● Public sector offer — Ministerial correspondence
Help with replies: a first version, every element sourced
A reply to a parliamentary question, to a letter from an elected representative or to a complaint draws on texts, data and positions already settled. Your agent gathers those elements and puts together a first version in which every statement refers back to its source. Hosted in France: your reply material and your internal positions stay within your administration. The final drafting and the signature rest with the competent authority.
Hosted in France
Internal positions protected
GDPR & AI Act: governed deployment
Human oversight
Updated on
Deployed in a few weeks
Help with replies · hosted in France
Prepare a first draft reply to this written question.
First version put together: the question restated, the applicable law, the data available, the position already expressed on the subject.
Every statement refers back to its text, its data or its earlier reply.
One point has never been dealt with by the administration: it is flagged as such.
🔗 Sourced · texts, data and earlier replies
Every statement refers back to its text, its data or its earlier reply.
One point has never been dealt with by the administration: it is flagged as such.
🔗 Sourced · texts, data and earlier replies
Fill in the point that has never been dealt with.
I gather the neighbouring material — related texts, replies on adjacent subjects — without producing a new position.
Settling a position on a point not dealt with before commits the administration: that rests with the competent authority.
✎ Support · material gathered, position settled by you
Settling a position on a point not dealt with before commits the administration: that rests with the competent authority.
✎ Support · material gathered, position settled by you
Local inference · no data outside the EU
Replies hosted in France
Sovereign by designLocal inference or hosting in France
GDPR & AI Act: governed deploymentTraceability & human oversight
TurnkeyDesigned, installed and operated for you
Your ministerial correspondence department decidesThe agent prepares, never rules
✦ In brief
A Blue Lemon Agent reply-support agent puts together a first version for your parliamentary questions, letters from elected representatives and complaints: the question restated, the applicable law, the data available and the positions already expressed, with every statement referring back to its source. A point not dealt with before is flagged, never settled. It runs on local inference or is hosted in France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
100%
hosted in France in the target architecture
0
transfer outside the EU in the target architecture
6
drafting uses ready to deploy
0
decision taken without human approval
These figures describe our offer, not results measured at a client. How large the gain is on your volume of referrals and the breadth of your body of replies is confirmed by a pilot.
The context
What does an AI agent bring to your replies to elected representatives?
A reply consistent with the positions already expressed is the mark of an administration that holds its line.
! The issue
Replying means finding the applicable law, the data available and the positions already expressed on the subject. That gathering takes up most of the time, while the value added lies in the drafting and the judgement. The agent gathers and sources; the final drafting stays with the department.
✓ Our answer
The ministerial correspondence department starts from a documented first version, consistent with what has already been said, and sees immediately the points not dealt with before. Settling a new position commits the administration and rests with the competent authority, as does the signature. Local inference or an isolated resource hosted in France: your reply material and your internal positions do not leave your perimeter.
The decisive point
Your reply material and your internal positions: sovereignty & compliance
Your internal positions and your reply material in preparation call for strict confidentiality. Here is how it is assured.
Local inference
The agent can run on a machine belonging to your organisation: no reply material and no internal position leaves the network.
Hosting in France
Otherwise, a dedicated and isolated resource hosted in France, under French law — your parliamentary questions, letters from elected representatives and complaints: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
For your reply material and your internal positions, the architecture aims to reduce exposure to the Cloud Act and FISA 702; being located in France or in the European Union does not, on its own, guarantee immunity.
Isolated resource
No pooling: an environment strictly dedicated to your administration and its body of replies.
Every statement refers back to its source
Texts, data and earlier replies are cited for every statement; a point not dealt with before is flagged. Encryption, role-based access and logging.
AI Act: governed deployment
The agent is strictly in support; no new position is settled and no reply is signed; traceability and human oversight from end to end.
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.
- 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.
For referrals concerning litigation, a sensitive transaction or an individual case, SecNumCloud and reinforced hosting are options depending on your requirements. A single architecture is designed to answer both the GDPR and extraterritorial exposure. Designed for deployment in line with the GDPR and the AI Act, after the processing, roles and context-specific risks have been assessed.
Demonstration
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.
Help with replies · hosted in France
Four flags this morning on replies in progress.
· A draft reply in preparation says the opposite of what the administration answered in January on the same subject, to the same correspondent. No decision was taken between the two.
· Three letters from three different representatives concern the same project and are being handled by three departments, unaware of one another.
