+33 (0)1 87 66 00 65 · Monday to Friday, 9am–6pm Free audit (15 min)
● Public sector — Education

The teachers' AI agent: lighten the preparation, give time back to teaching

Preparing materials, differentiating exercises, marking, writing reports and letters: these tasks absorb a considerable share of teachers' time — at the expense of what matters, the teaching relationship with pupils. Your AI agent absorbs that repetitive work. Hosted in France — on local inference or an isolated resource — pupils' data stays under control. The AI agent assists, the teacher decides.

Hosted in France Pupils' data protected GDPR & AI Act: governed deployment Human oversight

Updated on

Deployed in a few weeks
Teaching assistant · hosted in France
Prepare a three-lesson sequence on fractions for my Year 7 class, with differentiated exercises for pupils who are struggling and for the most advanced.
Sequence proposed: 3 lessons with objectives, running order and materials ready to print.
Differentiation: a scaffolded path (manipulation, guided diagrams) and an extension path (open problems). The exercises follow your schemes of work for the year.
For you to review and adjust to your class.
⛓ Source · your schemes of work + the official curriculum for the key stage
Can you help me write the report comments from my tracking notes?
Draft comments generated for the 28 pupils, from your observations and results: one draft per pupil, factual and supportive.
The assessment and the final wording stay yours — you correct, qualify and approve every comment.
✎ Action · drafts ready for review — the teacher approves
Local inference · no data outside the EU
Data 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 teacher decidesThe agent assists, it never assesses alone
✦ In brief

For teachers, a Blue Lemon Agent agent lightens the repetitive tasks — preparing materials, sequences and differentiated exercises, help with marking, letters and reports — to give time back to the teaching relationship. It runs on local inference or is hosted in France: pupils' data is never exposed to a foreign service, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. The agent stays strictly in support: no automated decision about a pupil. Live within a few weeks. Teachers and the administrative team write to it from Microsoft Teams, Slack or their email, and families reach it on WhatsApp Business, the website chat or email — with no account to create and nothing to install. Reaching the school from the tool people already have means less non-take-up of rights and equal access to the service. These connections are included in every plan, at no extra cost, within the number of connections your level includes.

100%
hosted in France in the target architecture
0
transfer outside the EU in the target architecture
9
uses ready to deploy on this scope
0
decision taken without human approval

Reference points describing our offer, not results measured at a client. The scale of the gain is confirmed by a pilot on your own scope.

The context

Why AI matters to schools — and why it worries them

Teachers want time for their pupils, but preparation, marking and administration take up a large part of it. Consumer AI promises relief — at the cost of pupils' data entrusted to foreign services, which is unacceptable for a public education service.

! The issue

The teacher is caught between the duty of continuity and equal treatment for every pupil, and a load of preparation, marking and administrative tasks that keeps growing. Yet most consumer AI tools amount to entrusting pupils' data, their work, observations and information about families to a third party, often hosted outside Europe and subject to the Cloud Act.

Our answer

AI is only of interest to schools if it is sovereign and confidential by design. Local inference or an isolated resource hosted in France, systematic human oversight, decisions reserved to the teacher: the time saved on preparation is never paid for in lost confidentiality, nor in automated decisions about pupils. The aim is not to replace the teacher, but to give them back time for their pupils.

The decisive point

Protecting pupils' data: sovereignty & compliance

A class means personal data of minors and families, among the most sensitive there is. Here is how the architecture of our agents protects it, lesson after lesson.

Local inference

The agent can run on a resource at the school: no pupil's work or data leaves the network, nothing passes through a cloud.

Hosting in France

Otherwise, a dedicated and isolated resource, hosted in France under French law — pupils' data: processing and access within the European Union targeted by the architecture.

Reduced extraterritorial exposure

Exposure of pupils' data to the Cloud Act and FISA 702 is reduced by design; location alone does not guarantee immunity.

One isolated resource per establishment

No pooling of pupil data: an environment strictly dedicated to your school.

Encryption & controlled access

Encryption in transit and at rest, role-based access (RBAC), strong authentication and logging.

AI Act: governed deployment

The agent is strictly in support; no automated decision about a pupil; 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.
  • The encryption mechanisms in transit and at rest, their components and key management are those documented for the architecture chosen.
  • Roles and permissions are configured and accepted for the identities and systems actually connected.
  • The events logged, their content, their retention period and who may access them are defined for the deployment chosen.
For the most sensitive data concerning minors, SecNumCloud and reinforced hosting options are available according to how demanding your requirements are. 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.

