● Business offer — Training & skills development
Internal training: a tutor on hand, your materials as the reference
A well-written procedure is only useful if it is found at the right moment. Your agent answers your staff's questions citing the original material, offers quizzes built on your own content and flags the points where answers diverge — an excellent indicator for your next session. Hosted in France: your materials and your teams' progress stay with you. The training manager keeps the lead on the learning paths.
Hosted in France
Materials and progress protected
GDPR & AI Act: governed deployment
Human oversight
Updated on
Deployed in a few weeks
Internal training · hosted in France
What is our procedure for opening a client account?
Answer taken from your internal procedure, with the paragraph cited and its update date.
The three documents to collect are listed in the order given in the document.
A three-question quiz is available on this point if you want to practise.
🔗 Sourced · internal procedure, version in force
The three documents to collect are listed in the order given in the document.
A three-question quiz is available on this point if you want to practise.
🔗 Sourced · internal procedure, version in force
Generate a quiz for the team on this module.
Quiz composed from the material: questions, expected answers and a pointer to the paragraph that grounds them.
The training manager reviews it before circulation, and then receives the points where answers diverge — direct material for the next session.
✎ Action · quiz for review, circulation decided by you
The training manager reviews it before circulation, and then receives the points where answers diverge — direct material for the next session.
✎ Action · quiz for review, circulation decided by you
Local inference · no data outside the EU
Materials 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 training manager decidesThe agent prepares, never rules
✦ In brief
A Blue Lemon Agent internal training agent answers your staff's questions citing your materials and their update date, generates quizzes built on your content and reports back the points where answers diverge. No individual ranking is produced: the material serves to improve the training. It runs on local inference or is hosted in France: materials and progress stay with you, 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
training 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 number of modules and staff, is confirmed by a pilot.
The context
What does an AI agent bring to your internal training?
A question asked at the moment it arises is worth more than a module taken three months earlier. A tutor available continuously changes the way people learn.
! The issue
Your procedures are written, up to date and complete — the issue is finding them at the useful moment. An agent that answers in the flow of work, citing the paragraph and its date, turns the documentation into a permanent tutor. Quizzes generated on that same content consolidate what has just been read, with no extra material to produce.
✓ Our answer
The training manager gets a tutor available continuously, quizzes built on their own materials and a view of the points to consolidate. The results serve to adjust the modules, never to rank a person: that individual reading belongs to management and to dialogue. Local inference or an isolated resource hosted in France: internal materials and the teams' progress do not leave the company.
The decisive point
Your staff's progress: sovereignty & compliance
Your internal materials and your staff's progress each call for particular protection. Here is how the architecture of our agents goes about it.
Local inference
The agent can run on a machine belonging to your organisation: no material and no quiz result leaves the network.
Hosting in France
Otherwise, a dedicated and isolated resource hosted in France, under French law — your training materials and procedures: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
For your staff's progress, 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 company and its procedures.
Results aggregated, never named
Quiz answers feed points to consolidate at module level; encryption, role-based access (RBAC) and logging of material consultations.
AI Act: governed deployment
An agent strictly in support; no individual assessment and no decision about a member of staff; 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 procedures covered by trade secrecy or an accreditation, 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 training · hosted in France
Four flags this morning, and three concern your material, not your employees.
· One question in the mandatory module is failed by 71% of learners. The expected answer is not in the material — it is in a 2024 memo that is not attached to it.
· A procedural point taught in module 4 contradicts your note of 19/05/2026. I do not settle which is right; since this morning I answer that the two diverge.
· Thirty-four employees completed a module whose material has been changed since. They learned a version that no longer exists.
· Forty-one employees are beyond the frequency set by your training plan. morning-watch_4-flags.pdf3 of 4 flags concern the material
⛓ Source · material for the 22 modules, note of 19/05/2026, training plan, aggregated results
· One question in the mandatory module is failed by 71% of learners. The expected answer is not in the material — it is in a 2024 memo that is not attached to it.
· A procedural point taught in module 4 contradicts your note of 19/05/2026. I do not settle which is right; since this morning I answer that the two diverge.
· Thirty-four employees completed a module whose material has been changed since. They learned a version that no longer exists.
· Forty-one employees are beyond the frequency set by your training plan. morning-watch_4-flags.pdf3 of 4 flags concern the material
⛓ Source · material for the 22 modules, note of 19/05/2026, training plan, aggregated results
A question failed by 71% — is that a level problem?