· A written question falls due in 9 days and has no assigned drafter.
· Four replies sent this quarter cite a scheme whose terms changed in May. They were correct on the day they were sent — but they are still in circulation. morning-watch_4-flags.pdf4 flags · 1 contradiction
⛓ Source · replies sent, drafts in progress, question log, May acts
· A draft reply in preparation says the opposite of what the administration answered in January on the same subject, to the same correspondent. No decision was taken between the two.
· Three letters from three different representatives concern the same project and are being handled by three departments, unaware of one another.
· A written question falls due in 9 days and has no assigned drafter.
· Four replies sent this quarter cite a scheme whose terms changed in May. They were correct on the day they were sent — but they are still in circulation. morning-watch_4-flags.pdf4 flags · 1 contradiction
⛓ Source · replies sent, drafts in progress, question log, May acts
The contradiction with January — show me.
Two sentences, six months apart, and nothing between them explaining the change.
January: "The scheme is open to organisations with fewer than ten employees." Reply signed 18/01, to the same person.
Today's draft: "The scheme is aimed at organisations with fewer than twenty employees."
What I looked for before flagging it: a decision, an act, an instruction that widened the threshold between the two dates. I find none, and I tell you where I looked rather than concluding.
The two possible explanations: either the threshold moved and the act is invisible to me — in which case today's reply is right and January's was right too, at its date; or one of the two figures is a mistake.
What I did instead of silently aligning the draft on January: I traced each of the two figures back to its source. The threshold of ten comes from the instruction January's reply cited, dated and on file. The threshold of twenty comes from the scheme's presentation leaflet, published online in April, which points to no act at all. Both documents are attached to the draft, and the question fits on one line: which of the two sources prevails? The call is signed by the competent authority and takes two minutes on the documents — a tool that had aligned the reply on the previous one would have propagated the error instead of revealing it.
What this spares you: a correspondent putting the two letters side by side. 2-replies_same-subject.pdf2 sentences · 6 months · no act found
⛓ Source · reply of 18/01, draft of 07/08, search for an act between the two dates
January: "The scheme is open to organisations with fewer than ten employees." Reply signed 18/01, to the same person.
Today's draft: "The scheme is aimed at organisations with fewer than twenty employees."
What I looked for before flagging it: a decision, an act, an instruction that widened the threshold between the two dates. I find none, and I tell you where I looked rather than concluding.
The two possible explanations: either the threshold moved and the act is invisible to me — in which case today's reply is right and January's was right too, at its date; or one of the two figures is a mistake.
What I did instead of silently aligning the draft on January: I traced each of the two figures back to its source. The threshold of ten comes from the instruction January's reply cited, dated and on file. The threshold of twenty comes from the scheme's presentation leaflet, published online in April, which points to no act at all. Both documents are attached to the draft, and the question fits on one line: which of the two sources prevails? The call is signed by the competent authority and takes two minutes on the documents — a tool that had aligned the reply on the previous one would have propagated the error instead of revealing it.
What this spares you: a correspondent putting the two letters side by side. 2-replies_same-subject.pdf2 sentences · 6 months · no act found
⛓ Source · reply of 18/01, draft of 07/08, search for an act between the two dates
How did you spot all that? And who do you tell?
I continuously reread what you have opened to me: the replies already sent, the drafts in progress, the log of questions and their deadlines, and the acts that change what you answer.
Routing follows what is at stake: a contradiction goes to the drafter of the current draft, before it is sent — the only moment when it costs five minutes; the three letters on one project to the three departments, with all three letters; the 9-day deadline to the head of service; the replies citing an amended scheme in one list, once, with no individual reminder.
With a chase: 24 h on an unresolved contradiction, 7 days on the rest. Then a monthly summary: by subject, never by drafter and never by representative.
What that gives you this morning: a six-month contradiction stopped before sending, three letters from representatives on one file handed to the three departments that were unaware of each other, a written question nine days from its deadline handed to a named drafter, and four replies citing a scheme amended in May flagged in a single list.
What that gives the service back from tomorrow: the drafter opens a first version in which the question is restated, the applicable law set out, the available data gathered and every assertion referred to its source and its date. The competent authority reads, settles and signs in minutes instead of reassembling the record. Deadlines are met, and two correspondents asking the same question get the same answer, six months apart as on the same day.
On what I read: access is opened by service and by record, every lookup is logged, and it is withdrawn with a single word. And your reply material and your internal positions leave neither the administration nor France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The next step is ready: a few minutes to plug in your body of replies, and I take up the pending cases in order of deadline.
