● Public sector offer — Recruitment
Recruitment support: your statutory criteria applied to every application
A public-service recruitment notice sets out conditions that can be verified: employment status, qualification required, length of experience, authorisation. Your agent examines each application against those conditions and states, for each one, what the file explicitly mentions. Where you have set objective criteria, including statutory conditions of access, it applies them and presents the applications sorted accordingly. Hosted in France: candidates' data stays within the administration. The panel and the department decide.
Hosted in France
Candidate data protected
GDPR & AI Act: governed deployment
Human oversight
Updated on
Deployed in a few weeks
Recruitment support · hosted in France
Examine the applications received against this recruitment notice.
Conditions taken from the notice: employment status, qualification required, length of experience, authorisation where applicable.
For each application, what the file explicitly mentions on each of those conditions.
The statutory conditions of access are applied and the resulting split is presented to you.
🔗 Sourced · the notice's conditions and the documents in the files
For each application, what the file explicitly mentions on each of those conditions.
The statutory conditions of access are applied and the resulting split is presented to you.
🔗 Sourced · the notice's conditions and the documents in the files
Can you rank the applications by merit?
I set out what each file establishes, condition by condition, with the wording to back it up.
Assessing merit rests with the panel and the department: it is a decision about people, which the European AI Act classes as high-risk.
✎ Support · material presented, the panel's judgement
Assessing merit rests with the panel and the department: it is a decision about people, which the European AI Act classes as high-risk.
✎ Support · material presented, the panel's judgement
Local inference · no data outside the EU
Applications hosted in France
Sovereign by designLocal inference or hosting in France
GDPR & AI Act: governed deploymentTraceability & human oversight
TurnkeyDesigned, installed and operated for you
Your recruitment department decidesThe agent prepares, never rules
✦ In brief
A Blue Lemon Agent recruitment support agent takes the conditions from your notice, states for each application what the file explicitly mentions and applies the statutory conditions of access you have set. A point not mentioned is flagged, never inferred; no assessment of merit is produced. It runs on local inference or is hosted in France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
100%
hosted in France in the target architecture
0
transfer outside the EU in the target architecture
6
recruitment uses ready to deploy
0
decision taken without human approval
These figures describe our offer, not results measured at a client. How large the gain is on your volume of applications and number of notices published is confirmed by a pilot.
The context
What does an AI agent bring to examining your applications?
Checking the conditions of access across every file, consistently and traceably, serves the candidates as much as the department.
! The issue
A recruitment notice sets out conditions that are verifiable and can be relied upon. Checking them on every file calls for methodical, steady reading; that is exactly what an agent does consistently, returning for each application what the file mentions, with the document to back it up.
✓ Our answer
The recruitment department has a consistent, traceable examination of the conditions of access, applied to every application. Assessing merit and deciding to appoint rest with the panel and the department: the European AI Act classes recruitment among the high-risk uses, and a decision about a person calls for a human eye. Local inference or an isolated resource hosted in France.
The decisive point
Candidates' personal data: sovereignty & compliance
Applications are personal data submitted for a particular procedure. Here is how they are protected.
Local inference
The agent can run on a machine belonging to your organisation: no application and no candidate data leaves the network.
Hosting in France
Otherwise, a dedicated and isolated resource hosted in France, under French law — your recruitment notices and your applications: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
For candidates' personal data, the architecture aims to reduce exposure to the Cloud Act and FISA 702; being located in France or in the European Union does not, on its own, guarantee immunity.
Isolated resource
No pooling: an environment strictly dedicated to your administration and its recruitment notices.
Every finding refers back to a document in the file
A condition not mentioned is flagged as such rather than inferred; encryption, role-based access, logging and purging at the retention deadline you set.
AI Act: governed deployment
The agent is strictly in support; no application is set aside, no merit is assessed and no appointment decision is taken; 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 recruitment to sensitive posts and applications involving a protected status, SecNumCloud and reinforced hosting are options depending on your requirements. A single architecture is designed to answer both the GDPR and extraterritorial exposure. Designed for deployment in line with the GDPR and the AI Act, after the processing, roles and context-specific risks have been assessed.
