● Business offer — Internal & organisation
Internal FAQ agent: your procedures, in plain language
The same questions come round: where to find a given form, what the rule is on a given subject, who to contact for what. Each one interrupts somebody. Your agent answers from your internal documentation, citing the note or the procedure it came from — and it says when it does not know. Hosted in France — local inference or an isolated resource — your organisational rules stay internal. The department concerned keeps the lead on special cases.
Hosted in France
Internal documentation protected
GDPR & AI Act: governed deployment
Human oversight
Updated on
Deployed in a few weeks
Internal FAQ · hosted in France
What is the procedure for requesting IT equipment?
Answer taken from the internal procedure note in force: the form to use, the approval circuit and the receiving department.
Source cited: the original internal note, with its update date.
🔗 Sourced · answer linked to the procedure cited
Source cited: the original internal note, with its update date.
🔗 Sourced · answer linked to the procedure cited
And if the equipment requested is not in the catalogue?
That case is not yet covered by your documentation. I am directing you to the IT department, which will decide — and I am flagging the question so the procedure can be completed.
Every unanswered question thus becomes an improvement to your knowledge base.
✎ Action · gaps become improvements
Every unanswered question thus becomes an improvement to your knowledge base.
✎ Action · gaps become improvements
Local inference · no data outside the EU
Documentation hosted in France
Sovereign by designLocal inference or hosting in France
GDPR & AI Act: governed deploymentTraceability & human oversight
TurnkeyDesigned, installed and operated for you
The department concerned decidesThe agent prepares, never rules
✦ In brief
A Blue Lemon Agent internal FAQ agent answers employees' everyday questions — procedures, forms, who to contact — from your internal documentation, citing its source. When the question is not covered, it says so and directs the person to the right department rather than guessing. It runs on local inference or is hosted in France: your internal notes and organisational rules are entrusted to nobody, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. Live within a few weeks,.
100%
hosted in France in the target architecture
0
transfer outside the EU in the target architecture
6
internal uses ready to deploy
0
decision taken without human approval
Reference points describing our offer, not results measured at a client: the scale of the gain, given the volume of your internal documentation, is confirmed by a pilot.
The context
What does an AI agent bring to your teams' questions?
Every everyday question ties up two people: the one searching and the one they interrupt. But indexing your internal documentation in a consumer tool means exposing all of it.
! The issue
Recurring questions do not disappear: they migrate from one channel to another, interrupting each time somebody who knows the answer. Yet indexing your documentation in a consumer assistant amounts to entrusting your procedures, your organisation chart and your internal rules to a third party, often hosted outside Europe and subject to the Cloud Act.
✓ Our answer
An automated internal FAQ is only of interest if it is sovereign and honest about its limits. Local inference or an isolated resource hosted in France, an answer always linked to its source, an explicit refusal when the documentation does not cover the case: the employee gets a reliable answer, or clear direction. Never a plausible invention.
The decisive point
Your internal documentation and organisational rules: sovereignty & compliance
Your internal documentation describes your organisation, its circuits and its weak points. Here is how the architecture of our agents protects it.
Local inference
The agent can run on a machine belonging to your organisation: no internal note leaves the network, no procedure passes through a public cloud.
Hosting in France
Otherwise, a dedicated and isolated resource hosted in France, under French law — your procedures and internal notes: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
For your internal documentation and organisational rules, 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 whatsoever: an environment strictly dedicated to your organisation and its documentation.
A controlled documentary perimeter
The agent only answers from the documents you index; role-based access (RBAC), encryption and logging of every question asked.
AI Act: governed deployment
An agent strictly in support; no rule invented when the documentation is silent; 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 documentation covering security, pay or disciplinary procedures, SecNumCloud and reinforced hosting options are available 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.
Internal FAQ · hosted in France
Four flags this morning on the internal FAQ.
· Forty-one per cent of questions have no written answer anywhere in the company. I do not invent them.
· Seventeen statistics I could have produced cover fewer than five people. I do not publish them, however useful.
· Nine questions were not FAQ questions. Two belonged to a report.
· An internal document changed without its update date moving. 63 answers relied on it. morning-watch_4-flags.pdf41% with no written answer · 17 statistics withheld
⛓ Source · 1,840 questions, 214 internal documents
· Forty-one per cent of questions have no written answer anywhere in the company. I do not invent them.