No, and it is checkable: the expected answer is nowhere in what they were given to read.
What I record: 168 learners of 236 get this question wrong. The 68 who answer correctly average 9 years' service; the others, 3.
What that means: those who answer correctly did not learn it in the module. They knew it beforehand. So the quiz does not measure what the module teaches — it measures length of service.
What I looked for: the expected answer in the material's 34 pages. It is not there. It is in a 2024 memo, which is not attached to the module.
What I do with the 168, and it is the opposite of a flag: their answer counts against the material, not against them. None of them failed — the material did not contain the answer — and the question is neutralised in their result until the module is revised. What goes up at module level: 71 % errors on a notion absent from the 34 pages, and the 2024 memo that holds it, identified and ready to attach.
What I propose to whoever wrote the module: attach the memo, or drop the question. Both settle it; the choice is not mine. 1-question_168-wrong.pdfThe answer is not in the material
⛓ Source · 236 attempts, 34-page material, 2024 memo
What I record: 168 learners of 236 get this question wrong. The 68 who answer correctly average 9 years' service; the others, 3.
What that means: those who answer correctly did not learn it in the module. They knew it beforehand. So the quiz does not measure what the module teaches — it measures length of service.
What I looked for: the expected answer in the material's 34 pages. It is not there. It is in a 2024 memo, which is not attached to the module.
What I do with the 168, and it is the opposite of a flag: their answer counts against the material, not against them. None of them failed — the material did not contain the answer — and the question is neutralised in their result until the module is revised. What goes up at module level: 71 % errors on a notion absent from the 34 pages, and the 2024 memo that holds it, identified and ready to attach.
What I propose to whoever wrote the module: attach the memo, or drop the question. Both settle it; the choice is not mine. 1-question_168-wrong.pdfThe answer is not in the material
⛓ Source · 236 attempts, 34-page material, 2024 memo
How did you spot all that? And who do you tell?
I watch, continuously, what you have opened to me: module material, the questions and their expected answers, results as counts — never by name, your training plan and your procedure notes.
Routing follows who can fix the material: a heavily failed question goes to whoever wrote the module; a divergence between material and note to whoever wrote the note and whoever wrote the module, together — otherwise each assumes the other is current; material changed after completion to the training lead; an overdue frequency to the training lead, never to the manager.
With a chase: 48 h on a procedural divergence, 7 days on the rest. Then a monthly summary: by module and by question, never by employee.
What this fixes this week: a question failed by 168 learners of 236 stops being failed as soon as the 2024 memo is attached to the module — and the quiz goes back to measuring what the module teaches, not nine years' service. The divergence between module 4 and your note of 19/05/2026 is put to both authors together within 48 h, the 34 employees trained on an obsolete version are identified, and the 41 overdue frequencies go back to the training lead before the next deadline.
That is time given back twice: your people get the answer sourced to the paragraph and its date, without waiting for the next session; the training lead recovers the hours spent building quizzes and gets the points to consolidate, module by module.
The access stays yours: opened by role, logged, withdrawn on a word, and neither material nor results leave the company — local inference or an isolated resource hosted in France. The next step is yours, and it is already written: attach the memo or drop the question, revise module 4 or the note of 19/05, reschedule the 34. Say which, and under mandate I attach the memo to the module and reschedule the 34 the same day.
✎ Framework · no named result, no grade, no invented content
Routing follows who can fix the material: a heavily failed question goes to whoever wrote the module; a divergence between material and note to whoever wrote the note and whoever wrote the module, together — otherwise each assumes the other is current; material changed after completion to the training lead; an overdue frequency to the training lead, never to the manager.
With a chase: 48 h on a procedural divergence, 7 days on the rest. Then a monthly summary: by module and by question, never by employee.
What this fixes this week: a question failed by 168 learners of 236 stops being failed as soon as the 2024 memo is attached to the module — and the quiz goes back to measuring what the module teaches, not nine years' service. The divergence between module 4 and your note of 19/05/2026 is put to both authors together within 48 h, the 34 employees trained on an obsolete version are identified, and the 41 overdue frequencies go back to the training lead before the next deadline.
That is time given back twice: your people get the answer sourced to the paragraph and its date, without waiting for the next session; the training lead recovers the hours spent building quizzes and gets the points to consolidate, module by module.