Demonstration
See the agent at work
4 real situations, taken from those that come up most often. Pick one: the exchange unfolds as it would in your organisation.
A scripted demonstration. These exchanges show how the agent behaves — its sources, its refusals, what it leaves to your teams. Nothing is sent from this page, no model is queried here, and the matters named are fictional. That is precisely what we promise your data.
The behaviours shown here — monitoring, automation rules, routing and reminders — are configured with you during deployment, from your tools, your rules and your thresholds.
The architecture points named in these exchanges — location, local execution, isolation, encryption, role-based access, logging — are not a guarantee attached to the demonstration: they are those of the architecture set out in your quotation, and verified before commissioning.
Recruitment support · hosted in France
Four flags this morning on your live recruitments.
· One criterion in your vacancy appears in none of the 47 applications received. It is not that nobody has it — it is that the advert does not ask for it to be stated.
· Eleven applications are incomplete, and nine are missing the same document. That document is only asked for at the end of the advert, after the form.
· The closing date of another recruitment falls in 4 days and the panel is not constituted.
· A vacancy published in June is still online although the post was filled on 21 July. Six applications have arrived since. morning-watch_4-flags.pdf4 flags · 6 wasted applications
⛓ Source · 3 published vacancies, 47 + 22 applications received, publication dates
· One criterion in your vacancy appears in none of the 47 applications received. It is not that nobody has it — it is that the advert does not ask for it to be stated.
· Eleven applications are incomplete, and nine are missing the same document. That document is only asked for at the end of the advert, after the form.
· The closing date of another recruitment falls in 4 days and the panel is not constituted.
· A vacancy published in June is still online although the post was filled on 21 July. Six applications have arrived since. morning-watch_4-flags.pdf4 flags · 6 wasted applications
⛓ Source · 3 published vacancies, 47 + 22 applications received, publication dates
The six applications that arrived afterwards — what do we do?
We answer them, and it is the only one of the four that engages the authority towards people.
What I record: the post was filled on 21/07, the vacancy stayed online, six people applied between 22/07 and today. They prepared a file for a post that no longer existed.
What I propose: take the vacancy down — it is urgent, a seventh may arrive today — and write to the six. The letter is prepared; it says what happened and on what date, with no wording implying an assessment that never took place.
What I have written, and what is left to sign: the letter to the six is drafted, addressed by name, dated; it says what happened and when, with no wording implying an assessment that never took place. Sending it is signed by the authority — a reply to an application comes from it, and this one contains an admission: the vacancy stayed online too long. How to say that is yours to choose; everything else is done, and the six are informed within the hour of your go-ahead.
What I propose next: flag any vacancy still online 48 hours after a post has been filled. Over the last twelve months, this happened four times and affected 19 applications. vacancy-left-online_6-applications.pdf6 people · letter prepared · 4 cases in 12 months
⛓ Source · date the post was filled, publication log, 6 later applications
What I record: the post was filled on 21/07, the vacancy stayed online, six people applied between 22/07 and today. They prepared a file for a post that no longer existed.
What I propose: take the vacancy down — it is urgent, a seventh may arrive today — and write to the six. The letter is prepared; it says what happened and on what date, with no wording implying an assessment that never took place.
What I have written, and what is left to sign: the letter to the six is drafted, addressed by name, dated; it says what happened and when, with no wording implying an assessment that never took place. Sending it is signed by the authority — a reply to an application comes from it, and this one contains an admission: the vacancy stayed online too long. How to say that is yours to choose; everything else is done, and the six are informed within the hour of your go-ahead.
What I propose next: flag any vacancy still online 48 hours after a post has been filled. Over the last twelve months, this happened four times and affected 19 applications. vacancy-left-online_6-applications.pdf6 people · letter prepared · 4 cases in 12 months
⛓ Source · date the post was filled, publication log, 6 later applications
How did you spot all that? And who do you tell?
I watch, continuously, what you have opened to me: your published vacancies, the applications received, the documents asked for and those that arrive, and the dates you set yourselves.
