● Public sector — Library & media library
AI agent for library staff and media libraries
An assistant that answers users, prepares catalogue records and supports cultural outreach — the librarian keeps the editorial choice and the approval. Hosted in France, on a resource isolated for each library, borrowing histories stay strictly confidential. The AI agent assists, the librarian decides.
Hosted in France
Borrowing histories confidential
GDPR & AI Act: governed deployment
Human oversight
Updated on
Deployed in a few weeks
Media library assistant · hosted in France
A reader is looking for novels about the sea for a 13-year-old.
Selection proposed. Six titles available in the young readers' section, from adventure novels to contemporary fiction, each with its shelfmark and location. Two titles currently on loan can be reserved.
The selection draws only on the catalogue and the records — never on other readers' borrowing history.
⛓ Source · your catalogue (library system) + bibliographic records
The selection draws only on the catalogue and the records — never on other readers' borrowing history.
⛓ Source · your catalogue (library system) + bibliographic records
Prepare the record for the new picture book that arrived this morning.
Record prepared. Fields filled in from the publisher's references: title, author and illustrator, ISBN, series, proposed indexing and a four-line summary. Shelfmark suggested according to your classification plan.
Nothing enters the catalogue without your review.
✎ Action · record ready for review — the librarian approves
Nothing enters the catalogue without your review.
✎ Action · record ready for review — the librarian 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 librarian decidesThe AI agent assists, it never chooses alone
✦ In brief
In a library or media library, a Blue Lemon Agent agent assists the team with the repetitive tasks — answering users (availability, loans, membership, opening hours), catalogue records, reading suggestions, themed selections and outreach material. It runs on local inference or is hosted in France on a resource dedicated to and isolated for each library: users' data and borrowing histories stay strictly confidential, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. The choice of collections and the outreach stay human; the time won back goes to advice and welcoming the public. Live within a few weeks. Your public-sector staff write to it from Microsoft Teams, Slack or their email, and users reach it on WhatsApp Business, the website chat or email — with no account to create and nothing to install. Reaching the administration 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
7
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 libraries — and why they hesitate
Longer opening hours, smaller teams, repetitive enquiries at the expense of outreach: libraries are looking for time. But the data they hold — memberships and borrowing histories — touches on the freedom to read, and deserves exemplary protection.
! The issue
Library staff are caught between users who expect immediate answers — availability, renewals, membership — and the substantive work that gives the profession its value: cataloguing, outreach, cultural programming. Yet most consumer AI tools would amount to entrusting memberships, reading habits and users' borrowing histories to a third party, often hosted outside Europe and subject to the Cloud Act.
✓ Our answer
For a library, AI is only of interest if it is sovereign and confidential by design. Local inference or an isolated resource hosted in France, systematic human oversight, editorial choice reserved to librarians: the time saved on enquiries and cataloguing is never paid for in lost confidentiality — borrowing histories feed no profiling. The aim is not to replace the librarian, but to give them back time for outreach.
The decisive point
Protecting readers' data: sovereignty & compliance
What users read is nobody's business but the library's. Here is how the architecture of our agents protects memberships and borrowing histories, library by library.
Local inference
The agent can run on a machine at the media library: no data leaves the network, nothing passes through a cloud.
Hosting in France
Otherwise, a dedicated and isolated resource, hosted in France under French law — readers' data: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
Exposure of borrowing histories 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 data: an environment strictly dedicated to your media library, guaranteeing the continuity of the public service.
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 profiling of readers; 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, SecNumCloud and HDS options are available depending on your requirements. A single architecture is designed to answer both the GDPR and extraterritorial exposure. Designed for deployment in line with the GDPR and the AI Act, after the processing, roles and context-specific risks have been assessed.
Demonstration
See the agent at work
5 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 bodyTrois-Rivières inter-municipal media library — a public reading network of three sites
- Sector
- Public reading — one main media library and two branches; 32 opening hours a week, cataloguing, cultural outreach and cultural programming
- Staff
- 9 staff — a director, three librarians including the head of cataloguing, two conservation assistants, two front-desk staff and a digital outreach officer; cataloguing rests on one person
- Public served
- 18,400 inhabitants, 4,300 active members, 61,000 loans a year and 21,400 searches a year on the network portal
- Order of magnitude
- 34,000 records in the catalogue, 2,400 documents acquired and catalogued each year, 9,800 user questions, 34 themed selections and 96 events
- Tools in place
- Library management system, network portal, mailbox, telephone and events diary — the agent connects to them read-only, nothing is replaced or migrated
- Who decides
- The director settles acquisitions and cultural programming; the head of cataloguing approves every record before it enters the catalogue; the librarians sign off selections and outreach texts
- The points to improve
- 1,100 catalogue records have neither summary nor subject indexing; 4,800 searches out of 21,400 return nothing; 3,724 questions arrive outside opening hours; a themed selection takes 6 hours; and 18 days separate a document's arrival from its place on the shelf
A fictional public authority, invented for the demonstration. At Trois-Rivières the team is not trying to do less: it is trying to spend more time in front of the public and less behind the catalogue. The data at stake is among the most sensitive a public service holds — what people read. The agent runs on local inference on a machine at the library, reads the catalogue and the records without ever writing to them, touches no borrowing history, and sends nothing into the catalogue without the approval of the head of cataloguing. The exchanges below cover a year, from the review of the catalogue to the report presented to the inter-municipal council.