The public body in this demonstration

Fictional public body

Collège des Quatre-Vents — state lower-secondary school with 486 pupils (fictional school)

Sector
State lower-secondary school covering the four years to age 15 — 18 forms, 11 subject teams, a reception unit for newly arrived pupils with little French, 36 teaching weeks a year
Headcount
41 teachers — 18 of them form tutors —, 4 pastoral staff and a two-person office; the 41 teachers are the users of the agent
Public served
486 pupils and 431 families; 12 pupils with little French received this year; 5 languages other than French declared at enrolment over the last three years
Order of magnitude
738 teaching sequences prepared a year — 18 per teacher; 540 sets of report comments a year (18 forms × 10 subjects × 3 terms), that is 14,580 individual comments; 54 form review boards; 9,400 messages from families received in the year
Tools in place
The school's digital workspace, the pupil-management software (lesson log, marks, absences), the 11 team schemes of work, the office's letter templates and Microsoft Teams for internal exchanges — the agent plugs in read-only, nothing is replaced and nothing is migrated
Who decides
Each teacher validates, corrects or discards anything that carries their name — no comment goes out without them; the form tutor settles the report text; the review board remains sovereign; the head teacher settles what is published to families and signs the school's letters
Points for improvement
A differentiated three-lesson sequence takes 5 hours, and only 246 of the year's 738 sequences actually carry both pathways; 1 h 15 per form per term on writing report comments alone, that is 540 sets in the year; 6,400 of the 9,400 messages from families are about 11 recurring questions; and 62 of the 431 families never open the digital workspace

At Collège des Quatre-Vents, 41 teachers prepare 738 sequences a year, write 14,580 report comments and receive 9,400 messages from families, with no extra staff. It is not the teaching that eats the time, it is everything around it: writing the same exercise twice at two levels, hunting for the homework where an observation was noted, answering the question about pick-up times for the eleventh time. The exchanges below span a school year, from reading the lesson log to the report presented to the governing board. The agent runs on local inference on a machine in the school, reads the lesson log, the schemes of work and the marks entered without ever writing to them, and no output concerning a pupil exists before the teacher has signed it.

This public body, its figures and the exchanges that follow were invented for the demonstration. They illustrate a common situation; they describe no real service.

