Chatbot and AI agent for administrative information
Deploy an agent that answers the public's questions about their procedures, around the clock, from reliable and cited sources — and that hands over to a public officer when needed. Hosted in France — on local inference or an isolated resource — the data stays under French law. The AI agent assists, the public officer decides.
Updated on
A direct link to the “School enrolment” online service has been added to the answer.
⛓ Source · the authority's validated document base + service-public.fr
No decision about entitlement is taken automatically.
✎ Action · routing prepared — the front-desk officer settles the borderline cases
For a local authority or a public body, a Blue Lemon Agent agent answers the public's questions about their administrative procedures — on the web and at the counter — 24/7, from a validated document base whose sources it cites. It routes people to the right department or online service and hands over to a public officer as soon as a case is uncertain. It runs on local inference or is hosted in France under French law, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. No decision about a person's rights is automated. Live within a few weeks. Your public-sector staff write to it from Microsoft Teams, Slack or their email, and your users reach it on WhatsApp Business, the website chat or email — with no account to create and nothing to install. Reaching the administration from the tool people already have means less non-take-up of rights and equal access to the service. These connections are included in every plan, at no extra cost, within the number of connections your level includes.
Reference points describing our offer, not results measured at a client. The scale of the gain is confirmed by a pilot on your own scope.
Why administrative information is a pressure point — and why authorities hesitate
The public expects an immediate answer, at any hour and in their own language, while counters and switchboards are saturated. But the data involved — residents' situations, ongoing procedures — belongs to the public service and cannot be entrusted to just any tool.
! The issue
The service is caught between people who want an answer 24/7, in several languages, and reception teams under pressure who repeat the same information all day long. Yet most consumer chatbots amount to routing residents' questions and personal situations through a third-party service, often hosted outside Europe and subject to the Cloud Act — incompatible with the equal treatment and confidentiality a public service owes.
✓ Our answer
An information agent is only of interest to the public service if it is sovereign and reliable by design. Local inference or an isolated resource hosted in France, answers anchored in a validated document base with cited sources, human oversight and routing to an officer as soon as a case is uncertain: the availability gained is never paid for in sovereignty lost. The aim is not to replace the front-desk officer, but to absorb repetitive questions so as to give back time for the situations that deserve it.
Public data and equal treatment: sovereignty & compliance
A public service answers every member of the public, with the same requirement of continuity and confidentiality. Here is how the architecture of our agents guarantees it, channel by channel.
Local inference
The agent can run on a machine belonging to the authority: no citizen question leaves the network, nothing passes through a cloud.
Hosting in France
Otherwise, a dedicated and isolated resource, hosted in France under French law — your data: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
Exposure of the public's questions to the Cloud Act and FISA 702 is reduced by design; location alone does not guarantee immunity.
Sourced answers
Every answer rests on a validated document base and cites its sources: no invention, and uncertain cases go to a person.
Encryption & controlled access
Encryption in transit and at rest, role-based access (RBAC), strong authentication and logging.
AI Act: governed deployment
An agent strictly in support: the user is told they are talking to an AI, and no decision about their rights is automated.
What depends on the architecture chosen These points are not general guarantees: they are settled deployment by deployment, in the quotation.
- The applicable location is that of the architecture set out in the quotation and verified before commissioning.
- Local execution is announced only for the configuration explicitly described and accepted in the quotation.
- The applicable isolation depends on the deployment mode set out in the quotation; no dedicated isolation is presumed.
- The encryption mechanisms in transit and at rest, their components and key management are those documented for the architecture chosen.
- Roles and permissions are configured and accepted for the identities and systems actually connected.
- The events logged, their content, their retention period and who may access them are defined for the deployment chosen.
See the agent at work
4 real situations, taken from those that come up most often. Pick one: the exchange unfolds as it would in your organisation.
A scripted demonstration. These exchanges show how the agent behaves — its sources, its refusals, what it leaves to your teams. Nothing is sent from this page, no model is queried here, and the matters named are fictional. That is precisely what we promise your data.
The behaviours shown here — monitoring, automation rules, routing and reminders — are configured with you during deployment, from your tools, your rules and your thresholds.
The architecture points named in these exchanges — location, local execution, isolation, encryption, role-based access, logging — are not a guarantee attached to the demonstration: they are those of the architecture set out in your quotation, and verified before commissioning.
· One answer in the knowledge base points to a form withdrawn from the site on 12/06. It has been served 34 times since.
