Draft
Letters, replies to residents, internal notes. The officer reviews and signs: the AI prepares, it never sends anything.
AI · local government
AI for your staff. Your residents' data stays with you.
An internal AI assistant, on an open model, hosted in France, reachable from your network only. Not a confidentiality promise: an architecture where nothing leaves, and that can be verified.
In production at the Grand Chambéry metropolitan authority, France.
Internal assistant
Hosted in France · nothing leaves ▍
Example · fictional authority and documents
What changes for a public body
Generative AI in local government: four constraints a company does not have.
- 01
Residents' data
Civil records, social care, planning, HR: what your staff handle cannot be sent to a service hosted outside Europe.
- 02
Usage that already exists
Your staff have discovered generative AI. Without an approved tool, they use it on their personal phones, out of sight.
- 03
A training obligation
Article 4 of the EU AI Act requires the people who use AI to be trained in it. You need a framework, and a usage policy.
- 04
Accountability
To elected members, to residents, to oversight bodies. A guarantee that can be verified is worth more than a contractual promise.
What your staff do with it
The uses of generative AI you find in every local authority.
Summarise
Council decisions, reports, meeting minutes: the essentials in a few lines, for officers and elected members alike.
Rephrase
Make an administrative text readable for a resident, or adapt a note for another audience.
Find
Once your documents are connected: regulations, procedures, contracts, with the source cited in every answer.
The Grand Chambéry case
Internal AI for the staff of a metropolitan authority.
Staff wanted to use AI. The authority could not send its documents to a service hosted outside Europe. Banning it meant usage moving to personal phones; allowing it meant data leaving.
We delivered a conversational assistant end to end, infrastructure included, then maintained it for twelve months. No outbound calls.
The deployed architecture
The authority's internal network
Staff
Named accounts, access from the internal network
Assistant
Open French model
Server
Hosted in France
External AI service— No outbound calls
What we set up
A tool, a framework, then your documents.
The foundation
An internal AI assistant
An open model, hosted in France or in your own infrastructure, reachable from your network only, with named accounts.
The framework
A usage policy and staff training
What can be handed to AI, what cannot, and how to check an answer. What the AI Act asks for.
Next
AI connected to your documents
Answers drawn from your council decisions, your regulations, your internal procedures, without them leaving.
Further reading: deploying AI without a single piece of data leaving and what GDPR lets you hand to AI. Our whole AI offer is on our AI page.
Deployment
From the first meeting to AI in service, in five steps.
- 01
2-hour diagnostic
Your staff's real usage, the data at stake, and what AI can take on.
Free - 02
Infrastructure
A server hosted in France or on your premises, access from your network only, named accounts.
Nothing leaves - 03
Policy and training
What can be handed to AI, what cannot, and how to check an answer.
AI Act, Article 4 - 04
Go-live
Staff log in with their own account, and every use is traced.
Turnkey - 05
Maintenance
Model and infrastructure updates, usage monitoring. At Grand Chambéry, twelve months.
Over time
Your questions
What local authorities ask us.
“Which AI model do you use?”
An open model, French or European, chosen for your staff's uses. At Grand Chambéry, it is an open French model. It is hosted in France, or directly in your infrastructure.
You always know where your data goes.
“Is it as good as ChatGPT?”
Not on the hardest tasks. On what an officer asks ten times a day, drafting, rephrasing, summarising a document, answering from an internal knowledge base, an open model is more than enough.
The right choice is the best model that meets the constraint.
“How much does it cost?”
Infrastructure and maintenance, not a per-user subscription. The cost depends on the model and on the number of simultaneous users. At Grand Chambéry, the infrastructure switches off outside working hours.
The cost does not climb with adoption.
“What about GDPR?”
AI hosted in France, with no outbound calls and named accounts, makes compliance much simpler: there is no transfer to frame, and every use is traceable.
The rules are set out in our article on AI and GDPR.
The free prototype
1week
to see AI wired into one of your real processes, with your real documents. For free.
2 h
We find where you are losing time, together
1 week
We build the prototype, for free
Then
You only pay if you keep it, at a fixed price
4 prototypes a month · by application · no commitment