Document summaries: the essentials, with the pages cited
A three-hundred-page report, a study, a case-law file: reading it in full is not always possible in the time available. Your agent produces a structured summary citing the page of every element kept, and flags what the document does not deal with. Hosted in France: your working documents stay within your administration. The agent summarises, your officers judge and decide.
Updated on
Every element kept refers back to its page in the document.
Two questions are not dealt with by the report: they are flagged as such.
🔗 Sourced · pages of the document, cited one by one
Drawing a conclusion for your administration rests with your officers: the document informs, it does not decide.
✎ Support · matching provided, human judgement
A Blue Lemon Agent summary agent handles reports, studies, files and case law and produces a structured summary in which every element refers back to its page. It flags what the document does not deal with rather than filling the gap. It runs on local inference or is hosted in France: your working documents stay with you, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
These figures describe our offer, not results measured at a client. How large the gain is on the volume of documents handled each month is confirmed by a pilot.
What does an AI agent bring to reading your long documents?
A summary in which every element is referenced lets you go straight to the useful passage.
! The issue
Working through a long document calls for drawing out the essentials and being able to go back to the exact passage. A summary with no reference forces a re-read; a summary referenced page by page makes the document usable straight away. That is the behaviour built into the agent's design.
✓ Our answer
Your officers have a structured, referenced summary and the list of what the document does not deal with. Drawing a conclusion for the administration rests with them: the document informs a decision, it does not take it. Local inference or an isolated resource hosted in France: your working documents and your files in progress do not leave your perimeter.
Your working documents and your files in progress: sovereignty & compliance
Your working documents and your files in progress call for strict confidentiality. Here is how it is assured.
Local inference
The agent can run on a machine belonging to your organisation: no document and no extract leaves the network.
Hosting in France
Otherwise, a dedicated and isolated resource hosted in France, under French law — your reports, studies and files: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
For your working documents and your files in 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: an environment strictly dedicated to your administration and its working documents.
Every element refers back to its page
The summary cites the page of every element kept and flags what the document does not deal with; encryption, role-based access and logging.
AI Act: governed deployment
The agent is strictly in support; no conclusion is presented as the administration's; 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.
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.
· A 214-page report arrived on 12/07 and nobody has opened it. It contains a deadline of 15 September, on page 187.
· Two versions of the same document are in circulation in two departments, with different figures in appendix 3.
· An internal note cites a report that does not exist in your library — under neither that title nor that date.
· Four reports due under an agreement have never been received. The oldest was due in January. morning-watch_4-flags.pdf4 flags · documents concerned
⛓ Source · document library, opening logs, agreements and their dates
The passage, quoted: "The authorities concerned shall submit their action programme by 15 September 2026 at the latest." Page 187, appendix 4, final paragraph.
What I checked before flagging it: your authority does appear in the list of authorities concerned, on page 12 of the same report. This is not a general deadline: it is yours.
What I did not do: summarise the 214 pages for you. That is not what you need this morning — you need to know that a deadline 39 days away has gone unnoticed.
The full summary is ready if you want it, structured into subject, findings, recommendations and useful appendices, each element pointing back to its page.
Three points are flagged as absent from the report although one would expect them. report-214p_deadline-p187.pdf1 deadline · 39 days · passage quoted
⛓ Source · report of 12/07, p. 12 and p. 187
Routing follows your organisation: the 15 September deadline to the chief executive and the department concerned, both together; the two diverging versions to the two departments that hold them, with appendix 3 of each side by side; the missing report to the author of the note that cites it; the four unreceived reports to the department that signed the agreement.
With a chase: 24 h on the deadline, 7 days on the rest. Then a monthly summary: by document, never by officer.
What those four flags are worth this morning: a statutory deadline of 15 September found on page 187 of a 214-page report nobody had opened — 39 days to prepare the action programme instead of discovering it out of time; two diverging versions reconciled before one department writes on the other's figures; four reports due under an agreement claimed back, the oldest since January.
What that gives the service back: the reading hours go back to casework, the resident gets an answer within the stated time, and two departments handle the same file on the same figures — that is where equal treatment starts. The access stays yours: opened by role, logged, withdrawn on a word, and your working documents do not leave the administration — local inference or an isolated resource hosted in France.
The next step is ready: the full summary of the 214 pages, structured into subject, findings, recommendations and useful appendices, each element pointing back to its page, and the three points flagged as absent from the report. Tell me who receives it and at what time: it goes out within minutes, and the conclusion the administration draws stays its own, informed by pages it can reopen.
✎ Proposal · watch and chases to be configured — you set the thresholds
Subject — pp. 3 to 9. Findings — 14 retained, from p. 22 to p. 156. Recommendations made — 9, pp. 158 to 174. Useful appendices — 4 of 11, with why the other seven are not.
What I keep apart, and it matters: the recommendations the report makes and the findings that might call for others. Those are two different things, and a summary that mixes them makes the report say what it did not write.
Three absences flagged, because one would expect them in a document of this kind:
· No costing of the nine recommendations.
· No implementation timetable, apart from the deadline in appendix 4.
· No mention of authorities under 3,500 residents, although the scope announced on p. 5 includes them.
The third is the one that concerns you most. summary_214p-report.pdf14 findings · 9 recommendations · 3 absences
⛓ Source · 214-page report, original pagination
For the costing: I searched the 214 pages for cost, budget, amount and euro. 17 occurrences, all in the budgetary context section, none attached to a recommendation.