· One item contradicts a municipal by-law. The base says the market is on Wednesday; the by-law of 04/07 moved it to Thursday.
· 41 residents abandoned their application at the same step of the support-payment form, over three weeks.
· "I do not have that information" answers have doubled on one topic alone: façade renovation grants. morning-watch_4-flags.pdf4 flags · source of each
⛓ Source · knowledge base, published by-laws, online journey analytics
What is established: the by-law of 04/07/2026 moves the weekly market from Wednesday to Thursday with effect from 01/09. Your knowledge-base entry "weekly market" is dated 11/03/2025 and still says Wednesday.
What that produced: the entry has been served 62 times since 04/07, 11 of them since 01/09.
What I have prepared: the corrected entry, the by-law cited as its reference with its date, the effective date at the top, and the message to the 11 residents who received the old answer since 1 September — those can still be caught, and the message is written for each of them.
What is yours to do: approve the update. A knowledge base speaks for the authority — that is why updating it is approved and dated, and it is what makes it possible, six months later, to say who answered what and on what basis. Two minutes of reading, and the next 62 answers are right.
And I looked further, since the mechanism is the same: across your 214 entries, 9 cite a text that a later by-law has amended. They are listed, by by-law date. market-entry_to-correct.pdf1 entry · 11 residents reachable
⛓ Source · by-law of 04/07/2026, entry of 11/03/2025, served-answer log
Routing follows your organisation: the stale entry and the contradiction to whoever owns the base; the 41 abandonments to the department that owns the procedure, because that is a form problem and not an information one; the unanswered topic to the department concerned, with the number of requests.
With a chase: 48 h on the contradiction — every day adds badly informed residents; 7 days on the rest. Then a weekly summary to the chief executive: by subject, never by officer.
What this morning already gives back to the service: an entry that contradicts a by-law and that has been served 62 times since 4 July, 11 of them since 1 September — those eleven residents can still be caught, and the list is made. From tomorrow: 34 answers that stop pointing to a form withdrawn on 12 June, a topic I knew nothing about — façade renovation grants — written down once and for all, and 41 abandoned applications dealt with where they happen, in the form, not at the counter. The time given back is your officers': the most repeated questions get a sourced, dated answer at any hour, and the counter officer keeps the situations that need an officer. The access is yours: base, by-laws and application journeys opened by role, logged, withdrawn with a word; residents' data stays inside your administration, under French law. Correcting and publishing commits the authority's word: that is not a prohibition, it is a mandate — written, limited to the entries you name, dated, withdrawn with a word. The "weekly market" entry is already drafted, with the by-law of 4 July as its reference: you approve it, it is online in minutes, dated and in the name of whoever approved it; without a mandate it waits for your review and I chase you at 48 h. The next step is ready: the list of the 11 residents to call back, and the list of entries older than the last act that concerns them.
✎ Proposal · watch and chases to be configured — you set the thresholds
What your base says: recycling collected on Fridays, entry "waste collection", updated 02/05/2026. 15 August 2026 falls on a Saturday.
What your public-holiday calendar says: collection moves to Saturday when the collection day is a holiday — but 15 August is not the collection day this year.
So: collection takes place on Friday the 14th, as normal.
What I flag to the resident, because it is honest: your base says nothing about a possible shift caused by the holiday the day after. I answer what the base says, I name what it does not, and I offer the department's number.
⛓ Source · entry of 02/05/2026 + the authority's public-holiday calendar
What the resident receives: "Recycling is collected on Fridays, so Friday 14 August." Then, underneath: "We have no information on a shift linked to the holiday on the 15th. If in doubt, the waste department can be reached on …"
Why the second sentence matters: a resident told "Friday" without qualification who finds their bins full on Saturday does not come back to ask. They conclude the authority's information is worthless.
And an observation: 17 entries in your base state a collection or opening day without addressing public holidays. That is not a drafting lapse, it is an entry template with no field for it.
I can flag every entry in that state, before the next run of holidays. I rewrite none of them. entries_17-without-holidays.pdf17 entries · same missing field
✎ Proposal · 17 entries with no holiday handling — rewriting stays with the department
Her question: conditions and documents for the travel allowance. Answered from your entry of 18/04/2026.
What her answers reveal, and which I flag to her without asserting anything: she mentions a child at secondary school and a household income she places below the threshold in your "school meals allowance" entry. Two of the authority's schemes could concern her.