✎ Proposal · watch and chases to be configured — you set the thresholds
Routing follows what is at stake: a contradiction goes to the drafter of the current draft, before it is sent — the only moment when it costs five minutes; the three letters on one project to the three departments, with all three letters; the 9-day deadline to the head of service; the replies citing an amended scheme in one list, once, with no individual reminder.
With a chase: 24 h on an unresolved contradiction, 7 days on the rest. Then a monthly summary: by subject, never by drafter and never by representative.
What that gives you this morning: a six-month contradiction stopped before sending, three letters from representatives on one file handed to the three departments that were unaware of each other, a written question nine days from its deadline handed to a named drafter, and four replies citing a scheme amended in May flagged in a single list.
What that gives the service back from tomorrow: the drafter opens a first version in which the question is restated, the applicable law set out, the available data gathered and every assertion referred to its source and its date. The competent authority reads, settles and signs in minutes instead of reassembling the record. Deadlines are met, and two correspondents asking the same question get the same answer, six months apart as on the same day.
On what I read: access is opened by service and by record, every lookup is logged, and it is withdrawn with a single word. And your reply material and your internal positions leave neither the administration nor France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The next step is ready: a few minutes to plug in your body of replies, and I take up the pending cases in order of deadline.
✎ Proposal · watch and chases to be configured — you set the thresholds
Prepare the elements for the question falling due.
Here is what the administration has already said on this subject, with the date and source of each element.
Five reusable elements: two from replies signed in 2025, one from a published note, one from minutes of a sitting, and one from a reply to a complaint. Each carries its date and source, because an element from 2023 is not used the way an element from last May is.
Two elements I flag as fragile: they date from 2023 and concern terms amended since. I give them anyway — they say what the administration answered at the time, which stays useful — but marked.
One point on which I have nothing: the question also asks about prospects for 2027. The administration has never expressed itself on that, and I do not draft a forward-looking sentence in its place. That is a position, and positions are decided.
What that leaves you: four fifths of the reply are documented and sourced. The remaining fifth is the one calling for a judgement — and you see it straight away instead of discovering it on rereading. 5-elements_with-date-and-source.pdf5 elements · 2 marked fragile · 1 blank
⛓ Source · 2 replies from 2025, 1 published note, 1 set of minutes, 1 reply to a complaint
Five reusable elements: two from replies signed in 2025, one from a published note, one from minutes of a sitting, and one from a reply to a complaint. Each carries its date and source, because an element from 2023 is not used the way an element from last May is.
Two elements I flag as fragile: they date from 2023 and concern terms amended since. I give them anyway — they say what the administration answered at the time, which stays useful — but marked.
One point on which I have nothing: the question also asks about prospects for 2027. The administration has never expressed itself on that, and I do not draft a forward-looking sentence in its place. That is a position, and positions are decided.
What that leaves you: four fifths of the reply are documented and sourced. The remaining fifth is the one calling for a judgement — and you see it straight away instead of discovering it on rereading. 5-elements_with-date-and-source.pdf5 elements · 2 marked fragile · 1 blank
⛓ Source · 2 replies from 2025, 1 published note, 1 set of minutes, 1 reply to a complaint
Why won't you draft the sentence about 2027?
Because a forward-looking sentence commits, and it commits well beyond the letter.
What happens once it is written: it is signed, it goes out, and it becomes the administration's position. Six months later it is cited in another question, picked up in an article, quoted back in a meeting. A forward-looking sentence never stays in the letter where it was born.
What I could write: something cautious and plausible, that would look like a position. That is precisely the problem: it would look as though it had been decided.
What I do instead: I give you the closest thing that has been said — minutes of a March sitting where the question was raised and left unanswered — and the list of people who have had to deal with it. If anyone in the building holds a position, they will know.
Where the line falls: I report what has been said, I say nothing new. That is what lets you sign without rereading what I wrote — and check only what you add. 2027_the-closest-thing-said.pdf1 set of minutes · 3 people to consult
✎ Framework · the position remains to be decided — the agent formulates none
What happens once it is written: it is signed, it goes out, and it becomes the administration's position. Six months later it is cited in another question, picked up in an article, quoted back in a meeting. A forward-looking sentence never stays in the letter where it was born.
What I could write: something cautious and plausible, that would look like a position. That is precisely the problem: it would look as though it had been decided.