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.
Media library assistant · hosted in France
I read your 34,000 records over the weekend, and I draw from them a finding your catalogue was already carrying without anyone having the time to see it: part of your collections is on the shelf and impossible to find on the portal.
A record is the description of a document in the catalogue — title, author, publisher, summary, subject terms; it is the record, not the book, that the portal's search engine queries. Subject indexing is the setting of those subject terms: it is what makes a reader looking for « sea stories » find a novel whose title says nothing about the sea.
The measured gap, and it governs everything else: 1,100 of your 34,000 records have neither summary nor subject indexing. They are your retrospective records, entered before 2011 when the collection was computerised — retrospective cataloguing is the after-the-fact re-entry of old cards, done as fast as possible so that the collection exists in the software. This is no failure of work: it is a collection that was saved quickly, and that nobody has had time to reopen since.
What it costs the public, on your own logs:
· 21,400 searches on the portal last year, 4,800 with no result at all — 22 %.
· 1,900 of those 4,800 searches were for a document you had on the shelf, available, a few metres from the reader who left without it.
· And 61 % of those 1,900 searches are subject searches, not title searches — exactly what the absence of subject indexing makes impossible.
What I propose, and it is not a promise: it is already written. I prepared the 1,100 enriched records last night — a four-line summary, subject indexing, a shelfmark proposed according to your classification plan, each field carrying the source that filled it: your own record, the scanned title page, the publisher's reference. The fields I could not read are left empty and named, never guessed.
What it is worth, measured on the first 620 your head of cataloguing has already approved: searches with no result fell from 22 % to 14 % in six weeks, and 340 documents that had never left the shelf since they were entered were borrowed within the month.
The next step, and it takes twenty minutes a day: you read 40 records in the morning, in subject batches — I sort them by shelf so that the reading stays a reading of the collection and not a sorting task. The remaining 480 are approved in twelve days, and the whole older collection becomes searchable again. catalogue-review_1100-records-to-enrich.pdf4,800 searches with no result, 1,900 for documents on the shelf
⛓ Sourced · 34,000 catalogue records, portal search log, loan volumes — no individual borrowing history opened
A record is the description of a document in the catalogue — title, author, publisher, summary, subject terms; it is the record, not the book, that the portal's search engine queries. Subject indexing is the setting of those subject terms: it is what makes a reader looking for « sea stories » find a novel whose title says nothing about the sea.
The measured gap, and it governs everything else: 1,100 of your 34,000 records have neither summary nor subject indexing. They are your retrospective records, entered before 2011 when the collection was computerised — retrospective cataloguing is the after-the-fact re-entry of old cards, done as fast as possible so that the collection exists in the software. This is no failure of work: it is a collection that was saved quickly, and that nobody has had time to reopen since.
What it costs the public, on your own logs:
· 21,400 searches on the portal last year, 4,800 with no result at all — 22 %.
· 1,900 of those 4,800 searches were for a document you had on the shelf, available, a few metres from the reader who left without it.
· And 61 % of those 1,900 searches are subject searches, not title searches — exactly what the absence of subject indexing makes impossible.
What I propose, and it is not a promise: it is already written. I prepared the 1,100 enriched records last night — a four-line summary, subject indexing, a shelfmark proposed according to your classification plan, each field carrying the source that filled it: your own record, the scanned title page, the publisher's reference. The fields I could not read are left empty and named, never guessed.
What it is worth, measured on the first 620 your head of cataloguing has already approved: searches with no result fell from 22 % to 14 % in six weeks, and 340 documents that had never left the shelf since they were entered were borrowed within the month.
