable, the scope is fixed.

  • Runs exactly what was programmed, and not one case more
  • Stops as soon as the situation falls outside the scenario
  • Needs a new rule for every business variant
  • Cannot read a document or a request written in plain language
  • Maintenance cost that grows with the number of rules
VS
With Axelor Agent Suite

A goal to reach

The agent picks the path, you set the limits.

  • Assesses the actual situation and chooses the right action
  • Handles the cases nobody planned for instead of blocking the process
  • Understands a request written in plain language
  • Chains several steps across several modules, end to end
  • Escalates to a human above the threshold you have set
Automation handles the happy path. It is the exceptions that cost you: they are what pulls your teams into coordination, chasing and judgement calls.
How it works

From request to action, in four steps

It is this complete chain that sets an autonomous agent apart from a tool that only answers questions.

01Step 01

You ask

In plain English, the way you would ask a colleague. No rules to configure.

02Step 02

It reads the context

It queries your actual data, not a generic corpus.

03Step 03

It decides

It weighs the options and picks the best one, within the scope you have set.

04Step 04

It acts

It writes to the ERP. Above your threshold, it waits for human approval.

Action completed, and fully traceable end to end
Build with AI

Describe it. AI builds it.

State what you need in plain language and AI generates a first version. You then take over in the Studio, without writing any code.

  • Agent Studio: generation driven by your description
  • Business objects, forms, fields, menus and dashboards
  • Manual refinement in the Studio, property by property
  • No development cycle and no redeployment
AI generating an application screen from a plain-language description in Axelor
Orchestrate with AI

Create your business processes in plain language

Describe the process you have in mind and AI generates its structure. You then adjust it visually, rather than starting from a blank page.

  • States, conditions and branches, including parallel branches with merge and synchronisation
  • Sub-processes to break down complex chains of steps
  • Human tasks and approvals sitting alongside the automated steps
  • Process instances and their states monitored from the admin console
Business process generated by AI in Axelor Agent Suite, with its states, branches and human tasks
Deploy agents

Agents that do more than answer

Give your agents access to the business tools and functions they need to assess a situation and trigger actions directly inside Axelor.

Tools, not just answers

Every agent comes with tools that map to real functions of the application.

In practice the agent understands the request, decides, calls the tool and writes to Axelor, within the user's permissions and the approvals in place.

The right agent for each need

Several specialised agents live side by side in the same interface. You call the one you need with a mention.

In practice Agent Studio to build, a sales agent for a customer file, and you switch agents without leaving the conversation.

Agents inside your processes

An agent is not confined to a conversation: it takes ownership of a step inside a process.

In practice the automated step is handled by the agent, the sensitive step stays a human task, both within the same chain.
Work with collaborative AI

AI that is genuinely collaborative

Work alongside your teams and your agents in a single space. Talk to colleagues, call on AI when you need it, and share the same working context.

A conversation worth keeping becomes a persistent space: you name it, you find it again, and it turns into the permanent home of a topic or a team.

  • Conversation sharing and group discussions
  • Three permission levels: view, take part, or query the agent
  • Two modes, human to human or with AI, with instant switching
  • A one-off question to AI without leaving the discussion
  • Mentions to call in a person or an agent
  • Notification when a user is invited or mentioned
  • Persistent spaces, with an owner and participants
  • Conversation summary on demand, including from a previous summary
Space | Nantes tender4 participants and 1 agent DiscussionAgent
ML Marie L. | Site manager

The supplier is quoting an 11-day delay on the valves.

TB Thomas B. | Buyer

@Procurement Agent could you compare the certified alternatives?

AI Procurement Agent

3 suppliers compared on lead time, price and certification. Only one can hold the date. Order drafted, waiting for approval.

Conversation summary

Supplier delay identified, certified alternative selected, replacement order awaiting approval from the procurement manager.

AI connected to your context

AI that knows your business context

Query AI straight from your data. It works in the context of the business record in front of you, not in a vacuum.

The context of the open record

From a sales order, a product or a customer, AI reasons on that specific record and on the data attached to it.

The history of your exchanges

Conversations can be attached to a business record and found again from it. The context does not get lost from one week to the next.

Your knowledge bases

Procedures, reference data and internal documents feed the answers, alongside your operational data.

Stay in control

AI that acts safely

Agents plug into Axelor's permission model. A user cannot get through AI what the application denies them.

  • The user's read and write permissions apply to the agent
  • Access to agents controlled by user, group and role
  • Requests are analysed before execution, with optional confirmation on sensitive actions
  • Axelor does not use your data to train any model, and you choose the provider
Permission and agent access controls for AI agents in Axelor
The platform

The full power of the Axelor platform

The building blocks below share the same data model and the same permission system.

Studio Business processes AI agents BI API Connectors Templates No-Code
Where it sits in the range

Built into the ERP, or deployed on its own

All four suites share the same agentic layer. What changes is the deployment model.

Integration

It adds to your information system, it does not replace it

You do not have to change ERP or switch off your business applications to deploy agents.

Open API

A documented REST API exposes the platform's objects and actions. Your existing applications trigger agents, and agents write back to them.

Application connectors

Accounting, payroll, CRM, e-commerce, supplier EDI, industrial sensors: the flows connect without heavy development.

Alongside the Axelor ERP

If you already run the Axelor ERP, the platform shares its data model and its permissions directly. There is no bridge to build.

Rollout

How to deploy agents without a transformation programme

The most common reason this kind of platform fails is a project that is too broad and launched all at once. The method below works one scope at a time.

Pick a first process

High volume, stable rules, limited risk. The aim of this first step is to measure a real gain, not to cover every case.

