Claude for teams and agents connected to the ERP
For companies that want to use AI in their daily work, with their own documents and their ERP. We work with Anthropic under a formal agreement: we train teams on Claude and put into production agents that work in Sage X3 and Cegid PRIMAVERA, with human validation at critical points.
- 01
Anthropic Partner Network
We work with Anthropic under a formal agreement.
- 02
Connected to the ERP through MCP
Agents in Sage X3 and Cegid PRIMAVERA.
- 03
A person decides
Above a limit, the agent proposes and someone approves.
Who it is for
Teams already using AI without rules
People using AI assistants on their own, without knowing what data they can put in or where it is kept.
Document-heavy areas
Contracts, manuals, invoices, manifests and support history that someone reads and summarises every day.
Finance and purchasing in the ERP
Matching invoices to orders, recurring postings and reconciliations that currently pass through several hands.
Development teams
People who maintain ERP customisations and integrations and want to use Claude Code on code and queries.
How an agent runs in production
AI proposes, a person approves at the critical points. Each type of task has a value limit and a confidence limit, written into the agent's configuration.
Trigger
An event in the ERP, an email or a user request starts the run.
Gathering
The agent reads data from the authorised systems and gathers the context it needs.
Proposal
Produces the output (posting, answer or document) with the reasoning and the source of the data.
Validation
Above the defined limit, a person approves, corrects or rejects.
Logging
The action and the decision are logged and auditable, and feed into the review of the limits.
What we use from Claude
Claude Team and Enterprise
Plans for organisations, with single sign-on (SSO), connectors and admin controls. By default, company content is not used to train the models.
Connectors
Connection to company sources, such as Microsoft 365, approved by the administrator and with permissions per tool.
Control and logging
On the Enterprise plan, role-based permissions (RBAC), audit logs and data retention rules.
Model Context Protocol (MCP)
Anthropic's open standard for connecting AI assistants to the systems where the data lives. We connect Claude to the ERP through MCP.
Claude Code
Anthropic's coding agent, which runs in the terminal, in the editor and on the web, for people who maintain code and integrations.
How it compares with the alternatives
Many companies already use AI in some form. The difference lies in the rules, the connection to the ERP and who is accountable for what the agent does.
| AI assistants used on people's own initiative | AI licences without support | In-house development | Claude AI | |
|---|---|---|---|---|
| Company content kept out of model training by default | partly | yes | partly | yes |
| Usage policy and data classification configured | no | partly | partly | yes |
| Training with the company's own documents and processes | no | no | partly | yes |
| Connection to Sage X3 and Cegid PRIMAVERA through MCP | no | partly | partly | yes |
| An agent with the permissions of the person who invokes it | no | no | partly | yes |
| Human validation queue above a limit | no | no | partly | yes |
| A record of each run, ready for audit | no | partly | partly | yes |
What we do with Claude
We train teams to use Claude day to day and put agents to work on repetitive tasks in the ERP.
Applied training
Hands-on sessions on each department's documents and processes, using the company's own cases.
Assistants on internal data
Search and answers across contracts, manuals and support history, citing the source of each answer.
Usage policy and governance
Usage rules, classification of sensitive data, logging of interactions and configuration of admin controls.
Agents connected to the ERP
Agents that read and propose postings in Sage X3 and Cegid PRIMAVERA via MCP, with the permissions of the user who invokes them.
Document processing
Reading invoices, manifests and delivery notes, matching them to orders and recording them in the source system.
Human validation
Above a value threshold, or when confidence is low, the case goes to a queue where a person approves, corrects or rejects it.
To get started
- Usage policyrules, sensitive data and Claude administration controls
- Applied trainingsessions by area, using the company's own cases
- First agenta reversible, low-value task verifiable by rule, with low limits
- Limit reviewautonomy increases when proposals are approved without correctionsafter weeks of history
Sevways is part of the Anthropic Partner Network and uses Claude in the agents it puts into production on Sage X3, Cegid PRIMAVERA and clients' own systems.
Works with
Systems where our agents read and write.
Sage X3
ERP for manufacturing, distribution and multi-company groups
View product →Cegid PRIMAVERA
Multi-company ERP with Angolan and Mozambican taxation
View product →GANTRY
Integration and EDI with terminals, shipping lines and customs
View product →RACON
Documents, alerts and approval workflows alongside the ERP
View product →
Frequently asked questions
Is our data used to train the models?
On the Team and Enterprise plans, Anthropic does not use company content to train models by default. We also set up the usage policy and the classification of data that can be shared.
Can the agent do things in the ERP that the user cannot?
No. The agent runs with the permissions of whoever invokes it, or of a profile created for it, designed the same way as people's profiles. The authorisation rules of Sage X3 or PRIMAVERA still apply.
What decisions can the agent make on its own?
Only those that are reversible, low in value and verifiable by rule. Payments, bank details and master data with tax impact always stay with a person. The article in Resources explains the criteria.
What is Sevways' relationship with Anthropic?
We work with Anthropic under a formal agreement. We use Claude in the agents we put into production on Sage X3, Cegid PRIMAVERA and clients' own systems.
Further reading
AI agents connected to the ERP: where it is safe to let them decide
Value limits, permissions inherited from the user and the design of the human validation queue.
An agent that looks for VAT errors: what it needs to know about each country
In September Cegid showed an agent that diagnoses the VAT settlement. An agent like this is only as good as the rules it knows, and the rules change from country to country.
Choose the first task for an agent
We go through your processes with you and choose a task that is repetitive, reversible and verifiable by rule.
Got a project? Let us talk.
Answer within 24 working hours.