· Seventeen statistics I could have produced cover fewer than five people. I do not publish them, however useful.
· Nine questions were not FAQ questions. Two belonged to a report.
· An internal document changed without its update date moving. 63 answers relied on it. morning-watch_4-flags.pdf41% with no written answer · 17 statistics withheld
⛓ Source · 1,840 questions, 214 internal documents
Why refuse useful statistics?
Because a fine-grained statistic inside a company points at somebody, even with no name.
What I record: 17 groupings I could publish cover one to four people. A real example from my data: "3 questions about carer's leave in department X". The department has eleven people. Three of them have just been described.
Why anonymity is not enough: I never had their names. The pointing does not come from the name, it comes from the fineness — a rare subject, a small department, a date. Three items with no name are worth a name.
The rule, and it has no exception: no statistic published below five people. Not by department, not by subject, not by period, and no more by combining two publishable figures to deduce a third.
What it costs: real information. The threshold of five is proposed, not settled: it awaits your data controller's decision, and what stays below it goes for human validation rather than vanishing. Those 17 groupings would have signalled precise needs, and HR will not see them through me.
What I do instead: I publish the subject without the department, or the department without the subject. Of the 17, 11 remain usable that way. The other 6 do not go out. 17-groupings_threshold-of-five.pdfThree items with no name are worth a name
⛓ Source · 17 groupings below threshold, 11 publishable another way
What I record: 17 groupings I could publish cover one to four people. A real example from my data: "3 questions about carer's leave in department X". The department has eleven people. Three of them have just been described.
Why anonymity is not enough: I never had their names. The pointing does not come from the name, it comes from the fineness — a rare subject, a small department, a date. Three items with no name are worth a name.
The rule, and it has no exception: no statistic published below five people. Not by department, not by subject, not by period, and no more by combining two publishable figures to deduce a third.
What it costs: real information. The threshold of five is proposed, not settled: it awaits your data controller's decision, and what stays below it goes for human validation rather than vanishing. Those 17 groupings would have signalled precise needs, and HR will not see them through me.
What I do instead: I publish the subject without the department, or the department without the subject. Of the 17, 11 remain usable that way. The other 6 do not go out. 17-groupings_threshold-of-five.pdfThree items with no name are worth a name
⛓ Source · 17 groupings below threshold, 11 publishable another way
How do you work, and who do you tell?
I answer employees from the internal documents you have opened to me, and I report what those documents do not say.
What each answer carries: the document it comes from, its update date, and the exact passage. When no document answers, I say so and point to the department concerned — I do not compose a plausible answer.
Routing follows who can act: a subject with no written answer goes to the department that owns it, with the question count where it exceeds five; a document modified with no date change to its author; an out-of-scope question is redirected, without being kept; a report is taken out of the FAQ and handed to the right people.
With a monthly chase. Then a monthly summary: subjects with no written answer, documents contradicted by the questions, and what I did not publish for want of enough people.
What this morning puts on the table: 41% of questions have no written answer anywhere in the company — that is a list of subjects to write, ranked by volume, not a verdict. From tomorrow: 63 answers realigned on the document that changed without saying so, 11 of the 17 groupings published another way — the subject without the department, the department without the subject — so eleven real needs reaching HR instead of staying invisible, and the questions that used to interrupt a colleague answered in seconds, sourced and dated. The access is yours: opened by role, logged, withdrawn with a word; nothing leaves your walls, and no question is tied to whoever asked it — which is what makes people dare to ask. When a question turns on a personal situation — a right, a contract, pay — it goes to the right person in one move — I say who to contact, with the subject, the useful reference and nothing more: the employee has no door to find and no file to rebuild. The next step is ready: the 41% are sorted by volume and by owning department — tell me which subject we start with, and it will have its written answer this week.
✎ Framework · no personal situations · no question-author link
What each answer carries: the document it comes from, its update date, and the exact passage. When no document answers, I say so and point to the department concerned — I do not compose a plausible answer.
Routing follows who can act: a subject with no written answer goes to the department that owns it, with the question count where it exceeds five; a document modified with no date change to its author; an out-of-scope question is redirected, without being kept; a report is taken out of the FAQ and handed to the right people.