Routing follows what is at stake: a vacancy left online goes to recruitment immediately, because people are working for nothing; an unconstituted panel to the manager, 4 days out; the incomplete applications to recruitment with the shared cause, not one by one; the missing criterion to whoever wrote the advert.
With a chase: 24 h on the online vacancy, 7 days on the rest. Then a monthly summary: by vacancy and by missing document, never by candidate.
What this morning has already produced: 47 applications examined against the conditions of your notice — status, required qualification, length of experience, clearance —, every finding traced back to the line in the file that grounds it; 11 incomplete applications brought back to their shared cause, a document asked for after the form; a vacancy taken down the same day and six candidates who get an answer; a panel to constitute four days out.
What that gives the service from tomorrow: the panel opens the session with the 47 files already read and sorted by your statutory criteria, each with the line of the file that places it. It no longer rereads to sort: it chooses. Every decision carries an officer's name, and I hand it back to them fully worked in minutes: the criterion applied, the entry that grounds it, and the proof that the same criterion was applied to the other 46. That is equal treatment, documented application by application — and it is the first thing an appeal looks at. Nothing is inferred from a name, an address, a photograph or a date of birth, and candidates' data never leaves the administration: I reach what you open, role by role, and every read is logged.
The next step is ready: the letter to the six is drafted, the 11 incomplete applications each have their named document request, and moving that document ahead of the form removes the cause for nine of them. Tell me who signs, and the four-day recruitment starts moving today.
✎ Framework · the three refusals hold whatever the settings
Routing follows what is at stake: a vacancy left online goes to recruitment immediately, because people are working for nothing; an unconstituted panel to the manager, 4 days out; the incomplete applications to recruitment with the shared cause, not one by one; the missing criterion to whoever wrote the advert.
With a chase: 24 h on the online vacancy, 7 days on the rest. Then a monthly summary: by vacancy and by missing document, never by candidate.
What this morning has already produced: 47 applications examined against the conditions of your notice — status, required qualification, length of experience, clearance —, every finding traced back to the line in the file that grounds it; 11 incomplete applications brought back to their shared cause, a document asked for after the form; a vacancy taken down the same day and six candidates who get an answer; a panel to constitute four days out.
What that gives the service from tomorrow: the panel opens the session with the 47 files already read and sorted by your statutory criteria, each with the line of the file that places it. It no longer rereads to sort: it chooses. Every decision carries an officer's name, and I hand it back to them fully worked in minutes: the criterion applied, the entry that grounds it, and the proof that the same criterion was applied to the other 46. That is equal treatment, documented application by application — and it is the first thing an appeal looks at. Nothing is inferred from a name, an address, a photograph or a date of birth, and candidates' data never leaves the administration: I reach what you open, role by role, and every read is logged.
The next step is ready: the letter to the six is drafted, the 11 incomplete applications each have their named document request, and moving that document ahead of the form removes the cause for nine of them. Tell me who signs, and the four-day recruitment starts moving today.
✎ Framework · the three refusals hold whatever the settings
Concretely, how do you help me with 47 applications?
I record what your vacancy asks for, and I tell you where I read it.
What your vacancy asks for explicitly: a qualification at a stated level, project-management experience, command of a named piece of software, and a driving licence. Four factual elements, stated in the advert.
What I produce: for each application, those four elements with the page and line where I found them — or the words "not found", which mean what they say: I did not find it, not that the person does not have it.
What that changes for the panel: it opens 47 files knowing where to look, instead of hunting four pieces of information across forty-seven different layouts. In a dry run this cut first-reading time by a factor of three — and the panel read all 47, as before.
What I apply on top, and these are your statutory conditions, exclusions included: the 47 applications are sorted against the vacancy's four conditions — 31 meet them all, 9 miss one, 7 mention none — and each placement points back to the line of the file that grounds it. The 16 files set aside stay one click away, with the condition that put them there: a condition misread is recovered in ten seconds.