Teaching assistant · hosted in France
I have read the 738 sequences in your lesson log, your 11 team schemes of work and the 9,400 messages families wrote to you this year — and three figures stand out that nobody had had the time to put side by side.
What the count says:
· 246 of the 738 sequences actually carry both pathways — one in three. The other 492 were prepared with no written support or extension route, and that is not a teaching choice: a differentiated three-lesson sequence takes 5 hours, and those five hours do not exist three times a week.
· 6,400 of the 9,400 messages from families are about 11 recurring questions — timings and pick-up, canteen, absences and notes, equipment, school trips, appointments with a teacher, access to the digital workspace, grants, transport, the year-10 work placement, the report and board calendar. Two messages in three already have their answer written somewhere in your school rules or your staff notices.
· 3 of your 11 team schemes of work still carry last year's ordering, although the teams changed it at the start-of-year meeting on 2 September — the minutes are in your digital workspace. Those 3 schemes govern 214 sequences of the year.
What I bring you this morning, already written: the 3 schemes are rewritten in their 2 September version, with the replaced line and the minute line that replaces it set side by side; the 11 answers to families are drafted, sourced on your school rules and staff notices, each with its date; and a complete three-lesson sequence on fractions for the first year, with its support pathway and its extension pathway, is waiting for you.
The gain, costed on your own volumes: preparing a differentiated sequence goes from 100 % to 18 % of the time it takes you today — from 5 hours to 54 minutes, and those 54 minutes are reading and adapting to your class, not writing.
The next step, and it takes fifty-four minutes: you read the fractions sequence, you correct what does not suit your class, and it is used on Monday morning. preparing-the-year_738-sequences-11-questions.pdf246 differentiated sequences out of 738, 6,400 messages on 11 questions
⛓ Sourced · lesson log for the last 3 years, 11 team schemes of work, minutes of the start-of-year meeting of 2 September, 9,400 messages from families
There are 41 of us, and nobody has an hour to give to yet another tool. Where do we start without turning this into a whole-school project?
With three subject teams out of eleven — and I have ranked them by the number of sequences they carry, not alphabetically.
What the count of your lesson log says:
· Mathematics: 168 sequences a year, 7 teachersand it is one of the 3 schemes that still carried the old ordering.
· French: 154 sequences, 6 teachers.
· History and geography: 132 sequences, 5 teachers.
Those three teams account for 454 of the 738 sequences: three teams out of eleven, 61 % of the school's preparation. The other eight open in January, two per fortnight, with nothing left waiting.
And I would advise against opening all eleven at once, with the figures behind it: the first three teams need three thirty-minute meetings — one per team, to settle the reference scheme of work. All eleven need eleven, that is five and a half hours to fit into timetables that do not overlapthree weeks' delay to going live for 39 % more sequences. Three teams in September, eight in January: the first sequence is used on Monday, and the whole year is covered without a single meeting being added to your calendar.
What those three meetings cost you, and what they give back: thirty minutes per team to settle the reference scheme of work and to tell me what you allow me to reuse from the last three years of the lesson log. Walking out, each teacher has in front of them the list of their 18 sequences for the year, each with the number of hours it took them last year and what it will take this time — 5 hours against 54 minutes.
What that will give over the year, and it is the figure I will defend in front of you in June: 612 of the 738 sequences prepared with me454 for the three September teams, 158 of the 284 sequences of the eight teams opened in January.
The next step I propose: that the three thirty-minute meetings be put in the diary this week, and that I compare your 11 schemes of work with the minutes of the latest team meeting every night. Your schemes will stop ageing in silence, and the September read-through will not have to be done again: it becomes a five-minute validation at each team meeting. preparing-the-year_738-sequences-11-questions.pdf454 sequences for 3 teams, 612 over the year, opening order
⛓ Sourced · count of sequences by subject team over 12 months, 11 schemes of work, timetables of the 41 teachers
All of that assumes you read my lesson log, my tracking notes, my pupils' marks and the families' messages. These are minors. Where does it all go?
Nothing goes anywhere. I run on local inference on a machine in the school, and what I read does not leave your network.
Local inference means the model computes on your machine: the text of a piece of work, an observation recorded about a pupil or a family's message never crosses an outside network to be processed. If the school would rather not administer a machine, the other route is an isolated resource hosted in France, dedicated to Collège des Quatre-Vents, shared with no other school.
What that changes, point by point:
· Pupil data trains no model, neither ours nor a third party's.
· I work read-only, and the technical account I read through has no right to write into your pupil-management softwarethat can be checked with one command, which is worth more than a promise.
· Encryption in transit and at rest, role-based accessrights follow the role: a teacher opens the forms that are theirs, and nothing else. 41 roles for your 41 teachers, and the log shows 0 out-of-role access since go-live.
· Hosting in France, under French law, architecture designed to reduce exposure to extraterritorial legislation, including against an American operator hosting in Europe — location alone not being enough to guarantee immunity, the exposure is documented for the configuration chosen.
· A complete log: who asked what, when, and what the system produced.
And on decisions, this is not caution, it is law: a decision producing legal effects concerning a person cannot be based solely on automated processing — that is Article 22 of the General Data Protection Regulation. A mark, a progression to the next year, a guidance decision are decisions; they remain those of the teacher and of the review board, and that is exactly what makes them reasoned and contestable in front of a family. The route that gives you the result without giving anything up is the one we are taking: I prepare, I cross-check, I set the entries alongside, and you sign in fifteen seconds instead of writing for fifteen minutes.
What I can do beyond that, the day the school decides to: having an AI system assess learning outcomes falls under the education heading of Annex III of the European artificial intelligence regulation — high risk, and not prohibited. So it is a compliance programme, not a wall: technical documentation, logging, named human oversight, information of the persons concerned. It is written and I put it on the table the day you want to go there; as long as the agent stays in a supporting role, you do not need it.
And the pupil, like the family, knows who they are talking to: I introduce myself as a digital assistant of the school in the first sentence, because the European artificial intelligence regulation requires that anyone interacting with an AI system be informed — and a teacher can be called in at any moment, in which case I pass the request on with the subject already noted.
The figure that sums all of this up: 0 pupil data out of the school network over the year, and 0 transfer outside the European Union.
What I propose: that I keep up to date the record the academy's data protection officer will ask you for — hosting, data processed, retention periods, who accesses what, and the mention of local inference. The first version is written and attached; it updates itself each time a new use is opened. technical-framework_where-pupil-data-lives.pdfLocal inference, read-only, 0 transfer outside the EU
✎ Framework · deployment architecture, technical account rights, first version of the processing record, Article 22 GDPR
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

The uses of AI in support of teachers

Each use corresponds to an agent we deploy. All work in support, subject to your approval, with no automated decision about pupils.