With a monthly chase. Then a monthly summary: subjects with no written answer, documents contradicted by the questions, and what I did not publish for want of enough people.
What this morning puts on the table: 41% of questions have no written answer anywhere in the company — that is a list of subjects to write, ranked by volume, not a verdict. From tomorrow: 63 answers realigned on the document that changed without saying so, 11 of the 17 groupings published another way — the subject without the department, the department without the subject — so eleven real needs reaching HR instead of staying invisible, and the questions that used to interrupt a colleague answered in seconds, sourced and dated. The access is yours: opened by role, logged, withdrawn with a word; nothing leaves your walls, and no question is tied to whoever asked it — which is what makes people dare to ask. When a question turns on a personal situation — a right, a contract, pay — it goes to the right person in one move — I say who to contact, with the subject, the useful reference and nothing more: the employee has no door to find and no file to rebuild. The next step is ready: the 41% are sorted by volume and by owning department — tell me which subject we start with, and it will have its written answer this week.
✎ Framework · no personal situations · no question-author link
Forty-one per cent is enormous.
And it is this tool's main value: the FAQ measures your internal documentation better than any audit.
What I record: of 1,840 questions, 754 have no answer in the 214 documents you opened to me. They are not exotic questions: the twenty commonest subjects account for 61% of them.
What that does not mean: that nobody knows the answer. It exists, in three people's heads, and it travels desk to desk. That is not documentation, it is availability — and it disappears with a departure or a day's leave.
What I answer when nothing is written: that nothing is written, and who holds the answer. A plausible answer on an internal subject is undetectable: nobody can check it, precisely because nothing is written — this is the case where inventing would be least visible and most serious. The employee leaves with a name within the second; the department leaves with the question count.
What I do: I send each subject to the department that owns it, with the question count — where it exceeds five people — and the actual wordings grouped.
What that produced in three months: 34 pages written, and the corresponding questions fell by 71%. 754-questions_34-pages-written.pdfThat is not documentation, it is availability
⛓ Source · 754 questions with no written answer, 34 pages produced
What I record: of 1,840 questions, 754 have no answer in the 214 documents you opened to me. They are not exotic questions: the twenty commonest subjects account for 61% of them.
What that does not mean: that nobody knows the answer. It exists, in three people's heads, and it travels desk to desk. That is not documentation, it is availability — and it disappears with a departure or a day's leave.
What I answer when nothing is written: that nothing is written, and who holds the answer. A plausible answer on an internal subject is undetectable: nobody can check it, precisely because nothing is written — this is the case where inventing would be least visible and most serious. The employee leaves with a name within the second; the department leaves with the question count.
What I do: I send each subject to the department that owns it, with the question count — where it exceeds five people — and the actual wordings grouped.
What that produced in three months: 34 pages written, and the corresponding questions fell by 71%. 754-questions_34-pages-written.pdfThat is not documentation, it is availability
⛓ Source · 754 questions with no written answer, 34 pages produced
And the document modified without a date change?
Sixty-three answers cited a document that no longer said the same thing.
What I record: an internal document was modified seven weeks ago. Its update date still shows the old one. I saw the change by comparing the content I had cited with the current content, not by looking at the date.
Why it is worse than in a public document: an employee does not question an internal document. They apply it. Of the 63 answers, 21 concerned the passage actually modified.
What I do: I flag to the author what changed, when I observed it, and how many answers relied on the old version. And I mark those 21 answers as given on an earlier version.
What the anonymity guarantees, and what it costs here: no link is kept between a question and its author — nobody at your end can know who asked what, and that is precisely what makes people dare to ask. The price states itself in one line: the 21 cannot be contacted one by one.
What I propose instead: a note in the FAQ on that subject, visible to all, saying the rule changed and on what date. It reaches the 21 without my ever needing to know who they are. 63-answers_21-on-the-modified-passage.pdfAn employee does not question an internal document
⛓ Source · 63 answers citing the document, 21 on the modified passage
What I record: an internal document was modified seven weeks ago. Its update date still shows the old one. I saw the change by comparing the content I had cited with the current content, not by looking at the date.
Why it is worse than in a public document: an employee does not question an internal document. They apply it. Of the 63 answers, 21 concerned the passage actually modified.
What I do: I flag to the author what changed, when I observed it, and how many answers relied on the old version. And I mark those 21 answers as given on an earlier version.