What stays with the panel, and here it is the law that says so: the appraisal of merit. A recruitment decision produces legal effects for the person, and article 22 of the GDPR gives everyone the right not to be subject to a decision based solely on automated processing. A merit ranking produces that decision, even if a human then approves it without reopening it — a sort by verifiable conditions does not: it records what the vacancy requires and what the file states. 4-elements_with-the-line.pdf4 elements · 47 files · no ranking
⛓ Source · vacancy published 02/07, 47 applications, art. 22 GDPR
What your vacancy asks for explicitly: a qualification at a stated level, project-management experience, command of a named piece of software, and a driving licence. Four factual elements, stated in the advert.
What I produce: for each application, those four elements with the page and line where I found them — or the words "not found", which mean what they say: I did not find it, not that the person does not have it.
What that changes for the panel: it opens 47 files knowing where to look, instead of hunting four pieces of information across forty-seven different layouts. In a dry run this cut first-reading time by a factor of three — and the panel read all 47, as before.
What I apply on top, and these are your statutory conditions, exclusions included: the 47 applications are sorted against the vacancy's four conditions — 31 meet them all, 9 miss one, 7 mention none — and each placement points back to the line of the file that grounds it. The 16 files set aside stay one click away, with the condition that put them there: a condition misread is recovered in ten seconds.
What stays with the panel, and here it is the law that says so: the appraisal of merit. A recruitment decision produces legal effects for the person, and article 22 of the GDPR gives everyone the right not to be subject to a decision based solely on automated processing. A merit ranking produces that decision, even if a human then approves it without reopening it — a sort by verifiable conditions does not: it records what the vacancy requires and what the file states. 4-elements_with-the-line.pdf4 elements · 47 files · no ranking
⛓ Source · vacancy published 02/07, 47 applications, art. 22 GDPR
And what exactly does "not found" mean?
That I did not find it. Nothing more, and the distinction is the whole subject.
Across the 47 files: the driving licence is "not found" 21 times. That does not mean 21 people lack a licence — it means 21 files do not mention one.
Why it is worded that way: if I wrote "no licence", the panel would read a fact. It would read a false fact in most cases, and the person would be set aside over information they were never asked to supply.
What it reveals about your vacancy: the licence is in the requirements, but the advert nowhere asks for it to be stated, and the form has no field for it. 21 "not found" out of 47 do not describe the candidates: they describe the advert.
What I propose: a question on the form, for future recruitments. And for this one, a group follow-up to the 21 — thirty seconds each, and it puts 21 files in the same position as the other 26. 21-not-found_what-it-says.pdf21 files · the advert at fault
✎ Proposal · a question to add to the form — 21 follow-ups prepared
Across the 47 files: the driving licence is "not found" 21 times. That does not mean 21 people lack a licence — it means 21 files do not mention one.
Why it is worded that way: if I wrote "no licence", the panel would read a fact. It would read a false fact in most cases, and the person would be set aside over information they were never asked to supply.
What it reveals about your vacancy: the licence is in the requirements, but the advert nowhere asks for it to be stated, and the form has no field for it. 21 "not found" out of 47 do not describe the candidates: they describe the advert.
What I propose: a question on the form, for future recruitments. And for this one, a group follow-up to the 21 — thirty seconds each, and it puts 21 files in the same position as the other 26. 21-not-found_what-it-says.pdf21 files · the advert at fault
✎ Proposal · a question to add to the form — 21 follow-ups prepared
Do you spot the incomplete applications?
I spot the ones where a supporting document is missing — and I tell you where what I do stops.
What I found: 11 incomplete applications out of 47, and 9 are missing the same supporting document: the qualification certificate, requested at the end of the advert, after the form. That is not a failing of the candidates, it is a placement failing in the advert, and it is fixed by moving the list of supporting documents above the form. I offer you the wording, you publish it.
What I do on supporting documents: I list those your advert requires, I state for each application which are present and which one is missing, by name, and I draft the message asking for it with the closing date. No application is set aside for a missing document: it is flagged to the department, which decides whether to chase — a candidate excluded without ever being asked for their document would have nothing to reproach themselves with, and neither would you.
