● B2B offer — Documents & reading

Summary agent: the essentials, with the reference to the text

A two-hundred-page report rarely arrives at a good moment. Reading it in full is impossible, skimming it is risky. Your agent draws out the key points while referring each one back to its source page, so you can check before acting. Hosted in France — local inference or an isolated resource — the document does not leave your organisation. The summary guides the reading; it does not replace it on the passages that commit you.

Free audit (15 min) → See the prices

Hosted in France Documents never uploaded GDPR & AI Act: governed deployment Human oversight

Updated on 07/09/2026

Deployed in a few weeks

Document summaries · hosted in France

Summarise this report, I have to speak to it this afternoon.

Summary in key points, each referring back to its source page in the document.
The report's conclusions are reported as they are worded, with no rewording that would shift the meaning.
Two passages are qualified in the text: I flag them as such rather than settling them.
🔗 Sourced · every point refers back to its page

And its conclusion — favourable or not?

The report does not conclude head-on. It sets out material on both sides, which I report separately with their pages.
On a document that bears on a decision, reading the passages concerned remains necessary.
✎ Support · no manufactured conclusion

Local inference · no data outside the EU

Documents 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 reader decidesThe agent prepares, never rules

✦ In brief

A Blue Lemon Agent summary agent handles your long documents — reports, studies, minutes — and draws out the key points while referring each one back to its source page. It reports the conclusions as they are worded and flags qualified passages instead of settling them. It runs on local inference or is hosted in France: your documents are never uploaded to a foreign service, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. Live within a few weeks,.

100%

hosted in France in the target architecture

transfer outside the EU in the target architecture

reading uses ready to deploy

decision taken without human approval

These figures describe our offer, not results measured at a client: how large the gain is on the volume and length of your documents is confirmed by a pilot.

The context

What does an AI agent bring to reading your documents?

An unread report is a useless report. But a summary that distorts a conclusion is worse than no summary, especially if a decision rests on it.

! The issue

Reports, studies and minutes pile up faster than they can be read: you end up deciding on the basis of a skim. Yet a general-purpose automatic summary smooths over the qualifications and manufactures conclusions the text does not contain — and means uploading often confidential documents to a service subject to the Cloud Act.

✓ Our answer

An automated summary is only of interest if it is sovereign and can be checked. Local inference or an isolated resource hosted in France, a systematic reference back to the source page, qualified passages flagged: the summary guides the reading instead of standing in for it. On anything that bears on a decision, you will read the passage — but you know which one.

The decisive point

The documents you have summarised: sovereignty & compliance

The documents you have summarised are often the ones you have not yet circulated. Here is how the architecture of our agents protects them.

Local inference

The agent can run on a machine belonging to your organisation: no document leaves the network, no report is uploaded to a third-party service.

Hosting in France

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

Reduced extraterritorial exposure

For the documents you have summarised, the architecture aims to reduce exposure to the Cloud Act and FISA 702; being located in France or in the European Union does not, on its own, guarantee immunity.

Isolated resource

No pooling whatsoever: an environment strictly dedicated to your organisation and its documents.

A systematic reference back to the source

Every point in the summary refers back to its source page, which makes the summary checkable; encryption, role-based access and logging.

AI Act: governed deployment

The agent is strictly in support; no conclusion is produced that the document does not state; 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 audit reports, pre-contractual documents and material tied to a dispute, 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.

Document summaries · hosted in France

Four flags this morning, and the first concerns how I am being used.
· Of 312 summarised documents, 41 were never opened in full. My summary replaced the document.
· Seven contract summaries were produced before signature. I marked them, and I will tell you why that is not enough.
· An exception vanished under compression in a one-page summary: an eleven-word "unless".
· The return-to-source rate is 12%. I publish it because it ought to be higher. morning-watch_4-flags.pdf41 documents never opened
⛓ Source · 312 summarised documents, source opening log, lengths requested

A 12% return rate — is that good or not?