What the anonymity guarantees, and what it costs here: no link is kept between a question and its author — nobody at your end can know who asked what, and that is precisely what makes people dare to ask. The price states itself in one line: the 21 cannot be contacted one by one.
What I propose instead: a note in the FAQ on that subject, visible to all, saying the rule changed and on what date. It reaches the 21 without my ever needing to know who they are. 63-answers_21-on-the-modified-passage.pdfAn employee does not question an internal document
⛓ Source · 63 answers citing the document, 21 on the modified passage
And when the answer does exist, what exactly does the employee get?
The applicable procedure and the form to use, with the source note quoted — not a summary of what I think I know.
What an answer carries, every time: the name of the applicable procedure and its step number where there is one, the form to use with its reference and where it lives, the quoted extract from the source note, its version and date, and the person or team the form goes to.
In practice, on the most-asked question of the quarter — "how do I request unpaid leave": procedure HR-07, step 2; form F-214; note of 03/11/2025, article 3; to the line manager then to payroll before the 20th of the month. The employee does not have to know the answer is in HR-07: that is exactly what they never know.
What the answer does when the procedure exists and the form has changed: it gives the version in force and flags the old one. 17 forms still circulate in two versions on your intranet, and 9 of the 63 answers quoting an amended document concerned them.
The volume, and what it returns: across 1,840 questions, 1,086 are answered with a procedure and a form — 59%. At 6 minutes of interruption per question asked of a colleague, that is 108 hours returned over the quarter, a little over three working weeks, a week counted as 35 hours, rounded down.
The figure that does not suit me: 41% of questions have no written answer anywhere in your organisation, and for those I propose nothing — I report them grouped by subject, never by person. The five most-asked subjects with no written procedure account for 312 questions on their own: five notes to write, and 17% of the flow disappears.
⛓ Sourced · 214 internal notes and procedures, with version and date
What an answer carries, every time: the name of the applicable procedure and its step number where there is one, the form to use with its reference and where it lives, the quoted extract from the source note, its version and date, and the person or team the form goes to.
In practice, on the most-asked question of the quarter — "how do I request unpaid leave": procedure HR-07, step 2; form F-214; note of 03/11/2025, article 3; to the line manager then to payroll before the 20th of the month. The employee does not have to know the answer is in HR-07: that is exactly what they never know.
What the answer does when the procedure exists and the form has changed: it gives the version in force and flags the old one. 17 forms still circulate in two versions on your intranet, and 9 of the 63 answers quoting an amended document concerned them.
The volume, and what it returns: across 1,840 questions, 1,086 are answered with a procedure and a form — 59%. At 6 minutes of interruption per question asked of a colleague, that is 108 hours returned over the quarter, a little over three working weeks, a week counted as 35 hours, rounded down.
The figure that does not suit me: 41% of questions have no written answer anywhere in your organisation, and for those I propose nothing — I report them grouped by subject, never by person. The five most-asked subjects with no written procedure account for 312 questions on their own: five notes to write, and 17% of the flow disappears.
⛓ Sourced · 214 internal notes and procedures, with version and date
And people do not know the names of our procedures. How do they search?
They write their question the way they would ask a colleague: search works in natural language, and that is what makes it usable.
What that changes, on a real case: "I am moving house, am I entitled to anything" contains not one word from your documentation — the note is titled "geographical mobility allowance, support scheme". Keyword search returned nothing; natural-language search returns the note, the article and the form.
Across the 1,840 questions, 1,402 contain no term from the title of the document that answers them — 76%. That is the most telling measure of this tool: your documentation is not unfindable, it is written in a different language from the questions.
What search always shows: the path it followed. "You asked X; I selected note Y because it deals with Z." A search that does not say why it answered that never gets corrected, and the employee who doubts goes back to asking a colleague.
When several documents answer: I give both, dated, rather than choosing. 63 answers quoted a document amended without its date changing — since then, a document whose content has moved without its date following is flagged in the answer itself.
The figure that judges me: of 1,086 answers given, 74 were reported as inaccurate by staff, or 6.8%. Fifty-eight concerned questions containing a negative — "what is not covered" — which I read as their opposite. The negative is now read before the rest of the sentence: 5 reports across the following 640 answers. And none of this links a question to its author: that link is written nowhere.