Where what I do stops, and who takes over: I check the presence of a supporting document and that its title matches what the advert asks for. Completeness checking in the proper sense — reading the document, checking its validity date, its issuer, its conformity to the expected template — belongs to an agent dedicated to reading supporting documents, which complements this one. I would rather tell you here than let you discover it on a live file.
What that gives back: checking the supporting documents of a 47-application campaign took 2 h 20; it now takes 15 minutes. Across 34 campaigns a year, more than 70 hours given back, rounded down.
The figure that does not flatter me: of my first 200 checks, 17 documents that were present were counted missing — 8.5 %: they were attached inside the same file as the CV, on pages 3 and 4, and I was looking for a separate file. Seventeen candidates were chased for a document they had sent. I now read every file to the end and tie the document to the page it sits on: over the next 600 checks, no unwarranted chaser.
⛓ Sourced · 11 incomplete applications of 47, 9 for the same document, 2 h 20 down to 15 min, 17 unwarranted chasers fixed
What I found: 11 incomplete applications out of 47, and 9 are missing the same supporting document: the qualification certificate, requested at the end of the advert, after the form. That is not a failing of the candidates, it is a placement failing in the advert, and it is fixed by moving the list of supporting documents above the form. I offer you the wording, you publish it.
What I do on supporting documents: I list those your advert requires, I state for each application which are present and which one is missing, by name, and I draft the message asking for it with the closing date. No application is set aside for a missing document: it is flagged to the department, which decides whether to chase — a candidate excluded without ever being asked for their document would have nothing to reproach themselves with, and neither would you.
Where what I do stops, and who takes over: I check the presence of a supporting document and that its title matches what the advert asks for. Completeness checking in the proper sense — reading the document, checking its validity date, its issuer, its conformity to the expected template — belongs to an agent dedicated to reading supporting documents, which complements this one. I would rather tell you here than let you discover it on a live file.
What that gives back: checking the supporting documents of a 47-application campaign took 2 h 20; it now takes 15 minutes. Across 34 campaigns a year, more than 70 hours given back, rounded down.
The figure that does not flatter me: of my first 200 checks, 17 documents that were present were counted missing — 8.5 %: they were attached inside the same file as the CV, on pages 3 and 4, and I was looking for a separate file. Seventeen candidates were chased for a document they had sent. I now read every file to the end and tie the document to the page it sits on: over the next 600 checks, no unwarranted chaser.
⛓ Sourced · 11 incomplete applications of 47, 9 for the same document, 2 h 20 down to 15 min, 17 unwarranted chasers fixed
Could you help me spot the best profiles?
No, and I would rather explain why than offer you a softened version.
What I am often asked for: a score, an order, a shortlist, or "just" flagging the obviously irrelevant files. All four come to the same thing — they decide which files get read attentively, and that decision is the recruitment decision.
What happens with a score, even advisory: the panel reads the best-scored first. It reads the last ones with the attention given to what it already believes is poor, and the order confirms itself. Nobody took an automated decision, and yet the machine chose.
What I never look at, whatever the settings: surname, first name, address, photograph, date of birth, nationality, family circumstances, institution attended. Those fields are not weighted to zero: they are not read.
Why the institution attended is on that list: because it is the best known proxy for social origin, and a tool reading it would discriminate without a single line of its configuration saying so.
What I do instead: I save you reading time, never judgement time. 8-fields_never-read.pdf8 fields · no weighting · an absence
✎ Framework · eight fields never read — not a zero weighting
What I am often asked for: a score, an order, a shortlist, or "just" flagging the obviously irrelevant files. All four come to the same thing — they decide which files get read attentively, and that decision is the recruitment decision.
What happens with a score, even advisory: the panel reads the best-scored first. It reads the last ones with the attention given to what it already believes is poor, and the order confirms itself. Nobody took an automated decision, and yet the machine chose.
What I never look at, whatever the settings: surname, first name, address, photograph, date of birth, nationality, family circumstances, institution attended. Those fields are not weighted to zero: they are not read.