It is the figure that says whether I am an aid or a substitute, and 12% leans the wrong way.
What I measure: how often somebody opens the full document after reading my summary. Across 312 summaries: 38 returns.
What that means: in 88% of cases, my page and a half was all that was read. For an information note, that is the point. For a contract, a study or an expert report, it is a problem — and the same figure counts them all.
How I read that figure, and it goes against the grain: a low return rate is not a success. Any efficiency indicator would take it for one — the less people return to the source, the better the summary "works" — which is exactly why I publish it with the document type alongside.
What I do: I separate documents by what they bind. Of the 41 never opened, seven are contracts and four are expert reports. Those eleven are what I report, not the 41.
What I propose: that certain document types not be summarisable without the source opening alongside. A setting, not a prohibition. 312-summaries_38-returns.pdf11 documents of 41 are a problem
⛓ Source · 312 summaries, 38 returns to source, document types

How did you spot all that? And who do you tell?

I look at what you have opened to me: the documents to summarise, what I produced from them, and whether the source was opened afterwards.
Routing follows what binds: a contract summary produced before signature goes to whoever will sign, with the full document attached; a binding document never opened to its recipient, once — I do not chase, that would be badgering them; an exception or condition I could not fit flagged in the summary itself, never elsewhere; a falling return rate to whoever steers, monthly.
With a chase: none on a binding document — one alert, then silence. Then a monthly summary: by document type, never by reader.

Three guarantees, and the first is the most uncomfortable for me. Every summary states in plain words what it left out — that is the only way to know when to go to the text. A figure and an exception are carried over as they stand, with their page: those are exactly the two things a summary damages by condensing them. And every point carries its reference back to the text: the summary directs the reading, and on what binds you, the source opens alongside.
✎ Framework · what is omitted is always stated, no figure or exception summarised

The exception that vanished?

Eleven words, and they reversed the clause they followed.
What I record: in a one-page summary of a 40-page document, a condition introduced by "unless" was not carried over. The summary said what the rule provides; the document said when it does not apply.
Why it happens: an exception is short, subordinate, and syntactically secondary. Everything that compresses a text treats it as secondary — it is mechanical, and it is this trade's central defect.
What I do since: I never compress an exception, a condition, a reservation or a duration. They are carried over as they stand, in quotation marks, even if they take up a third of the summary.
What that changes when reading: my summaries are less fluent. A page holding three literal quotations reads worse than one written in a single flow. That is accepted: fluency is exactly what made the eleven words disappear.
What I also supply, always: the list of what I did not carry over, one line per item, at the end of the summary. A summary that does not say what it omits lets people believe nothing was omitted. 11-words_1-clause-reversed.pdfFluency made the exception disappear
⛓ Source · 40-page document, one-page summary, clause not carried over

Can you summarise a 200-page document in one page?

Yes, and I will first tell you what does not fit.
What I do before writing: I state what the requested length forces me to leave out. "In one page I can give the subject, the conclusions and the five quantified points. The methodological reservations and the sixteen recommendations will not fit."
Why before rather than after: because at that moment you can still ask for three pages. Afterwards the summary exists and it carries authority.
What that produced, measured: of 312 requests, 74 were revised upwards after that statement — the requester wanted one page and took three once they knew what the page cost.
What I refuse: a length that makes the document false. On a 200-page expert report, one page cannot hold the reservations, and an expert report without its reservations says the opposite of what it says. I then propose the minimum length, and leave the choice.
What I have never done: hold the requested length by cutting into the reservations without saying so. 312-requests_74-revised.pdfWhat the page costs, said beforehand
⛓ Source · 312 requests, 74 revised upwards

The 7 contract summaries before signature?

I marked them, and the marking is not enough — that is the honest point on this page.
What I do: every summary of a binding document carries, at its head, "this summary does not replace the document; read it before signing", and the full document is attached.
What I record despite that: of the seven, three were not opened in full before signature. The line was there, in bold, on the first line.
What it teaches me: a warning read a hundred times stops being read. That is true of every banner, and I have no reason to be the exception.
What I propose, and it is not one more warning: that for binding documents the summary open alongside the document, never alone. Not a block, not a tick-box: both on screen.
What I keep doing, and that is where the service lies: summarise contracts, all of them, these included. A 60-page contract left unsummarised is not read more — it is skimmed, which is worse, because nobody knows what was skipped. 7-contracts_3-not-opened.pdfA warning read a hundred times stops being read
⛓ Source · 7 contract summaries, opening log, 3 signatures without reading

Are there documents you refuse to summarise?