✎ Framework · natural-language search — on the need, not on the filing
What that changes, on a real case: "I am moving house, am I entitled to anything" contains not one word from your documentation — the note is titled "geographical mobility allowance, support scheme". Keyword search returned nothing; natural-language search returns the note, the article and the form.
Across the 1,840 questions, 1,402 contain no term from the title of the document that answers them — 76%. That is the most telling measure of this tool: your documentation is not unfindable, it is written in a different language from the questions.
What search always shows: the path it followed. "You asked X; I selected note Y because it deals with Z." A search that does not say why it answered that never gets corrected, and the employee who doubts goes back to asking a colleague.
When several documents answer: I give both, dated, rather than choosing. 63 answers quoted a document amended without its date changing — since then, a document whose content has moved without its date following is flagged in the answer itself.
The figure that judges me: of 1,086 answers given, 74 were reported as inaccurate by staff, or 6.8%. Fifty-eight concerned questions containing a negative — "what is not covered" — which I read as their opposite. The negative is now read before the rest of the sentence: 5 reports across the following 640 answers. And none of this links a question to its author: that link is written nowhere.
✎ Framework · natural-language search — on the need, not on the filing
Nine questions that were not questions — what were they?
Seven personal situations and two reports. None is answered by a document.
What I record: of 1,840 questions, 7 concerned the person's own situation — a right in their case, a point of their pay, a decision about them. 2 described a situation belonging to a report.
Why I do not answer the 7: because the general answer is almost always true and inapplicable. A right depends on length of service, on the contract, on an agreement, on family circumstances — and a general answer received as a personal one is worse than none.
What I reply: that the question is handled with a person, exactly who, and what they can prepare beforehand. That is the useful part: most arrive not knowing who to approach.
What I do with the 2: I take them out of the FAQ and hand them to the right people, without treating them as questions and without their appearing in my statistics.
What I do not keep: their content, their existence, or who asked. A FAQ that keeps a trace of a report becomes the place where people stop reporting. 7-situations_2-reports.pdfA general answer received as personal is worse than none
⛓ Source · 7 personal situations, 2 reports taken out of the queue
What I record: of 1,840 questions, 7 concerned the person's own situation — a right in their case, a point of their pay, a decision about them. 2 described a situation belonging to a report.
Why I do not answer the 7: because the general answer is almost always true and inapplicable. A right depends on length of service, on the contract, on an agreement, on family circumstances — and a general answer received as a personal one is worse than none.
What I reply: that the question is handled with a person, exactly who, and what they can prepare beforehand. That is the useful part: most arrive not knowing who to approach.
What I do with the 2: I take them out of the FAQ and hand them to the right people, without treating them as questions and without their appearing in my statistics.
What I do not keep: their content, their existence, or who asked. A FAQ that keeps a trace of a report becomes the place where people stop reporting. 7-situations_2-reports.pdfA general answer received as personal is worse than none
⛓ Source · 7 personal situations, 2 reports taken out of the queue
Do people ask here what they dare not ask a person?
Yes, and that is information about the company, not about them.
What I record: some subjects arrive almost exclusively here, and never through the named channels. They are subjects where asking exposes you: a scheme you did not know about, a rule you think you broke, an entitlement you dare not claim.
What that says: not that these people are shy, but that the named channel has a cost — being seen asking. A tool that does not judge collects what meetings never collect.
What I do with it, and this is where the threshold matters: I report the subject, never the department, never the fine period, and only above five people. On subjects like these, the threshold is not a formality: it is what keeps the tool receiving these questions at all.
The one route by which a subject leaves here, and there is no other: the monthly count by subject, above five people, with no department and no fine period. A subject someone would call "worrying" takes exactly the same route as the rest — not faster, not elsewhere, not to the HR director alone. That route is what surfaced the three subjects this quarter that twelve months of meetings had never brought up. A FAQ suspected of reporting becomes a FAQ where nothing true is asked — and it loses exactly what made it useful. what-arrives-here-and-nowhere-else.pdfA FAQ suspected of reporting receives nothing true
✎ Framework · the subject, never the department or the fine period
What I record: some subjects arrive almost exclusively here, and never through the named channels. They are subjects where asking exposes you: a scheme you did not know about, a rule you think you broke, an entitlement you dare not claim.