Why the institution attended is on that list: because it is the best known proxy for social origin, and a tool reading it would discriminate without a single line of its configuration saying so.
What I do instead: I save you reading time, never judgement time. 8-fields_never-read.pdf8 fields · no weighting · an absence
✎ Framework · eight fields never read — not a zero weighting
The panel unconstituted 4 days out — can you do anything?
I can tell you what is blocking, and this time it is not an oversight.
What I record: the panel must have three members under your own resolution, two are appointed, and the third must hold a particular capacity that your resolution specifies.
What I looked for: people who have sat in that capacity on your previous panels. I find four, with the date of the panel each sat on. Three are still in post.
What I hand you, and what is left for you to do: the four people who have already sat in that capacity, the date of the panel each sat on, the three still in post, the resolution setting the rule, and the availability three of them have already sent for other meetings that week. The name is chosen and signed for: constituting a panel is a decision, and a name produced by a tool would weigh on it without anyone knowing why that one. You appoint one, and the summons goes out straight after.
And a calendar remark: across your last six recruitments, the panel was constituted on average 11 days before the closing date. Here there are 4 — not abnormal, but the low end of your usual range. panel_what-is-blocking.pdf4 people who have sat · no name proposed
⛓ Source · resolution setting the composition, 6 previous panels, shared calendars
What I record: the panel must have three members under your own resolution, two are appointed, and the third must hold a particular capacity that your resolution specifies.
What I looked for: people who have sat in that capacity on your previous panels. I find four, with the date of the panel each sat on. Three are still in post.
What I hand you, and what is left for you to do: the four people who have already sat in that capacity, the date of the panel each sat on, the three still in post, the resolution setting the rule, and the availability three of them have already sent for other meetings that week. The name is chosen and signed for: constituting a panel is a decision, and a name produced by a tool would weigh on it without anyone knowing why that one. You appoint one, and the summons goes out straight after.
And a calendar remark: across your last six recruitments, the panel was constituted on average 11 days before the closing date. Here there are 4 — not abnormal, but the low end of your usual range. panel_what-is-blocking.pdf4 people who have sat · no name proposed
⛓ Source · resolution setting the composition, 6 previous panels, shared calendars
A candidate asks what the tool did with their file. What do I say?
You show them, and the answer fits on one page.
What the tool did with their file: it looked in it for four elements stated in the advert, found three, gave the page of each, and wrote "not found" for the fourth.
What the tool did not do: it gave them no score, compared them to nobody, placed them in no order, and read neither their name, nor their address, nor their date of birth.
What that gives you: an answer that is true, checkable and short. Most authorities cannot answer that question, not because they have something to hide, but because nobody kept a record of what the tool looked at.
What is kept to make it possible: the elements recorded, their page, and the date. Nothing else — and above all no trace of a judgement, since none was made.
The document is ready, and it is handed to any candidate who asks for it, as it stands. what-the-tool-did_1-file.pdf1 page · handed to the candidate as it stands
⛓ Source · elements recorded on the candidate's file, art. 22 GDPR
What the tool did with their file: it looked in it for four elements stated in the advert, found three, gave the page of each, and wrote "not found" for the fourth.
What the tool did not do: it gave them no score, compared them to nobody, placed them in no order, and read neither their name, nor their address, nor their date of birth.
What that gives you: an answer that is true, checkable and short. Most authorities cannot answer that question, not because they have something to hide, but because nobody kept a record of what the tool looked at.
What is kept to make it possible: the elements recorded, their page, and the date. Nothing else — and above all no trace of a judgement, since none was made.
The document is ready, and it is handed to any candidate who asks for it, as it stands. what-the-tool-did_1-file.pdf1 page · handed to the candidate as it stands
⛓ Source · elements recorded on the candidate's file, art. 22 GDPR
What do you keep out of all this?
The strict minimum needed to answer that question, and nothing that outlives the recruitment.
What is kept: the elements recorded, the page each was read on, the date, and the missing documents flagged.