The wording used: "From what you tell me, two other allowances exist and could concern you. I cannot verify your eligibility: here are the conditions, and the department that establishes it."
What I give her in the same message: the conditions of both schemes, the documents to gather, the department responsible and its opening hours — and, for one of the two, the fact that she has already filed two of the four documents required this year. She arrives at the counter with half the file already done.
What is left to the officer, and it is the resident's guarantee: establishing eligibility. It is established on documents, by a person, and opening the entitlement carries their name — that is what makes the decision reasoned and open to challenge. An entitlement opened by a machine would be neither. entitlements-flagged_2-schemes.pdf2 schemes · conditions, no conclusion
⛓ Source · entries for the 3 schemes, facts stated during the call
Two figures that belong together: of those 214, 61 residents asked for details of the scheme flagged, and 23 went on to apply. The other 153 did not follow up, which is their right.
Why those 153 are not contacted again, and it is a rule rather than an oversight: a flagged entitlement is mentioned once, in the exchange where it arises. Repeated, it becomes cold outreach — and on social circumstances, that is intrusive. A resident's silence is an answer.
What I did with the 153, without naming any of them: I looked at where the exchange stopped. On 88 of them, the conversation breaks off at the line listing the documents required — that is not a refusal, it is a wall. The same information would fit in three lines rather than eleven, and I am proposing the rewrite.
A proposal, if you approve it: a quarterly report to the social action centre, by scheme and never by resident: how often flagged, how many asked for details, how many applied, and at which line the exchange stops. That is the measure of unclaimed entitlements you do not have today, and it is built without any nominative file. unclaimed-entitlements_quarter.pdf214 flags · 3 schemes
✎ Proposal · report by scheme — no resident is contacted again
What the base says, served on chat and by phone: online pre-application recommended, then an appointment at the town hall with the listed documents.
What the front desk does, per your internal instruction of 09/01/2026: officers accept files without an online pre-application for people over 75 and for those referred by the social action centre.
That flexibility appears nowhere in the base. An 80-year-old told by chat that a pre-application is required gives up, or arrives with needless worry. Over the quarter, 31 exchanges involved people stating they were over 75.
What I have drafted: the paragraph carrying the exception, in the exact wording of your 09/01 instruction, ready to be added to the entry — and its short version for the telephone answer.
What is yours to do, and it is no formality: deciding to publish it. An internal instruction is not public information until you decide it is, and publishing it changes its status: what was a front-desk flexibility becomes an enforceable rule the counter will have to keep. That is exactly the question to put, and it is better put now than in front of an unhappy resident. gaps_base-and-front-desk.pdf4 gaps · internal instruction behind each
⛓ Source · knowledge base, internal instruction of 09/01/2026
· Identity cards — online pre-application, exception for the over-75s and social-centre referrals.
· Nursery registration — the base says applications from 1 to 31 January; the front desk accepts late files in case of a job relocation.
· Village hall — the base asks for two months' notice; the front desk goes down to a fortnight out of season.
· Hosting certificate — the base does not mention the appointment option the front desk offers as a matter of course.
All four point the same way: the front desk is more flexible than the authority announces. That is good news about the service and a problem of equal treatment — whoever comes in person benefits, whoever phones does not.
What I can do, on your approval: flag every internal instruction that contradicts a published entry, to whoever owns the base and to whoever issued the instruction. The decision whether to publish the exception stays entirely open: some have good reasons not to be.
✎ Support · 4 gaps documented — whether to publish stays with the authority
What I measured across your 214 standard answers: the average sentence runs to 34 words; 61 answers use at least one term that exists only inside the administration; 18 answers exceed 400 words for a question that calls for 60.
The plain-language version, and what it keeps: one idea per sentence, the answer in the first two lines, the procedure in numbered steps, and the administrative term kept and then explained rather than deleted — because the user will meet it again on the form. The answer on façade renovation grants goes from 412 to 118 words and keeps all 9 pieces of information from the original, which I set side by side for checking.
The draft transcription under the FALC rules: I prepare it — one idea per sentence, a pictogram per step, dates written out in full — and it remains a draft subject to human validation until the review involving the people concerned has taken place. Without that step it is not FALC and we do not call it that: calling an unreviewed text FALC would advertise an accessibility that does not exist.