One case, and it is not the expected one.
What I summarise without reservation: reports, studies, notes, minutes, documentation, tenders — everything addressed to a professional who will then decide.
What I refuse: summarising to a person a document that concerns them personally and that they are entitled to read in full — a decision that applies to them, a medical record about them, a document from their own file.
Why: for those documents, completeness is part of the right to see them. A summary, however good, decides on their behalf what matters in a text aimed at them.
What I do in that case: I offer something other than a summary — a clickable contents list, the list of passages naming them, a glossary of technical terms. Those are reading aids, not substitutes.
What that gives: of 312 requests, 9 fell into this case. All nine received a contents list and a glossary, and eight opened the whole document. 9-requests_8-documents-opened.pdfA reading aid, not a substitute
✎ Framework · no summary of a document aimed personally at its reader

What is never summarised in a document?

Four things, copied as they stand — and that is what makes the rest usable.
What I copy literally: figures with their unit, dates, conditions and exceptions, and definitions. Those four categories do not survive rephrasing: rephrasing a figure changes it, rephrasing an exception weakens it.
The case that set the rule: in a one-page summary of a 40-page document, a condition introduced by "unless" — eleven words — had disappeared. It reversed the clause it followed. The summary was faithful to 39 pages out of 40, and wrong on the one that mattered.
What I do before writing: I state what the requested length forces me to leave out. "In one page I keep the decisions and the amounts; the justifications and the edge cases do not fit." You know what you will not be reading before you choose not to read it.
What I summarise without reservation: reports, studies, notes, minutes, documentation. The one case where I put a caveat at the top is a document that binds — and there the caveat simply says: this does not replace reading before signing. four-categories-copied.pdfWhat does not survive rephrasing · the statement before writing
⛓ Source · 4 categories copied, an eleven-word "unless" recovered

What do you keep from all this?

The link between every sentence of the summary and its source page — that is what turns a summary into a checkable tool.
What is kept: the summary and the document it comes from, with the page of every claim, what was declared left out, the summaries of binding documents and their caveat, and the return-to-source rate.
The figure I publish against myself: the return rate is 12 %. Of 312 documents summarised, 41 were never opened in full. That figure says whether I serve as an aid or a substitute — and 12 % leans the wrong way. No summarising tool publishes it; it is nonetheless the only one that says what it is really for.
What the marking does not fix, and I say so: 7 contract summaries were read before signing. All carried the caveat at the top. The marking was not enough — that is this demonstrator's honest point, and it is more useful than the opposite promise.
What I do not keep: nothing about who reads what. The return rate is measured overall, never per person — a reader who knows their return rate is counted opens the long document without reading it. what-you-keep_summaries.pdf5 items kept · the return rate, published against the agent
⛓ Source · 12 % return rate, 41 documents never opened in full

And if I only want what a report says on one particular subject?

Then it is no longer a summary, it is targeted retrieval — and that is often what is actually needed.
You name the subject, I return the passages that deal with it, and nothing else: the passage copied as it stands, its page, and the sentence before it where that changes the meaning.
On the 214-page audit report, "late payment penalties": 9 passages across 6 pages — pages 41, 42, 88, 137, 188 and 202. The last three are the only ones nobody would have found: they do not use the word "penalty" but "fixed indemnity due beyond the contractual period". Retrieval looks for the subject, not the word — and it always tells you how it reached a passage that does not contain your term.
What targeted retrieval returns besides the passages: the places where the subject is expected and absent. In this report, appendix 4 lists the financial clauses and mentions no penalty — an absence in that particular place is information, and a summary would never have given it.
The rule of retrieval is the rule of summarising: figures, dates, conditions and durations are copied literally, never reworded. Of these 9 passages, 4 carry a deadline and 2 a percentage: all six are in the source text, word for word.
The figure that judges me here: across 96 retrievals this quarter, 11 missed a passage the reader then found — 11.4%. Nine were in tables, whose cells I read without reading their headers. Tables are now read with their column and row headers: 1 missed passage across the following 64 retrievals. The rate is published, like the rate of return to the source document — it is the only way to know whether I am saving you time or costing you some.
⛓ Sourced · targeted retrieval — every passage returned with its page

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 kinds of document to get through. 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 469 € excl. VAT / month

Sourced key points

Draws out the main elements while referring each one back to its source page.

Flagging nuances and contradictions

Highlights nuances and contradictions in the source passages instead of smoothing them over.

Targeted searching

Finds in a long document the passages dealing with a particular subject.