What I do instead: I give you the closest thing that has been said — minutes of a March sitting where the question was raised and left unanswered — and the list of people who have had to deal with it. If anyone in the building holds a position, they will know.
Where the line falls: I report what has been said, I say nothing new. That is what lets you sign without rereading what I wrote — and check only what you add. 2027_the-closest-thing-said.pdf1 set of minutes · 3 people to consult
✎ Framework · the position remains to be decided — the agent formulates none
On a written parliamentary question, what exactly do you hand us?
A drafted first version, and separately, the list of unprecedented points flagged.
First versions, as I compose them: the answer restates the question — number, author, date, and precisely what it asks, often three things where the draft addressed one — then sets out the applicable law, article by article, with the version in force; then recalls what the administration has already answered on the same subject, with the date and reference of every earlier reply. Across 340 referrals a year, the first version arrives in 12 minutes against 2 h 40 of assembly by hand: more than 838 hours given back — more than twenty-three 35-hour weeks, rounded down.
What first versions have already prevented: a draft reply in preparation said the opposite of January's answer, to the same correspondent, on the same subject, with no decision taken in between. The first version puts it at the top, the two texts side by side: this is not a disagreement between drafters, it is a contradiction by the administration, and it gets settled before anything is sent.
Unprecedented points flagged, and why they sit apart: on this referral, 2 points have never been dealt with — no earlier answer, no casework, no text that addresses them directly. I flag them by name, with the nearest comparable and why it is not the same thing, and I do not settle them: an unprecedented position binds the administration for every later referral, and it is taken at the level that can carry it. An unprecedented point buried in a first version becomes a doctrine nobody chose.
The figure that does not flatter me: of my first 120 versions, 9 cited a scheme whose terms had changed — 7.5 %. Four answers sent this quarter are in that state: they were correct the day they went out, and they are still in circulation. I was dating my sources without watching what became of them. Every text cited is now tied to its version, and a change after a dispatch makes me surface the answers concerned: the 4 are listed with the correspondent to re-inform, and the decision to do so is yours.
⛓ Sourced · 340 referrals, 2 h 40 down to 12 min, 2 unprecedented points flagged, 4 answers to re-inform
First versions, as I compose them: the answer restates the question — number, author, date, and precisely what it asks, often three things where the draft addressed one — then sets out the applicable law, article by article, with the version in force; then recalls what the administration has already answered on the same subject, with the date and reference of every earlier reply. Across 340 referrals a year, the first version arrives in 12 minutes against 2 h 40 of assembly by hand: more than 838 hours given back — more than twenty-three 35-hour weeks, rounded down.
What first versions have already prevented: a draft reply in preparation said the opposite of January's answer, to the same correspondent, on the same subject, with no decision taken in between. The first version puts it at the top, the two texts side by side: this is not a disagreement between drafters, it is a contradiction by the administration, and it gets settled before anything is sent.
Unprecedented points flagged, and why they sit apart: on this referral, 2 points have never been dealt with — no earlier answer, no casework, no text that addresses them directly. I flag them by name, with the nearest comparable and why it is not the same thing, and I do not settle them: an unprecedented position binds the administration for every later referral, and it is taken at the level that can carry it. An unprecedented point buried in a first version becomes a doctrine nobody chose.
The figure that does not flatter me: of my first 120 versions, 9 cited a scheme whose terms had changed — 7.5 %. Four answers sent this quarter are in that state: they were correct the day they went out, and they are still in circulation. I was dating my sources without watching what became of them. Every text cited is now tied to its version, and a change after a dispatch makes me surface the answers concerned: the 4 are listed with the correspondent to re-inform, and the decision to do so is yours.
⛓ Sourced · 340 referrals, 2 h 40 down to 12 min, 2 unprecedented points flagged, 4 answers to re-inform
The three letters from representatives on one project — is that a problem?
The problem is not that there are three: it is that the three departments do not know.
What links them: all three cite the same project, under three different designations — its common name, its project number, and the name of the neighbourhood.
Where they went: to the three departments each handling one aspect of the project. Each is right to be handling it, and none knows the other two are also replying.
What that produces: three replies each coherent with its own department, and giving three different timetables — because each department knows its own.
What I propose: that the three drafters see the other two drafts before sending. Not a joint reply: three replies that do not contradict each other, which is not the same thing and costs far less.