Included in your agent The 6 capabilities essential to this promise are included, at no extra cost.
From 690 € incl. VAT / month

Preparing materials & sequences

Generation of sequences, worksheets and lesson materials aligned with the curriculum and your schemes of work — for review and adaptation.

Differentiated exercises

Production of exercises at several levels (scaffolding and extension) to answer the range within the class.

Help with marking & feedback

Draft feedback and comments from your observations; the teacher assesses, qualifies and approves.

Letters & reports

Drafts letters to families, class council meeting minutes and notes using your templates.

Plain language & easy-read draft

Preparation of a plain-language version and a draft easy-read (FALC) transcription, subject to human validation.

Answers to families

Answering parents' recurring questions: organisation, procedures, practical information about the school.

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
Other needs our agents cover Each card says where the matching agent stands: available, on quote, or still being architected.

Organisation & diary

Preparing class meetings, scheduling meetings with parents and tracking administrative deadlines.

On quote View the agent page
Does your need fall outside this?

In 15 minutes we identify the agent that will give your staff the most time back — without oversizing the project.

Book the free audit Build your agent
The gain

How much time can a teacher win back?

By lightening the preparation of materials and the production of feedback, a teacher can aim for a noticeable reduction in the time spent on repetitive tasks — reinvested in the teaching relationship and in supporting pupils.

Preparing a differentiated sequence
Today · done by hand
Prepared by the agent, to approve
Writing the comments for a class
Today · done by hand
Prepared by the agent, to approve
Answering a parent's recurring question
Today · done by hand
Automatic
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

An agent in support of teaching teams

A teaching agent (preparation, help with marking, administration), installed and operated for you. Choose according to how the school is organised. Prices exclude VAT — annual subscription, the time it takes for the gains to settle in.

Agility

Setup + controlled subscription

7,420 € incl. VAT setup
then 690 € 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
Order →
The simplest Serenity

All inclusive, no setup fee

1,100 € 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
Order →
100% Sovereign

On site, you own it

11,666 € incl. VAT setup
then 930 € 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)
Order →
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 a school

Pupils' data never leavesLocal inference or an isolated resource hosted in France; no pupil's work or data entrusted to a foreign third party.
Data in France, under French lawNative location and minimisation for pupils' data; architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The teacher keeps the decisionThe agent produces materials, exercises and drafts that can be checked; no assessment or decision about a pupil is automated.
Human oversight & traceabilityMonitoring and logging frame the preparation of materials & sequences; compliant with the requirements of the AI Act.
Frequently asked questions

Your questions, our answers

Does the agent assess the pupils?
No. It assists the teacher: it prepares materials, exercises and draft feedback, but the assessment, the marking and the teaching stay entirely with the teacher. No decision about a pupil is automated.
Is pupils' data protected?
Yes. Data about pupils, their work and their families is sensitive, all the more so as they are often minors: our agents run locally within the school or are hosted in France, 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.
Does it fit the framework of the state education service?
The use is strictly in support, with no automated decision about pupils. The teacher reviews and approves every piece of content produced; human oversight and traceability are systematic, in line with the requirements of the AI Act.
Can AI really help prepare lessons?
Yes, to a large extent. An AI agent generates sequences, materials and differentiated exercises aligned with the curriculum and your schemes of work. The teacher moves from writing to reviewing and adapting — far quicker — and keeps control of the teaching choices.
Do we have to change tools or digital environment?
No. The agent complements your existing tools (learning platform, workspaces, document templates) with no migration imposed. We adapt the integration to your school's digital environment.
How long does it take to deploy an agent?
A few weeks as a rule, after a free audit that identifies the use most helpful to the teams, then a phase of design, integration and testing before going live and handing over to the teachers.
Which tools can families use to reach the agent?
The ones they already have. The agent answers on WhatsApp Business, the website chat and email: a family has no account to create and no application to install. Exchanges cover organisation and procedures — timetables, outings, appointments — and never a pupil's progress, which goes through the school's own channels and through the teacher. This is a lever for access — WhatsApp and the telephone reach families an online portal never does, which serves equal access. Internally, teachers and the administrative team talk to the agent from Microsoft Teams, Slack or their email, without switching tools. Oversight runs from a web dashboard. These connectors rely on open standards, including the MCP protocol; they are included in every plan, at no extra cost, within the number of connections your level includes. Only the fees charged by the platforms themselves — WhatsApp Business bills per conversation — are passed on at actual cost, with no margin, outside the subscription.
Let's talk

Let's size up the potential in your school

15 minutes to identify the use most helpful to teachers — hosted in France, supervised, with no commitment.