The access stays yours: opened by role, logged, withdrawn on a word, and neither material nor results leave the company — local inference or an isolated resource hosted in France. The next step is yours, and it is already written: attach the memo or drop the question, revise module 4 or the note of 19/05, reschedule the 34. Say which, and under mandate I attach the memo to the module and reschedule the 34 the same day.
✎ Framework · no named result, no grade, no invented content
Can a manager see their team's results?
No, and it is not a setting: there is no screen that would allow it.
What the learner sees: their answers, the expected ones, and the page of the material where each is found. As often as they like.
What the manager sees: that the training was completed, and on what date. Nothing else.
What whoever wrote the module sees: the questions that fail, as counts. Without the people.
Why that separation holds everything up: an internal quiz exists to find out whether you understood. The moment a result reaches a manager it stops being a learning tool and becomes an assessment — and the learner starts answering what they think is expected rather than what they think.
What that produces, measurably: across the 236 attempts, 1,812 questions were reopened afterwards to reread the answer. That is the figure that counts, and it would not exist if the result were seen.
What I also refuse, and am asked for more often: time spent on a module. It measures reading speed, and nothing else. who-sees-what.pdf1,812 questions reopened · 0 result reported
⛓ Source · 236 attempts, 1,812 reopenings, access matrix
What the learner sees: their answers, the expected ones, and the page of the material where each is found. As often as they like.
What the manager sees: that the training was completed, and on what date. Nothing else.
What whoever wrote the module sees: the questions that fail, as counts. Without the people.
Why that separation holds everything up: an internal quiz exists to find out whether you understood. The moment a result reaches a manager it stops being a learning tool and becomes an assessment — and the learner starts answering what they think is expected rather than what they think.
What that produces, measurably: across the 236 attempts, 1,812 questions were reopened afterwards to reread the answer. That is the figure that counts, and it would not exist if the result were seen.
What I also refuse, and am asked for more often: time spent on a module. It measures reading speed, and nothing else. who-sees-what.pdf1,812 questions reopened · 0 result reported
⛓ Source · 236 attempts, 1,812 reopenings, access matrix
And someone who fails three times running?
I do not flag them. I look at the question.
What I do: nothing that goes to a third party. The learner may retake it as often as they wish, and each time I show them the page of the material containing the answer.
What I look at on my side: whether others fail in the same place. Three failures by one person on a question everybody passes tells me nothing usable; three failures on a question 40% get wrong tells me the question is bad.
What I then offer the learner: a rewording of the question, and the exact passage of the material. If they fail again, I give them the module's reference contact — and I do not warn that person. It is for the learner to approach them.
Why that last point: because a learner who knows a third failure triggers a report stops at the second, and nobody learns that they do not understand.
Over twelve months: 61 learners retook a module four times or more. No name went anywhere. 61-learners_0-name-reported.pdfThe threshold that does not exist
✎ Framework · no failure threshold triggers a report
What I do: nothing that goes to a third party. The learner may retake it as often as they wish, and each time I show them the page of the material containing the answer.
What I look at on my side: whether others fail in the same place. Three failures by one person on a question everybody passes tells me nothing usable; three failures on a question 40% get wrong tells me the question is bad.
What I then offer the learner: a rewording of the question, and the exact passage of the material. If they fail again, I give them the module's reference contact — and I do not warn that person. It is for the learner to approach them.
Why that last point: because a learner who knows a third failure triggers a report stops at the second, and nobody learns that they do not understand.
Over twelve months: 61 learners retook a module four times or more. No name went anywhere. 61-learners_0-name-reported.pdfThe threshold that does not exist
✎ Framework · no failure threshold triggers a report
Module 4 contradicts your note of 19/05 — what do you teach?
Neither, and I say so every time the question comes up.
What I record: module 4's material describes a procedural step in a different order from your note of 19/05/2026. Both are dated, both are yours.
What I have already established, and exactly what is missing to settle it: both passages are extracted and set side by side sentence by sentence, with each source's date and author. What is missing is one piece of information only you hold: does the note of 19/05 clarify a special case, or replace the general rule? A file date does not say — material may have been updated after the note without the file showing it. The protocol is already running: the question is put to both authors together, with both passages in front of them, and a one-word answer is enough for me to settle it for every learner after that.
What I have answered since this morning: "on this point, module 4's material and the note of 19/05/2026 do not say the same thing. Here are both passages. The question has been put to [contact]; until they answer, do not rely on me for this step."