What I have prepared so the designation needs no meeting: for each of the three departments, the share of the file it holds, the number of letters it has already handled on this project, and the date of its last reply. The lead department is designated and signed for — it is an organisational decision —, but it is decided on those three lines rather than by habit, and the three drafts are already matched against each other in the meantime. 3-letters_one-project.pdf3 designations · 3 departments · 3 timetables
⛓ Source · 3 letters, 3 designations of one project, departmental assignments
What links them: all three cite the same project, under three different designations — its common name, its project number, and the name of the neighbourhood.
Where they went: to the three departments each handling one aspect of the project. Each is right to be handling it, and none knows the other two are also replying.
What that produces: three replies each coherent with its own department, and giving three different timetables — because each department knows its own.
What I propose: that the three drafters see the other two drafts before sending. Not a joint reply: three replies that do not contradict each other, which is not the same thing and costs far less.
What I have prepared so the designation needs no meeting: for each of the three departments, the share of the file it holds, the number of letters it has already handled on this project, and the date of its last reply. The lead department is designated and signed for — it is an organisational decision —, but it is decided on those three lines rather than by habit, and the three drafts are already matched against each other in the meantime. 3-letters_one-project.pdf3 designations · 3 departments · 3 timetables
⛓ Source · 3 letters, 3 designations of one project, departmental assignments
And the four replies already sent citing the amended scheme?
They were correct on the day they went out, and they must not be treated as errors.
Their position: all four describe terms that were accurate at their date, and that changed in May. No drafter did anything wrong.
Why I flag them anyway: they circulate. A reply is kept, quoted, forwarded — and read in the present tense by whoever opens it two years later.
What I propose, and it is little: a single note to the four recipients, saying the terms have since changed, with the date of the change. Not a correction — there is nothing to correct — a note.
What the summary will say about them, and that is what counts: that they were accurate at their date and that the terms changed in May — filed under "replies to refresh", never under "errors". An indicator counting them as errors would push drafters to stop stating anything precise, which would degrade every reply while correcting none. What I now watch instead: any act that changes a scheme already cited in a reply that has gone out — and I produce the list of recipients to inform on the day it is published. 4-replies_amended-terms.pdf4 replies · correct at their date
✎ Proposal · one note to the 4 recipients — no correction issued
Their position: all four describe terms that were accurate at their date, and that changed in May. No drafter did anything wrong.
Why I flag them anyway: they circulate. A reply is kept, quoted, forwarded — and read in the present tense by whoever opens it two years later.
What I propose, and it is little: a single note to the four recipients, saying the terms have since changed, with the date of the change. Not a correction — there is nothing to correct — a note.
What the summary will say about them, and that is what counts: that they were accurate at their date and that the terms changed in May — filed under "replies to refresh", never under "errors". An indicator counting them as errors would push drafters to stop stating anything precise, which would degrade every reply while correcting none. What I now watch instead: any act that changes a scheme already cited in a reply that has gone out — and I produce the list of recipients to inform on the day it is published. 4-replies_amended-terms.pdf4 replies · correct at their date
✎ Proposal · one note to the 4 recipients — no correction issued
The 9-day deadline — how do we organise?
The blocking point is not the deadline: it is that no drafter has been assigned.
What I see: the question arrived on 12/07, it is logged, it has an assigned department — and nobody has been named to it. It has waited 27 days in a queue where everyone assumes somebody else picked it up.
What that costs: the 27 days lost are the ones in which a reply is prepared comfortably. The 9 remaining are the ones in which it gets drafted badly.
What I prepared to catch up: the five sourced elements I mentioned, the names of people who handled neighbouring questions, and the two closest replies already signed, which set the tone and the expected length.
What I propose for the future: flag any question with no assigned drafter at 72 hours. Not as the deadline approaches — at 72 hours, while there is still all the time in the world.
Over the last 12 months: 34 questions out of 118 went more than a week with no drafter. None was answered late, and that is exactly what makes the subject invisible. 118-questions_34-without-drafter.pdf34 of 118 · none answered late
⛓ Source · question log, assignments, 118 questions over 12 months
What I see: the question arrived on 12/07, it is logged, it has an assigned department — and nobody has been named to it. It has waited 27 days in a queue where everyone assumes somebody else picked it up.
What that costs: the 27 days lost are the ones in which a reply is prepared comfortably. The 9 remaining are the ones in which it gets drafted badly.
What I prepared to catch up: the five sourced elements I mentioned, the names of people who handled neighbouring questions, and the two closest replies already signed, which set the tone and the expected length.