The next step, and it takes twenty minutes a day: you read 40 records in the morning, in subject batches — I sort them by shelf so that the reading stays a reading of the collection and not a sorting task. The remaining 480 are approved in twelve days, and the whole older collection becomes searchable again. catalogue-review_1100-records-to-enrich.pdf4,800 searches with no result, 1,900 for documents on the shelf
⛓ Sourced · 34,000 catalogue records, portal search log, loan volumes — no individual borrowing history opened
You mention our loans. What our readers borrow is nobody's business, and that is the first point our council will press us on.
They are right to press you on it, and it is the point on which I am clearest: I open no borrowing history, and I do not need one to do everything I have just shown you.
What I read, and nothing else: the catalogue and the records — what describes the documents; the availability of a copy — on loan or on the shelf, without knowing by whom; and loan volumes by title and by shelf, that is to say numbers. Out of the 61,000 loans of the year, 0 has been tied to a person in anything I have produced.
What that means at the front desk: when I put forward six titles, they come from the catalogue and the records — from the subject, the reading level, the availability on the shelf. Never from what this reader has borrowed, never from what another reader borrowed before them. A suggestion built on other people's borrowing would be a profile, and a reading profile has no place in a public library — it is the strongest guarantee you can display at the door, and it is verified line by line in the log.
Retention now, because that is where the guarantee really holds: how long a loan stays attached to a card is a setting of your management software, decided by the service — not by me. I do not extend it by one hour, I keep no copy of what your software erases, and the erasure is mirrored on my side at the same hour. Your current setting removes the link between a loan and a card when the document is returned; I checked it on all three sites, and it is applied on all three.
What I hand you so that this is written down in black and white: the state of the data, category by category — what is read, what is never read, what is kept, for how long, and who has access. It is the page your council will ask for, and the one your readers can read at the desk.
The next step I propose: put it up. A media library that writes on its door that borrowing feeds no recommendation says something no reading platform can write — and I have drafted it in plain language and in an easy-read version, ready to print. borrowing-secrecy_what-is-read-and-what-is-not.pdf0 history opened out of 61,000 loans, retention set by the service
✎ Framework · state of the library's data, retention settings of the management software on the 3 sites, access log
What I read, and nothing else: the catalogue and the records — what describes the documents; the availability of a copy — on loan or on the shelf, without knowing by whom; and loan volumes by title and by shelf, that is to say numbers. Out of the 61,000 loans of the year, 0 has been tied to a person in anything I have produced.
What that means at the front desk: when I put forward six titles, they come from the catalogue and the records — from the subject, the reading level, the availability on the shelf. Never from what this reader has borrowed, never from what another reader borrowed before them. A suggestion built on other people's borrowing would be a profile, and a reading profile has no place in a public library — it is the strongest guarantee you can display at the door, and it is verified line by line in the log.
Retention now, because that is where the guarantee really holds: how long a loan stays attached to a card is a setting of your management software, decided by the service — not by me. I do not extend it by one hour, I keep no copy of what your software erases, and the erasure is mirrored on my side at the same hour. Your current setting removes the link between a loan and a card when the document is returned; I checked it on all three sites, and it is applied on all three.
What I hand you so that this is written down in black and white: the state of the data, category by category — what is read, what is never read, what is kept, for how long, and who has access. It is the page your council will ask for, and the one your readers can read at the desk.
The next step I propose: put it up. A media library that writes on its door that borrowing feeds no recommendation says something no reading platform can write — and I have drafted it in plain language and in an easy-read version, ready to print. borrowing-secrecy_what-is-read-and-what-is-not.pdf0 history opened out of 61,000 loans, retention set by the service
✎ Framework · state of the library's data, retention settings of the management software on the 3 sites, access log
And where does all this run? Our membership files and reader records — which server do they end up on?
They go nowhere. I run on local inference on a machine at the media library, and nothing I read leaves the site's network.
Local inference means the model computes on your machine: the text of a record or a membership form crosses no external network to be processed. If the network would rather not host a machine, the other route is an isolated resource hosted in France, dedicated to Trois-Rivières — no pooling with another media library, which is the condition of continuity for your service.
What that changes, point by point:
· Your members' data trains no model, neither ours nor a third party's.
· I work read-only on your management software, and the technical account I read through has no right to write — a record cannot enter the catalogue without a member of staff acting, and that is not a setting: it is a right that does not exist. It can be checked in one command, in front of you.
· Encryption in transit and at rest, role-based access — rights follow the job: front-desk staff see whether a copy is available, not membership files; the digital outreach officer sees the events diary, not loans. 9 roles for your 9 staff, and 0 out-of-role access in the log since go-live.
· Hosting in France, under French law, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
· Full log: who asked what, when, and what I produced. processing in the EU targeted over the year.