Set the agent's scope

Define the tools it can use, the data it can reach and the threshold above which a decision goes to a human.

Observe it in real conditions

Run the agent with approval on every action, review its decisions, then raise the autonomy threshold step by step once confidence is established.

Extend to neighbouring processes

Reuse the same foundation on adjacent processes, one department at a time. Agents work together rather than in isolation.

Definitions

The vocabulary of agentic AI, in plain terms

These terms get used a lot and often mean very different things. Here is what they mean at Axelor.

Agentic platform

An environment for building applications and processes from a plain-language description, then deploying autonomous agents that carry out tasks and take decisions within a defined scope.

Autonomous AI agent

Software that pursues a goal rather than a list of instructions. It assesses the situation, chooses the action and carries it out on your data, without a human triggering each step.

Process orchestration

Coordinating the steps, roles and business rules of a process that runs across several departments and several applications, with visibility on progress and ownership.

No-Code

The ability to build and change an application without writing code: screens, fields, menus and rules are all set through configuration. At Axelor it is also the layer that makes what AI produced visible and editable.

Multi-LLM

The ability to run the platform on different language models, hosted in the cloud or on your own servers, without depending on a single provider.

Human approval threshold

A limit you set for each agent, above which a decision is no longer carried out automatically but submitted to a person. It can be based on an amount, a criticality level or a type of action.

FAQ

Frequently asked questions about Agent Suite

What is Axelor Agent Suite?
Axelor Agent Suite is an enterprise agentic AI platform. It does three things: it builds applications and screens from a plain-language description, it orchestrates business processes with their states, branches, human tasks and approvals, and it deploys autonomous agents that carry out real actions inside the application. Unlike the agentic capabilities built into Axelor Service Suite, Axelor Build Suite and Axelor Factory Suite, Agent Suite can also be deployed as a standalone platform, independently of the ERP.
What is an agentic AI platform?
An agentic AI platform lets you build applications and processes from a plain-language description, then deploy autonomous agents that carry out tasks and take decisions within a defined scope. One point sets it apart from conventional automation: automation runs a rule written in advance, whereas an agent assesses a situation and chooses the action, within the limits it has been given.
How does an agent actually act inside the application?
Through tools, which map to real functions of the application. The agent understands the request, decides which action fits, calls the matching tool and writes to Axelor. That mechanism is what sets it apart from a conversational assistant: it does not produce an answer, it produces an effect on your data, within the user's permissions and any required approvals.
What does collaborative AI mean?
It means that people and agents work in the same conversation space. A conversation can be shared, several people can take part, and there are three permission levels: view, take part, or query the agent. Two modes coexist, human to human and with AI, and you switch between them instantly. A conversation worth keeping can become a persistent space, named and retrieved later like a standing working group.
Does AI know the context of my data?
Yes. From an open business record, a sales order or a product for instance, AI reasons in the context of that record and the data attached to it. Conversations can be linked to a record and found again from it, so the context is not lost from one week to the next. Your internal knowledge bases complement your operational data.
How do I get the gist of a long conversation?
You can request a summary at any point: AI returns the key information from a conversation, even after dozens of messages. You can summarise from the start, produce a new summary based on the previous one, and delete a generated summary. The summary takes into account the business record the conversation is linked to.
Can AI be used to bypass application permissions?
No, and that is by design. Agents plug into Axelor's permission model: the user's read and write permissions apply to the agent, and a user cannot get through AI what the application denies them. Access to the agents themselves is controlled by user, group and role, so nobody can call on an agent they have no access to. Requests are analysed before execution, and sensitive actions can require confirmation.
Can a process be created without building it by hand?
Yes. You describe the process you want in plain language and AI generates its structure. You then adjust it visually rather than modelling everything from a blank page. The engine that runs it handles states, conditions, parallel branches and their merge, sub-processes, along with human tasks and approvals.
Do I need the Axelor ERP to use Agent Suite?
No. That is the difference with Axelor Service Suite, Axelor Build Suite and Axelor Factory Suite, where the agentic capabilities are built into the ERP. Agent Suite can also be deployed as a standalone platform, above your existing information system, through its APIs and connectors. You can therefore start with the platform without replacing your management applications.
Do I need developers to create agents or screens?
Not for everyday work. An agent is configured from the interface by describing its goal, its tools and its scope. For screens, AI generates a first version that your teams refine in the No-Code Studio. Technical profiles remain useful for specific integrations and for exposing new tools through the API.
Is my data used to train a model?
Axelor does not use your data to train any model. Beyond that, processing depends on the model provider you choose and on their own terms: that is precisely the point of the Multi-LLM approach, which lets you pick a provider whose policy suits you, or run the agents on a model hosted on your own infrastructure when the data is sensitive enough to warrant it. You also choose how the platform is hosted: cloud, private cloud, a sovereign European provider, or your own servers.
What does a rollout look like?
One scope at a time, rather than all at once. Most projects start with a high-volume, low-risk process, which makes it possible to measure the real gain and to calibrate the approval thresholds. Neighbouring processes are added afterwards, one department at a time. This incremental approach avoids the long transformation programme, which is the most common reason this kind of platform fails.
How is this different from Axelor's agentic ERP?
They are two ways into the same agentic layer. The agentic ERP refers to the Axelor ERP with its AI capabilities built in, for a company that wants to run its business and get the benefit of agents inside its modules. Agent Suite is the platform itself, sold and deployable independently, for an organisation that wants to orchestrate and automate processes above an information system that is already in place.

See Agent Suite on your own processes

Our experts build an application and an agent in front of you, on a case taken from your own organisation.