What is not kept: no ranking, no score, no candidate profile, no comparison between applications, and no data from one recruitment reused in another.
The last line deserves a word: a candidate not selected in March who applies again in October is a new candidate. A tool that remembered them would apply to their second application what it retained of the first — and the person would be judged on a file they have replaced.
What the monthly summary contains: the documents most often missing, the vacancy criteria nobody states, and the deadlines met. Three indicators about your adverts and your circuits, never about the people who answer them. what-is-kept.pdf4 items kept · 5 impossible
✎ Framework · retention periods to be set by the authority
What is kept: the elements recorded, the page each was read on, the date, and the missing documents flagged.
What is not kept: no ranking, no score, no candidate profile, no comparison between applications, and no data from one recruitment reused in another.
The last line deserves a word: a candidate not selected in March who applies again in October is a new candidate. A tool that remembered them would apply to their second application what it retained of the first — and the person would be judged on a file they have replaced.
What the monthly summary contains: the documents most often missing, the vacancy criteria nobody states, and the deadlines met. Three indicators about your adverts and your circuits, never about the people who answer them. what-is-kept.pdf4 items kept · 5 impossible
✎ Framework · retention periods to be set by the authority
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 stages in examining applications. All these uses work in support, subject to your approval.
Included in your agent
The 4 capabilities essential to this promise are included, at no extra cost.
From 895 € incl. VAT / month
Reading the applications
Finds in each file what the notice's conditions require.
Statutory conditions applied
Applies the conditions of access you have set, exclusions included.
Traceability of findings
Refers every finding back to the wording in the file that supports it.
Supporting documents
For checking that documents are complete, a dedicated agent completes the picture.
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
Apply transparent criteria and avoid automated decisions
Preserve human validation, recourse and audit trail
Protect HR data with fine-grained access and retention periods
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.
Integrate with the HR information system/ATS and measure quality, turnaround time and fairness
Other needs our agents cover
Each card says where the matching agent stands: available, on quote, or still being architected.
HR agent — employee support, onboarding and HR documents
HR base, 1–2 integrations.
On quote View the agent page →Need to go further?
These agents handle a different business process, with their own owner and their own price. They are added to this one.
Does your need fall outside this?
Book the free audit →
Build your agent
In 15 minutes we identify the agent that will give your staff the most time back — without oversizing the project.
The gain
How much time can a department give back to interviews?
By taking on the checking of conditions, the effort shifts towards meeting the candidates. How large the gain is depends on your volume and remains to be confirmed by a pilot.
Checking the conditions of access
Reading the files condition by condition
Preparing the selection file
Indicative figures, not contractual, to be confirmed by a pilot on your volume of applications and number of notices published. Assessing merit and deciding to appoint rest with the panel and the department: these are decisions about people.
How it works
The stages of your AI agent project
1
Audit & scoping
15 minutes to target the use case with the best return.
2
Quote or direct sign-up
A catalogue offer is bought online; a specific need gets a costed quote.
3
Design
We design the agent and its guardrails.
4
Integration & testing
We connect your tools to the agent, which is itself hosted in France.
5
Rollout
Going live and training your team.
6
Operation
Continuous supervision and improvement.
Pricing
One package, one agent
A recruitment support agent (statutory conditions, traceability, human control), installed and operated for you.
Agility
Setup + controlled subscription
10,080 € incl. VAT setup
then 895 € incl. VAT/month — you invest at installation and pay a reduced subscription. Ideal for keeping the cost under control over time.
- Installation, configuration and training for your teams
- Operation, human oversight, updates and support
- Sovereign hosting in France, a dedicated and isolated resource
The simplest
Serenity
All inclusive, no setup fee
1,455 € incl. VAT /month
all inclusive, immediate start. No upfront investment: a single subscription. Ideal for starting quickly and simply.
- Setup included (installation, configuration, training)
- Operation, human oversight, updates and support
- Sovereign hosting in France, managed end to end
100% Sovereign
On site, you own it
14,876 € incl. VAT setup
then 1,155 € incl. VAT/month · + hardware from 2,989 € (one-off purchase, in addition) — a sovereign computer installed on your premises, maintained remotely. Models run locally, your data returned at the end of the contract. 36-month commitment.