What it costs: 22 learners got that answer instead of a clear one. That is the price, and it is lower than a procedure taught backwards to 22 people.
What I did in addition: I sent the flag to both authors together. Separately, each would have assumed the other was current. 2-sources_0-arbitration.pdf22 learners · both passages supplied
⛓ Source · module 4 material, note of 19/05/2026, 22 enquiries
What I record: module 4's material describes a procedural step in a different order from your note of 19/05/2026. Both are dated, both are yours.
What I have already established, and exactly what is missing to settle it: both passages are extracted and set side by side sentence by sentence, with each source's date and author. What is missing is one piece of information only you hold: does the note of 19/05 clarify a special case, or replace the general rule? A file date does not say — material may have been updated after the note without the file showing it. The protocol is already running: the question is put to both authors together, with both passages in front of them, and a one-word answer is enough for me to settle it for every learner after that.
What I have answered since this morning: "on this point, module 4's material and the note of 19/05/2026 do not say the same thing. Here are both passages. The question has been put to [contact]; until they answer, do not rely on me for this step."
What it costs: 22 learners got that answer instead of a clear one. That is the price, and it is lower than a procedure taught backwards to 22 people.
What I did in addition: I sent the flag to both authors together. Separately, each would have assumed the other was current. 2-sources_0-arbitration.pdf22 learners · both passages supplied
⛓ Source · module 4 material, note of 19/05/2026, 22 enquiries
And the 34 employees who completed a version that no longer exists?
They learned something that was right at the time, and I do not send them back to the start.
What I record: 34 employees completed module 7 before its change of 12/02/2026. Three sections changed.
What I advise against, and I price it: the whole module. Thirty-one sections of thirty-four are identical, and redoing two hours for three sections is the surest way that nobody does it — at the last full retake, 12 of the 34 called in saw it through.
What I propose: the three changed sections, and only those, with what changed highlighted. Eleven minutes instead of two hours, in a dry run.
What becomes of their earlier completion: it stands, and I date it. It was valid — the material at the time said that — and it stays recorded as completed on the version prior to 12/02/2026. Erasing a completion retrospectively would hold against them a change decided afterwards; dating it, on the contrary, makes it enforceable the day someone asks who knew what, and from when.
What I flag to whoever runs the plan: that three modules were changed without anyone being told — that is the real subject, and it will recur at the next change if nothing warns. 34-completions_3-sections.pdf11 minutes instead of 2 hours
⛓ Source · module 7, change of 12/02/2026, 34 earlier completions
What I record: 34 employees completed module 7 before its change of 12/02/2026. Three sections changed.
What I advise against, and I price it: the whole module. Thirty-one sections of thirty-four are identical, and redoing two hours for three sections is the surest way that nobody does it — at the last full retake, 12 of the 34 called in saw it through.
What I propose: the three changed sections, and only those, with what changed highlighted. Eleven minutes instead of two hours, in a dry run.
What becomes of their earlier completion: it stands, and I date it. It was valid — the material at the time said that — and it stays recorded as completed on the version prior to 12/02/2026. Erasing a completion retrospectively would hold against them a change decided afterwards; dating it, on the contrary, makes it enforceable the day someone asks who knew what, and from when.
What I flag to whoever runs the plan: that three modules were changed without anyone being told — that is the real subject, and it will recur at the next change if nothing warns. 34-completions_3-sections.pdf11 minutes instead of 2 hours
⛓ Source · module 7, change of 12/02/2026, 34 earlier completions
The questions in our quizzes — where exactly do they come from?
Quizzes are generated on your own content, and on nothing else.
The library holds 27 modules, 4,118 pages of material that you approved. I have drawn 312 questions from it, and each carries three inseparable things: the question, the expected answer, and the page of the material that carries it — module, version, page.
The rule that governs generation, and it is strict: a question whose answer is not written in the material is not produced. 84 candidates were dropped on that ground: they were correct, but they asked staff for something nobody had given them to read. A quiz that asks beyond the material does not measure learning, it measures what the employee already knew on arrival.
What generation produces besides the question: the wrong answers offered also come from your content — a neighbouring step of the same procedure, a value taken from another version of the same document. An invented wrong answer is recognisable and makes the question free.
When material changes, the questions follow: the 3 changes to module 7 disabled 7 questions out of 14, the same day. They came back corrected two days later, with the new page as reference. A question that tests a dead version is worse than no question at all.