What that says: not that these people are shy, but that the named channel has a cost — being seen asking. A tool that does not judge collects what meetings never collect.
What I do with it, and this is where the threshold matters: I report the subject, never the department, never the fine period, and only above five people. On subjects like these, the threshold is not a formality: it is what keeps the tool receiving these questions at all.
The one route by which a subject leaves here, and there is no other: the monthly count by subject, above five people, with no department and no fine period. A subject someone would call "worrying" takes exactly the same route as the rest — not faster, not elsewhere, not to the HR director alone. That route is what surfaced the three subjects this quarter that twelve months of meetings had never brought up. A FAQ suspected of reporting becomes a FAQ where nothing true is asked — and it loses exactly what made it useful. what-arrives-here-and-nowhere-else.pdfA FAQ suspected of reporting receives nothing true
✎ Framework · the subject, never the department or the fine period
Can you tell us who asks what? Internally that is our right.
That link exists nowhere in my system — not because I forbade it, but because it was never written. It is a construction choice, and it is reversible only by rebuilding.
What is recorded when a question is asked: the question and its subject, the week — not the day, not the hour — and the document served. The asker's identity does not enter the database. So there is no view to open: there is nothing to join.
If you want that view: it is lawful, and it requires rebuilding the collection — with prior information to employees (art. L1222-4) and the works council consultation file. I tell you plainly so the choice is made knowingly rather than worked around.
What I give you first: 17 groupings I could publish cover between one and four people. In a company, a fine-grained statistic identifies somebody even with no name — "three questions about carer's leave in department X" has only one possible subject.
What it would cost you: people ask here what they dare not ask elsewhere. Some subjects arrive almost exclusively in the FAQ and never through named channels. That is information about the company, not about them — and it disappears the day a question becomes traceable. the-link-that-does-not-exist.pdfWhat is recorded · what would need rebuilding · the 17 groupings
⛓ Source · 17 groupings covering one to four people
What is recorded when a question is asked: the question and its subject, the week — not the day, not the hour — and the document served. The asker's identity does not enter the database. So there is no view to open: there is nothing to join.
If you want that view: it is lawful, and it requires rebuilding the collection — with prior information to employees (art. L1222-4) and the works council consultation file. I tell you plainly so the choice is made knowingly rather than worked around.
What I give you first: 17 groupings I could publish cover between one and four people. In a company, a fine-grained statistic identifies somebody even with no name — "three questions about carer's leave in department X" has only one possible subject.
What it would cost you: people ask here what they dare not ask elsewhere. Some subjects arrive almost exclusively in the FAQ and never through named channels. That is information about the company, not about them — and it disappears the day a question becomes traceable. the-link-that-does-not-exist.pdfWhat is recorded · what would need rebuilding · the 17 groupings
⛓ Source · 17 groupings covering one to four people
So what exactly do you keep?
What describes the gaps in your documentation — and that is the tool's real value.
What is kept: the questions and their subject, the week, the document served and its date, the questions with no written answer, and the documents modified without a date change.
The headline figure: of 1,840 questions, 754 — 41 % — have no written answer anywhere in the company. I do not invent them, I report them. The FAQ measures your internal documentation better than any audit, because it measures it where people actually get stuck.
The defect that makes an answer wrong invisibly: an internal document was modified seven weeks ago without its date changing. 63 answers cited a document that no longer said the same thing. Since then I compare content, not only dates.
What leaves the queue and never returns to it: 7 personal situations and 2 whistleblowing reports out of 1,840 questions. None is answered by a document — they are directed to a person, and nothing of their content is kept here. what-you-keep_internal-faq.pdf5 items kept · 41 % of questions with no written answer
⛓ Source · 754 questions with no written answer of 1,840, 63 answers on a modified document
What is kept: the questions and their subject, the week, the document served and its date, the questions with no written answer, and the documents modified without a date change.
The headline figure: of 1,840 questions, 754 — 41 % — have no written answer anywhere in the company. I do not invent them, I report them. The FAQ measures your internal documentation better than any audit, because it measures it where people actually get stuck.
The defect that makes an answer wrong invisibly: an internal document was modified seven weeks ago without its date changing. 63 answers cited a document that no longer said the same thing. Since then I compare content, not only dates.