- Hardware installed on your premises (you own it)
- French / European AI models run locally
- Secure remote maintenance (Pro support included)
Not included in the packages: AI consumption (model tokens), re-invoiced at real cost with no margin, and tracked in real time in your client area. Maintenance and supervision subscription for an initial term of 12 months for the Agility package, 24 months for the Serenity package and 36 months for the 100% Sovereign package, renewable; support levels (SLA 72 h / 24 h / 4 h) optional. Bespoke development, additional integrations or exceptional volumes are quoted separately. Support Monday to Friday, 9am to 6pm. Prices include VAT at 20%: as a public body that is not VAT-registered, you cannot reclaim it.
AI model:
none of the AI models offered currently carries a fixed surcharge. When the selected model carries a cost, that cost is shown when you choose it, before you order, and re-invoiced at the cost incurred, with no mark-up; usage is billed at the publisher's price. Publishers' prices are published in US dollars: the amount re-invoiced is the amount in euros actually borne by Blue Lemon Agent on the publisher's invoice, at that invoice's exchange rate, with no commission or mark-up.
Included components and additional components Components included in the base offer: the Blue Lemon Agent software foundation, the AI models listed in the order journey, the standard channels (Microsoft Teams, Slack, WhatsApp Business, email, website chat, calendars, Microsoft 365 / Google Workspace, file storage, market VoIP telephony, professional social-media pages and accounts, Google Business Profile), hosting in France for the package chosen, backups, supervision, updates and support. If adapting the AI agent to your constraints, your needs or your requests requires other paid components — a third-party publisher's software licence, paid API access to one of your applications, hosting of health data, for which French law requires an HDS-certified host (art. L. 1111-8 of the French Public Health Code), SecNumCloud-qualified hosting, a speech synthesis service, particular hardware —, they are offered to you as an option or on quotation and re-invoiced at the cost incurred; nothing is committed without your written agreement. Where the artificial intelligence model you choose entails an additional cost, that cost is shown to you before you order and re-invoiced to you at the cost incurred, with no margin.
What to expect
Go-live
2 to 3 weeks
Agent designed, channels connected, team trained.
Steady state
4 to 7 weeks
After a few weeks of real use, once the agent's behaviour matches what you expect.
Indicative estimate, adjusted to the options you keep. It is not a delivery commitment.
Our commitment
Four guarantees that matter to your recruitment
Candidates' data stays with youLocal inference or an isolated resource hosted in France; no application entrusted to a third party, no data used to train a model.
Data in France, under French lawCandidates' personal data: minimisation and location in France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
Your recruitment department keeps the decisionThe agent produces applications examined against your statutory criteria, which can be checked and altered; no approval is automated.
Human oversight & traceabilityOn your volume of applications and number of notices published: systematic logging and tracking, in line with the AI Act.
Going further
Related resources
Frequently asked questions
Your questions, our answers
Does the agent rank candidates by merit?
No. It applies the objective conditions of access you have set and presents what each file establishes. Assessing merit rests with the panel and the department.
Can it be given exclusion conditions?
Yes, provided they are objective and lawful: a statutory condition, a qualification required, a length of experience, a mandatory authorisation. Every application of them is recorded.
How is bias avoided?
The agent produces no assessment: it reports what the file mentions, condition by condition, with the document to back it up. What is not mentioned is flagged, never inferred.
How is the European AI Act respected?
Recruitment is one of the high-risk uses. The agent prepares and documents without taking any automated decision about a person, with human control at every step.
How long are applications kept?
For the period you set, with automatic purging at the deadline, in line with the GDPR.
How long does it take to deploy this agent?
A few months as a rule, depending on the volume of applications and the number of notices published, after a free audit then phases of design, integration and testing.
Related use cases
Other agents for your human resources
Let's talk
Let's size up the potential in your recruitment
15 minutes to frame your notices and your statutory conditions — hosted in France, supervised, with no commitment.