For the timetable: I searched for dates and deadlines. One enforceable deadline only, the one in appendix 4; the other dates recall past events.
For small authorities: the scope announced on p. 5 covers "all authorities". The 3,500-resident threshold appears nowhere, and the 14 findings rest on a sample of 42 authorities the smallest of which has 8,200 residents, p. 21.
What I do not say: that the report is inapplicable to small authorities. I say it did not observe them — the consequence is yours. 3-absences_search-method.pdf3 absences · how each was searched for
✎ Support · search method given — the scope stays for the service to judge
What is identical: the body of the document, the first 40 pages, and nine appendices out of eleven.
What differs: appendix 3, the table of projected allocations. The March version shows €1.42m; the June version shows €1.28m.
Who holds what: finance works from the March one, technical services from the June one. Both are preparing a joint document for September.
What I record, and what gets settled: two dates and two figures — a later version is not always the operative one, a document can be corrected and the correction withdrawn. Which version is authoritative is named by the two departments, before September's joint document, and it takes one line to say.
What I have prepared: both appendices side by side, line by line, with the six lines that differ — the €140,000 gap sits on two lines only. appendix-3_two-versions.pdf6 diverging lines · 2 carry the gap
⛓ Source · March version, June version, library deposit logs
What I can do: when a document enters the library and one with the same title or origin is already there, flag it to both holders, with each one's date and the pages that differ.
What I leave in place, and why: both versions, withdrawing none and marking none obsolete. An older version can still be the operative one, and a document library that tidies up on its users' behalf becomes a library nobody uses. What I add instead: the flag to both holders, dated, with the list of pages that differ — which is what it takes to settle it in two minutes.
What your library holds today: I searched for duplicates. 184 documents exist in at least two versions, 31 of them with differing content — the other 153 are strict duplicates, which is a housekeeping matter and not an information one.
Of the 31, seven differ on figures. Those are the ones I would surface first. library_184-duplicated-documents.pdf184 duplicates · 31 diverging · 7 on figures
✎ Proposal · flagged on entry — no version withdrawn
· Annual report from a concession holder — due in January, never received. 7 months. The agreement provides for it at article 12.
· Interim report from a grant-funded association — due in April. 4 months.
· Inspection report from a body — due in June. 2 months.
· Activity report from a joint authority — received in April, but never filed in your library. It is sitting in a shared mailbox.
The fourth is not a delay: the document exists, it arrived, and nobody knows. It is the easiest to fix and the most invisible.
What I have written, and what is left to sign: all three chasers are ready, each citing the article of the agreement that provides for the document, its due date and the delay in months. They are signed by the department that signed the agreement — a chaser on a contractual obligation comes from it, and that is what makes it enforceable. Three signatures, and seven months of delay stop being silent. 4-expected-reports.pdf3 delays · 1 received but never filed
⛓ Source · 4 agreements and their articles, document library, shared mailbox
What I did: reread your live agreements and record every document the other party must produce, with its date and the article providing for it. 34 agreements, 41 documents expected over twelve months.
What that changes: today those dates live inside the agreements, which is to say nowhere. Nobody holds the list, and an unreceived report produces no signal — an absence is silent by nature.
What I can do, if you approve: flag each expected document fifteen days before its date to the signing department, then on the date, then fifteen days late. No chaser is sent by me: I warn the person who will sign it.
And a quarterly summary to the chief executive: documents expected, received, late. By agreement, never by organisation — the point is keeping a calendar, not scoring partners. agreements_41-expected-documents.pdf34 agreements · 41 documents · 11 before December
✎ Proposal · 41 dates tracked — chasers stay with the signing department
Your case is not here? That is exactly what a 15-minute conversation is for. Book the free audit →
What does the agent actually do?
One agent, several kinds of document. All these uses work in support, subject to your approval.
Structured summary
Draws out the subject, the findings, the recommendations and the useful annexes.
Pages cited
Refers back to the page of the document for every element kept.
Gaps flagged
States the questions the document does not deal with.
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.
In 15 minutes we identify the agent that will give your staff the most time back — without oversizing the project.
How many documents can a department work through?
By taking on the summarising and the referencing, the effort shifts towards analysis and decision. How large the gain is depends on your volume and remains to be confirmed by a pilot.
The stages of your AI agent project
Audit & scoping
15 minutes to target the use case with the best return.
Quote or direct sign-up
A catalogue offer is bought online; a specific need gets a costed quote.
Design
We design the agent and its guardrails.
Integration & testing
We connect your tools to the agent, which is itself hosted in France.
Rollout
Going live and training your team.
Operation
Continuous supervision and improvement.
One package, one agent
A document summary agent (structure, pages cited, gaps), installed and operated for you.
Setup + controlled subscription
- Installation, configuration and training for your teams
- Operation, human oversight, updates and support
- Sovereign hosting in France, a dedicated and isolated resource
All inclusive, no setup fee
- Setup included (installation, configuration, training)
- Operation, human oversight, updates and support
- Sovereign hosting in France, managed end to end
On site, you own it
- Hardware installed on your premises (you own it)
- French / European AI models run locally
- Secure remote maintenance (Pro support included)
Four guarantees that matter to your documents
Related resources
Your questions, our answers
What documents does the agent handle?
How do I check an element of the summary?
What does it do if a question is not dealt with?
Does the agent draw conclusions?
Are our documents protected?
How long does it take to deploy this agent?
Other agents for your administration
Let us estimate the potential across your documents
15 minutes to frame your volumes and your types of document — hosted in France, supervised, with no commitment.