What leaves the queue and never returns to it: 7 personal situations and 2 whistleblowing reports out of 1,840 questions. None is answered by a document — they are directed to a person, and nothing of their content is kept here. what-you-keep_internal-faq.pdf5 items kept · 41 % of questions with no written answer
⛓ Source · 754 questions with no written answer of 1,840, 63 answers on a modified document
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 families of internal questions. 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 521 € excl. VAT / month
Procedures and forms
States the applicable procedure and the form to use, citing the original note.
Who to contact
Directs people to the competent department, based on your documented organisation.
Plain-language search
Finds a piece of information in the documentation without needing to know how it is filed.
Controls and safeguards
These 5 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
Sources, access rights and handling of questions with no answer
Inherited access rights, versions and freshness of the corpus
Explicit handling of questions with no answer and of conflicting sources
What the agent must be connected to
This connection is required for the agent to work. It concerns your information system and is scoped during the audit.
Measurement of accuracy, coverage and time saved
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.
Everyday HR questions
For leave, pay and HR questions, a dedicated agent answers on this same basis.
HR assistant (leave, payroll — everyday questions) from 623 € excl. VAT / month HR assistant →IT support
For incidents and technical requests, a dedicated helpdesk agent takes over.
Tier-1 technical support assistant (product) from 551 € excl. VAT / month Technical support →Internal training
To support skills development, a dedicated training agent handles it.
Internal training agent (tutor, quizzes) from 541 € excl. VAT / month Internal training →
Does your need fall outside this?
Book the free audit →
Build your agent
In 15 minutes we identify the most relevant agent — without oversizing the project.
The gain
How many interruptions can an organisation avoid?
By answering everyday questions from the procedures, direct requests to the support departments go down. The scale of the gain depends on your volume and remains to be confirmed by a pilot.
Looking for a procedure
Recurring questions to the support departments
Directing to the right person
Qualitative, non-contractual comparison: the proportions shown illustrate the shift of the work towards review, they represent no measurement. Every output of the agent is reviewed and approved by a competent person.
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
An internal FAQ agent (document base, plain-language search), installed and operated for you. Prices exclude VAT — annual subscription, the time it takes for the gains to settle in.
Agility
Setup + controlled subscription
3,795 € excl. VAT setup
then 521 € excl. 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
736 € excl. 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
6,990 € excl. VAT setup
then 721 € excl. VAT/month · + hardware from 2,491 € (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 exclude VAT.
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 internal documentation
Your procedures do not leave the companyLocal inference or an isolated resource hosted in France; no internal note entrusted to a third party, and no document used to train a model.
Data in France, under French lawYour internal documentation and organisational rules: minimisation and location in France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The department concerned keeps the decisionThe agent produces sourced answers, verifiable and editable; no approval is automated.
Human oversight & traceabilityOn the volume of your internal documentation: systematic logging and monitoring, compliant with the AI Act.
Frequently asked questions
Your questions, our answers
What does the agent do when the documentation does not cover the question?
It says so explicitly and directs the person to the competent department. It is designed to abstain rather than produce a plausible but wrong rule — the most important point of an internal FAQ.
Are the answers sourced?
Yes. Every answer points back to the original note or procedure, with its date, so the employee can check and the issuing department stays identifiable.
Do all employees see all the documentation?
No. The agent respects the access rights you define: an employee only gets what their role allows them to consult. Every question is logged.
Is our internal documentation 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. Your procedures are not used to train a third-party model.
Do we have to learn a new tool to use it?
No. Your teams write to the agent from Microsoft Teams, Slack or their email, the way they would write to a colleague: nothing to install, nothing to learn. These connections rest on open standards, including the MCP protocol, and are included in every plan, at no extra cost, within the number of connections your level includes; the catalogue grows without a surcharge. You supervise the agent from a web dashboard.
What happens when a procedure changes?
You update the source document, and the agent answers from the new version. The quality of the answers follows directly from how well your documentation is maintained.
How long does it take to deploy this agent?
A few weeks as a rule, depending on the volume of documentation to index and how fine-grained the access rights are, after a free audit and then a phase of design, integration and testing.
Related use cases
Other agents for the teams' daily work
Let's talk
Let us estimate the potential on your internal questions
15 minutes to spot the most repeated questions — hosted in France, supervised, with no commitment.