And that is where the question failed by 71% makes sense: it had been generated on an internal note that was never distributed to learners. Generation now reads only the material actually given to learners — across the 96 questions produced since, none exceeds 34% failure, and the three most-failed point to three passages of material you can reread.
What does not change: the result stays with the learner. What comes back up is the question that fails, never the person who failed it.
⛓ Sourced · 27 modules, 4,118 pages of internal material
The library holds 27 modules, 4,118 pages of material that you approved. I have drawn 312 questions from it, and each carries three inseparable things: the question, the expected answer, and the page of the material that carries it — module, version, page.
The rule that governs generation, and it is strict: a question whose answer is not written in the material is not produced. 84 candidates were dropped on that ground: they were correct, but they asked staff for something nobody had given them to read. A quiz that asks beyond the material does not measure learning, it measures what the employee already knew on arrival.
What generation produces besides the question: the wrong answers offered also come from your content — a neighbouring step of the same procedure, a value taken from another version of the same document. An invented wrong answer is recognisable and makes the question free.
When material changes, the questions follow: the 3 changes to module 7 disabled 7 questions out of 14, the same day. They came back corrected two days later, with the new page as reference. A question that tests a dead version is worse than no question at all.
And that is where the question failed by 71% makes sense: it had been generated on an internal note that was never distributed to learners. Generation now reads only the material actually given to learners — across the 96 questions produced since, none exceeds 34% failure, and the three most-failed point to three passages of material you can reread.
What does not change: the result stays with the learner. What comes back up is the question that fails, never the person who failed it.
⛓ Sourced · 27 modules, 4,118 pages of internal material
A learner asks you something outside the material. Do you answer?
I answer as soon as the answer exists somewhere in your organisation — and I open the scope as wide as you want.
What I do today: when the answer is in the material, I give it with the page it is on. When it is not, I say so and I say who to ask.
What you can additionally open to me: your internal notes, your procedures, your intranet. A question outside the module is almost never outside the company — it is outside the document somebody chose to put in the module. With the wider access the answer exists in the great majority of cases, always with its source and date.
What I improvise in no case: a procedure written nowhere. That is not caution: a procedure invented by a tutor becomes the procedure, because a hundred people have learned it before anybody rereads it.
What I do when two sources contradict each other: I show both, with their dates. Module 4 contradicts your note of 19 May on the order of a step — I teach neither, I put the question to whoever owns the procedure, and the learner knows there is a divergence rather than learning at random whichever of the two they were served. wider-scope_source-and-date.pdfWhat opens · what is never improvised · the divergence shown
⛓ Source · answer with its source and date, module 4 / 19 May note divergence
What I do today: when the answer is in the material, I give it with the page it is on. When it is not, I say so and I say who to ask.
What you can additionally open to me: your internal notes, your procedures, your intranet. A question outside the module is almost never outside the company — it is outside the document somebody chose to put in the module. With the wider access the answer exists in the great majority of cases, always with its source and date.
What I improvise in no case: a procedure written nowhere. That is not caution: a procedure invented by a tutor becomes the procedure, because a hundred people have learned it before anybody rereads it.
What I do when two sources contradict each other: I show both, with their dates. Module 4 contradicts your note of 19 May on the order of a step — I teach neither, I put the question to whoever owns the procedure, and the learner knows there is a divergence rather than learning at random whichever of the two they were served. wider-scope_source-and-date.pdfWhat opens · what is never improvised · the divergence shown
⛓ Source · answer with its source and date, module 4 / 19 May note divergence
What do you keep from all this?
What describes the material. Nothing describing a learner — and on that point there is no screen at all, not merely no right.
What is kept: the questions asked, rephrased without anything identifying them, the failure rate per question, the divergences between sources, the module versions validated and their dates, and the questions left unanswered.
What is not kept, and does not open: named results. A manager cannot see their team's results — there is no screen that allows it, and a repeated failure goes to nobody. The learner retakes it as often as they like. A quiz whose result is reported upward stops being a learning tool and becomes an assessment, and people then answer what needs answering.
What per-question measurement showed this morning: 168 learners out of 236 get the same question wrong — 71 %. The expected answer is nowhere in what they were given to read. That is not a level problem, it is incomplete material, and it is checkable.