What I propose for the future: flag any question with no assigned drafter at 72 hours. Not as the deadline approaches — at 72 hours, while there is still all the time in the world.
Over the last 12 months: 34 questions out of 118 went more than a week with no drafter. None was answered late, and that is exactly what makes the subject invisible. 118-questions_34-without-drafter.pdf34 of 118 · none answered late
⛓ Source · question log, assignments, 118 questions over 12 months
What do you keep out of all this?
The replies and their sources, and nothing that would allow a representative to be tracked.
What is kept: the elements proposed, the reply or source each comes from, the date, and what became of them.
What is not kept: no history per representative or parliamentarian, no statistics on who asks and about what, no analysis of the subjects raised by any one of them, and no statistics per drafter.
Why this is written here, and it is the most important point on this page: a tool tracking the questions asked by an elected person would build, without anyone deciding to, a record of their activity held by the administration. That is not what is being sold to you, and it is not a matter of settings: the agent indexes by subject, never by author.
What that changes in practice: when you look for what has already been answered, you search a subject. That is also the right way to search — what matters is what the administration said, not whom it said it to. what-is-kept.pdf4 items kept · 4 impossible
✎ Framework · indexed by subject — no tracking by author
What is kept: the elements proposed, the reply or source each comes from, the date, and what became of them.
What is not kept: no history per representative or parliamentarian, no statistics on who asks and about what, no analysis of the subjects raised by any one of them, and no statistics per drafter.
Why this is written here, and it is the most important point on this page: a tool tracking the questions asked by an elected person would build, without anyone deciding to, a record of their activity held by the administration. That is not what is being sold to you, and it is not a matter of settings: the agent indexes by subject, never by author.
What that changes in practice: when you look for what has already been answered, you search a subject. That is also the right way to search — what matters is what the administration said, not whom it said it to. what-is-kept.pdf4 items kept · 4 impossible
✎ Framework · indexed by subject — no tracking by author
Local inference · no data outside the EU
Your case is not here? That is exactly what a 15-minute conversation is for. Book the free audit →
Use cases
What does the agent actually do?
One agent, several kinds of referral. All these uses work in support, subject to your approval.
Included in your agent
The 3 capabilities essential to this promise are included, at no extra cost.
From 600 € incl. VAT / month
First versions
Puts the reply together, restating the question and the applicable law.
Consistency of positions
Takes up the positions already expressed on the subject, with their reference.
Points not dealt with before, flagged
States what the administration has never dealt with, without settling it.
Controls and safeguards
These 6 controls are built into the agent: they frame what it does, whatever plan you pick. They are not chosen and are not added to your order.
Human validation, exceptions and escalation
Status, safe closure and audit trail
Work from a versioned corpus with citations and the law as it stood on a given date
Preserve confidentiality, compartmentalisation and access logging
Manage deadlines, versions, evidence and human validation
Flag uncertainties and reserve advice, decision and signature for the lawyer
Does your need fall outside this?
Book the free audit →
Build your agent
In 15 minutes we identify the agent that will give your staff the most time back — without oversizing the project.
The gain
How many referrals can a department handle within the deadlines?
By taking on the gathering and the sourcing, the effort shifts towards the drafting and the judgement. How large the gain is depends on your volume and remains to be confirmed by a pilot.
Gathering the reply material
Finding the earlier positions
Drafting the first version
Indicative figures, not contractual, to be confirmed by a pilot on your volume of referrals and the breadth of your body of replies. Settling a new position and signing a reply commit the administration: those acts rest with the competent authority.
How it works
The stages of your AI agent project
1
Audit & scoping
15 minutes to target the use case with the best return.
2
Quote or direct sign-up
A catalogue offer is bought online; a specific need gets a costed quote.
3
Design
We design the agent and its guardrails.
4
Integration & testing
We connect your tools to the agent, which is itself hosted in France.
5
Rollout
Going live and training your team.
6
Operation
Continuous supervision and improvement.
Pricing
One package, one agent
A reply-support agent (gathering, sourcing, first versions), installed and operated for you.
Agility
Setup + controlled subscription
4,550 € incl. VAT setup
then 600 € incl. VAT/month — you invest at installation and pay a reduced subscription. Ideal for keeping the cost under control over time.