What that gives on every channel: the plain-language version is the same on chat, on the telephone and at the counter — your front-desk officer reads on screen the text the user reads at home, which puts an end to the two contradictory wordings. Over three months of trial, requests for re-explanation fell from 23 % to 9 % of exchanges, and the façade renovation topic stopped generating answers with no information.
The figure that does not flatter me: of my first 80 plain-language versions, 9 had dropped the name of the competent department — 11 % — because I was reading it as a footer signature. Nine people wrote to the wrong department and waited three more weeks. I set six unalterable mentions — competent department, appeal deadline and route, closing date, file reference, amount, signatory — and I block any version that loses one: over the next 300 versions, no loss.
⛓ Sourced · 214 standard answers measured, 412 words down to 118 keeping 9 pieces of information, re-explanations from 23 % to 9 %
Your case is not here? That is exactly what a 15-minute conversation is for. Book the free audit →
The facets of administrative information
Each facet corresponds to an agent we deploy. All of them work in support, under the supervision of your officers.
Answers on procedures, sources cited
Answering questions on everyday procedures (civil registry, schools, planning, social) from a validated base, with sources to back them.
24/7 and multichannel
The same agent on the web and at the counter, to answer at any hour.
Support for the front-desk officer
Prompting the officer at the counter with the right answer and the expected documents, for consistent and faster information.
Plain language & easy-read draft
Preparation of a plain-language version and of a draft easy-read (FALC) transcription, submitted for human validation.
Tier-1 customer support agent (website)
FAQ chatbot, human escalation, messaging channels included at no extra cost.
On quote View the agent page →Translating answers and letters
What this agent prepares — the sourced answer and its plain-language version — is handed to the Accessibility and inclusion agent, which produces the translation and the draft easy-read (FALC) transcription. Human validation stays with the public officer before anything reaches the user.
On quote View the agent page →Need to go further?
These agents handle a different business process, with their own owner and their own price. They are added to this one.
Help with form filling & non-take-up
Supporting users through an online procedure, step by step, to reduce drop-outs and the non-take-up of rights.
Support with online procedures from 780 € incl. VAT / month Help with procedures →Appointment booking
What this agent prepares — the user's request and the department concerned — is handed to the Appointment booking and management agent, which books, moves and confirms the slot. The scope of this page stops at information; human validation stays with the public officer.
Booking and managing appointments from 810 € incl. VAT / month Discover the agent →Multi-agency France Services reception
Strictly in support (administrative). Information and routing.
Multi-agency France Services reception from 765 € incl. VAT / month Discover the agent →Information and cataloguing support in libraries
Strictly in support (administrative and documentary).
Information and cataloguing support in libraries from 840 € incl. VAT / month Discover the agent →In 15 minutes we identify the agent that will give your staff the most time back — without oversizing the project.
How much time can a service win back?
By absorbing repetitive questions about procedures, a reception service can aim for a sharp reduction in the volume handled by officers — reinvested in supporting complex situations and tackling non-take-up.
The stages of your AI agent project
Audit & scoping
15 minutes to target the use case with the best return.
Quote or direct sign-up
A catalogue offer is bought online; a specific need gets a costed quote.
Design
We design the agent and its guardrails.
Integration & testing
We connect your tools to the agent, which is itself hosted in France.
Rollout
Going live and training your team.
Operation
Continuous supervision and improvement.
Three options, one agent
An administrative information agent (web, counter, phone), installed and operated for you. Choose according to how you are organised. Prices exclude VAT — annual subscription, the time it takes for the gains to settle in.
Setup + controlled subscription
- Installation, configuration and training for your teams
- Operation, human oversight, updates and support
- Sovereign hosting in France, a dedicated and isolated resource
All inclusive, no setup fee
- Setup included (installation, configuration, training)
- Operation, human oversight, updates and support
- Sovereign hosting in France, managed end to end
On site, you own it
- Hardware installed on your premises (you own it)
- French / European AI models run locally
- Secure remote maintenance (Pro support included)
Four guarantees that matter to a public service
Related resources
Your questions, our answers
Can the answers be wrong?
Do we have to change our website or our tools?
How is the AI Act taken into account?
Is the public's data protected?
Does the agent answer outside opening hours?
What happens when the agent does not know the answer?
Does the agent state that it is an artificial intelligence?
How long does it take to deploy an agent?
Which tools can users use to reach the agent?
Other solutions for relations with the public
Let us estimate the potential in your service
15 minutes to identify the most requested procedures — hosted in France, supervised, with no commitment.