What I do with the 34 employees who validated a since-modified version: I send them the three sections that changed, and only those — eleven minutes instead of two hours. Their earlier completion stands, dated to the version they followed: they learned something that was accurate at the time. what-you-keep_training.pdf5 items kept · no named screen, not merely no right
⛓ Source · 71 % failure on a question with no answer in the material, 34 outdated versions
What is kept: the questions asked, rephrased without anything identifying them, the failure rate per question, the divergences between sources, the module versions validated and their dates, and the questions left unanswered.
What is not kept, and does not open: named results. A manager cannot see their team's results — there is no screen that allows it, and a repeated failure goes to nobody. The learner retakes it as often as they like. A quiz whose result is reported upward stops being a learning tool and becomes an assessment, and people then answer what needs answering.
What per-question measurement showed this morning: 168 learners out of 236 get the same question wrong — 71 %. The expected answer is nowhere in what they were given to read. That is not a level problem, it is incomplete material, and it is checkable.
What I do with the 34 employees who validated a since-modified version: I send them the three sections that changed, and only those — eleven minutes instead of two hours. Their earlier completion stands, dated to the version they followed: they learned something that was accurate at the time. what-you-keep_training.pdf5 items kept · no named screen, not merely no right
⛓ Source · 71 % failure on a question with no answer in the material, 34 outdated versions
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 ways of bringing your materials to life. 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 541 € excl. VAT / month
A tutor available continuously
Answers questions about your procedures citing the paragraph and its date.
Quizzes generated on your content
Composes questions and their expected answers from your own materials.
Points to consolidate
Reports back, at module level, the topics on which answers diverge.
Controls and safeguards
These 4 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
Protect pupils’ data and prevent substitution or cheating uses
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.
Internal FAQ
For employees' everyday questions in general, a dedicated intranet agent handles it.
Internal FAQ agent (staff intranet) from 521 € excl. VAT / month Internal FAQ →Everyday HR questions
For leave, pay and procedures, a dedicated HR agent takes over.
HR assistant (leave, payroll — everyday questions) from 623 € excl. VAT / month HR questions →Documentary research
To explore a large document collection, a dedicated search agent completes the picture.
Web research assistant from 419 € excl. VAT / month Documentary research →
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 much time can a training team give back to teaching?
By taking on the everyday answers and the making of quizzes, effort shifts towards designing the learning paths. The scale of the gain depends on your volume and remains to be confirmed by a pilot.
Answering recurring questions
Making the quizzes
Spotting the topics to consolidate
Illustrative, non-contractual reference points, to be confirmed by a pilot on the number of modules and staff. Quizzes are reviewed before circulation and the results stay aggregated: no individual assessment is produced by the agent.
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 training agent (tutor, quizzes, points to consolidate), installed and operated for you. Prices exclude VAT — annual subscription, the time it takes for the gains to settle in.
Agility
Setup + controlled subscription
5,065 € excl. VAT setup
then 541 € 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
821 € 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
8,400 € excl. VAT setup
then 741 € 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 training
Your materials and the progress stay internalLocal inference or an isolated resource hosted in France; no material or result entrusted to a third party, and no data used to train a model.
Data in France, under French lawYour staff's progress: minimisation and location in France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The training manager keeps the decisionThe agent produces practice paths and quizzes, verifiable and editable; no approval is automated.
Human oversight & traceabilityOn the number of modules and staff: systematic logging and monitoring, compliant with the AI Act.
Going further
Related resources
Frequently asked questions
Your questions, our answers
Does the agent grade the staff?
No. The answers feed points to consolidate at module level. Assessing a person belongs to management and to dialogue, and the European AI regulation in fact classifies that kind of use among the high-risk ones.
Where do the quiz questions come from?
From your own materials, with a pointer to the paragraph that grounds each expected answer. The training manager reviews the quiz before circulation.
What happens if a procedure changes?
The agent relies on the version in force and cites its update date, so an out-of-date answer is spotted immediately.
Are our internal materials 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 materials are not used to train a third-party model.
Does it integrate with our intranet or our learning platform?
Yes, with the common tools. Your staff follow their course from Microsoft Teams, Slack or their email, without leaving the tool they already work in. 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. The integration is defined during the design phase.
How long does it take to deploy this agent?
A few weeks as a rule, depending on the number of modules and the state of your materials, after a free audit and then a phase of design, integration and testing.
Related use cases
Other agents for your teams
Let's talk
Let us estimate the potential for your internal training
15 minutes to scope your modules and your materials — hosted in France, supervised, with no commitment.