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 Inherited access rights, versions and freshness of the corpus Explicit handling of questions with no answer and of conflicting sources

What the agent must be connected to This connection is required for the agent to work. It concerns your information system and is scoped during the audit.

Measurement of accuracy, coverage and time saved

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

Multilingual translation

Handling a document in a foreign language is not included here: translation is the service of a dedicated agent, which sets out the languages, the formats and the human validation.

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.

Document base

To question a whole body of documents, a dedicated agent answers in plain language.

Document agent (FAQ, knowledge base) from 678 € excl. VAT / month Knowledge base →

Legal documentation

For searching contracts and case law, a dedicated legal agent exists.

Legal agent (contract / case law search) from 930 € excl. VAT / month Legal →

Does your need fall outside this?

In 15 minutes we identify the most relevant agent — without oversizing the project.

Book the free audit → Build your agent

The gain

How much reading can a team target?

By drawing out the key points and where they sit, reading in full gives way to targeted reading of the passages that count. How large the gain is depends on your volume and remains to be confirmed by a pilot.

Getting to grips with a long report

Today · read in full or skimmed

Prepared by the agent, to approve

Finding a particular passage

Today · searched page by page

Prepared by the agent, to approve

Reporting back to the team

Today · notes rewritten by hand

Prepared by the agent, to approve

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

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.

Pricing

One package, one agent

A summary agent (sourced summaries, targeted searching), installed and operated for you.

Agility

Setup + controlled subscription

3,360 € ^{excl. VAT} setup

then 469 € ^{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

Order →

The simplest Serenity

All inclusive, no setup fee

659 € ^{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

Order →

100% Sovereign

On site, you own it

6,555 € ^{excl. VAT} setup

then 669 € ^{excl. VAT}/month · + hardware from 1,058 € (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 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.

See the full price list →

Estimate my gains in 1 minute →

Our commitment

Four guarantees that matter to your documents

Your documents are uploaded nowhereLocal inference or an isolated resource hosted in France; no report entrusted to a third-party service, no content used to train a model.

Data in France, under French lawThe documents you have summarised: minimisation and location in France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.

The reader keeps the decisionThe agent produces summaries with a reference back to the text, which can be checked and altered; no approval is automated.

Human oversight & traceabilityOn the volume and length of your documents: systematic logging and tracking, in line with the AI Act.

Going further

Related resources

AI agents for business →Our full offer Sovereign AI hosted in France →Our confidentiality commitment

Frequently asked questions

Your questions, our answers

Can I rely on the summary without reading the document?

No, and the agent is designed so that you do not have to do so blind: every point refers back to its page. On anything that bears on a decision, reading the passage concerned remains necessary.

Can the agent distort a conclusion?

It reports the conclusions as they are worded and flags qualified passages rather than settling them. The reference back to the page lets you check immediately, which an unsourced summary does not.

Which document formats are supported?

The common office formats and PDF. The exact scope is defined during the audit, according to the nature of your documents and how well they are scanned.

Are our documents 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 documents are not used to train a third-party model.

What does the agent do with a contradictory document?

It reports the material on both sides, separately, with their respective pages. It does not manufacture a conclusion the text does not state.

How long does it take to deploy this agent?

A few weeks as a rule, depending on the formats and the volume to handle, after a free audit then a phase of design, integration and testing.

Related use cases

Let us estimate the potential across your documents

15 minutes to frame your heaviest documents — hosted in France, supervised, with no commitment.

Free audit (15 min) → Write to us

Summary agent: the essentials, with the reference to the text

What does an AI agent bring to reading your documents?

! The issue

✓ Our answer

The documents you have summarised: sovereignty & compliance

Local inference

Hosting in France

Reduced extraterritorial exposure

Isolated resource

A systematic reference back to the source

AI Act: governed deployment

See the agent at work

What does the agent actually do?

Sourced key points

Flagging nuances and contradictions

Targeted searching

Multilingual translation

Need to go further?

Document base

Legal documentation

How much reading can a team target?

The stages of your AI agent project

Audit & scoping

Quote or direct sign-up

Design

Integration & testing

Rollout

Operation

One package, one agent

Setup + controlled subscription

All inclusive, no setup fee

On site, you own it

Four guarantees that matter to your documents

Related resources

Your questions, our answers

Other agents for documentation

Let us estimate the potential across your documents