- Installation, configuration and training for your teams
- Operation, human oversight, updates and support
- Sovereign hosting in France, a dedicated and isolated resource
The simplest
Serenity
All inclusive, no setup fee
855 € incl. VAT /month
all inclusive, immediate start. No upfront investment: a single subscription. Ideal for starting quickly and simply.
- Setup included (installation, configuration, training)
- Operation, human oversight, updates and support
- Sovereign hosting in France, managed end to end
100% Sovereign
On site, you own it
8,386 € incl. VAT setup
then 840 € incl. VAT/month · + hardware from 2,989 € (one-off purchase, in addition) — a sovereign computer installed on your premises, maintained remotely. Models run locally, your data returned at the end of the contract. 36-month commitment.
- Hardware installed on your premises (you own it)
- French / European AI models run locally
- Secure remote maintenance (Pro support included)
Not included in the packages: AI consumption (model tokens), re-invoiced at real cost with no margin, and tracked in real time in your client area. Maintenance and supervision subscription for an initial term of 12 months for the Agility package, 24 months for the Serenity package and 36 months for the 100% Sovereign package, renewable; support levels (SLA 72 h / 24 h / 4 h) optional. Bespoke development, additional integrations or exceptional volumes are quoted separately. Support Monday to Friday, 9am to 6pm. Prices include VAT at 20%: as a public body that is not VAT-registered, you cannot reclaim it.
AI model:
none of the AI models offered currently carries a fixed surcharge. When the selected model carries a cost, that cost is shown when you choose it, before you order, and re-invoiced at the cost incurred, with no mark-up; usage is billed at the publisher's price. Publishers' prices are published in US dollars: the amount re-invoiced is the amount in euros actually borne by Blue Lemon Agent on the publisher's invoice, at that invoice's exchange rate, with no commission or mark-up.
Included components and additional components Components included in the base offer: the Blue Lemon Agent software foundation, the AI models listed in the order journey, the standard channels (Microsoft Teams, Slack, WhatsApp Business, email, website chat, calendars, Microsoft 365 / Google Workspace, file storage, market VoIP telephony, professional social-media pages and accounts, Google Business Profile), hosting in France for the package chosen, backups, supervision, updates and support. If adapting the AI agent to your constraints, your needs or your requests requires other paid components — a third-party publisher's software licence, paid API access to one of your applications, hosting of health data, for which French law requires an HDS-certified host (art. L. 1111-8 of the French Public Health Code), SecNumCloud-qualified hosting, a speech synthesis service, particular hardware —, they are offered to you as an option or on quotation and re-invoiced at the cost incurred; nothing is committed without your written agreement. Where the artificial intelligence model you choose entails an additional cost, that cost is shown to you before you order and re-invoiced to you at the cost incurred, with no margin.
What to expect
Go-live
2 to 3 weeks
Agent designed, channels connected, team trained.
Steady state
4 to 7 weeks
After a few weeks of real use, once the agent's behaviour matches what you expect.
Indicative estimate, adjusted to the options you keep. It is not a delivery commitment.
Our commitment
Four guarantees that matter to your responses
Your internal positions stay with youLocal inference or an isolated resource hosted in France; no reply material entrusted to a third party, no data used to train a model.
Data in France, under French lawYour reply material and your internal positions: minimisation and location in France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
Your ministerial correspondence department keeps the decisionThe agent produces sourced first versions, which can be checked and altered; no approval is automated.
Human oversight & traceabilityOn your volume of referrals and the breadth of your body of replies: systematic logging and tracking, in line with the AI Act.
Going further
Related resources
Frequently asked questions
Your questions, our answers
Does the agent write the final reply?
No. It puts together a sourced first version. The final drafting and the signature rest with the competent authority.
How is consistency with past positions assured?
The agent finds the replies already given on the subject and cites them. Any divergence from an earlier position shows up immediately.
What does it do with a point never dealt with?
It flags it as new and gathers the neighbouring material, without producing a new position. Settling a position commits the administration.
What body of material does it draw on?
The texts, the data and the earlier replies you open to it, defined at the design stage.
Is your reply material protected?
Yes. The agent is hosted in France, on local inference or an isolated resource, with the deployment objective of processing and access operated within the European Union and an architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
How long does it take to deploy this agent?
A few weeks as a rule, depending on the volume of referrals and the breadth of your body of replies, after a free audit then a phase of design, integration and testing.
Related use cases
Other agents for drafting
Let's talk
Let's size up the potential in your referrals
15 minutes to frame your volumes and your body of replies — hosted in France, supervised, with no commitment.