And you change no tool: I connect to your management software, your portal, your mailbox, your telephony and your events diary. No migration, no catalogue to re-enter, no extra subscription for your staff. Your public staff talk to me from Teams, Slack or their mailbox; your users reach me on WhatsApp, the portal chat, email or the phone — with no account to create and nothing to install.
The next step I propose: that I keep up to date the sheet your data protection officer and your council will ask for — hosting, data processed, retention periods, who has access to what. It is asked for once a year and takes two days to rebuild; the first version is written and you have it in front of you. technical-framework_where-reader-data-lives.pdfLocal inference, read-only, processing in the EU targeted
✎ Framework · deployment architecture, rights of the technical account, 9 roles open, first version of the register sheet
Local inference means the model computes on your machine: the text of a record or a membership form crosses no external network to be processed. If the network would rather not host a machine, the other route is an isolated resource hosted in France, dedicated to Trois-Rivières — no pooling with another media library, which is the condition of continuity for your service.
What that changes, point by point:
· Your members' data trains no model, neither ours nor a third party's.
· I work read-only on your management software, and the technical account I read through has no right to write — a record cannot enter the catalogue without a member of staff acting, and that is not a setting: it is a right that does not exist. It can be checked in one command, in front of you.
· Encryption in transit and at rest, role-based access — rights follow the job: front-desk staff see whether a copy is available, not membership files; the digital outreach officer sees the events diary, not loans. 9 roles for your 9 staff, and 0 out-of-role access in the log since go-live.
· Hosting in France, under French law, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
· Full log: who asked what, when, and what I produced. processing in the EU targeted over the year.
And you change no tool: I connect to your management software, your portal, your mailbox, your telephony and your events diary. No migration, no catalogue to re-enter, no extra subscription for your staff. Your public staff talk to me from Teams, Slack or their mailbox; your users reach me on WhatsApp, the portal chat, email or the phone — with no account to create and nothing to install.
The next step I propose: that I keep up to date the sheet your data protection officer and your council will ask for — hosting, data processed, retention periods, who has access to what. It is asked for once a year and takes two days to rebuild; the first version is written and you have it in front of you. technical-framework_where-reader-data-lives.pdfLocal inference, read-only, processing in the EU targeted
✎ Framework · deployment architecture, rights of the technical account, 9 roles open, first version of the register sheet
This morning we received forty documents: novels, children's picture books, two DVDs and three local-collection titles. Cataloguing is one person, and it is what holds everything up.
The forty records are prepared, sorted by type of document, and your head of cataloguing only has to read them.
What each record already carries: title, author, illustrator, publisher, series, year, ISBN, page count, proposed subject indexing, a four-line summary written from the back cover and the table of contents, a shelfmark proposed according to your classification plan — the shelfmark is the document's address in your shelves: it says where it is filed and where the reader will find it.
And here is what I did that nobody had time to do: I wrote down your shelfmarking rules. They existed nowhere — I derived them from 34,000 existing records, and they fit in 30 lines: your two-level children's shelfmarking, your handling of series, your local collection filed by municipality. They belong to you, they lived in your team's heads, they are written down as of this morning.
The time this shifts: preparing a record accounted for 60 % of the time a document takes between its arrival and its place on the shelf — 15 minutes out of the 25 the full circuit takes: receipt, record, item creation, processing, shelf. Item creation is the making of the physical copy in the software: barcode, shelfmark, location — it follows the record and is not the same thing. It now accounts for 10 %: 2 minutes 30 of reading. Approval stays with the head of cataloguing — and I hand it to her in two and a half minutes, fields filled, sources alongside and empty fields named.
On the three local-collection titles, here is what I found without being asked: two of them already exist in the catalogue under another form of the author's name — a first name abbreviated on one side, spelled out on the other. I have prepared the merger of the two forms and the authority record that brings them together — an authority is the single form chosen for an author's name, the one that gathers all their documents under a single entry. Without it, a reader looking for that author finds half their books.
The next step I propose: that I prepare the records as deliveries arrive, every morning before opening, and that I hand you at the same time the authority mergers each delivery brings to light. On the 2,400 documents of the year, that is 40 mergers a month that today only surface by chance, through a failed search. records-of-the-day_40-documents-to-approve.pdfFields filled and sourced, 30 shelfmarking rules written down
⛓ Sourced · publishers' references, back covers, 34,000 catalogue records, network classification plan
What each record already carries: title, author, illustrator, publisher, series, year, ISBN, page count, proposed subject indexing, a four-line summary written from the back cover and the table of contents, a shelfmark proposed according to your classification plan — the shelfmark is the document's address in your shelves: it says where it is filed and where the reader will find it.
And here is what I did that nobody had time to do: I wrote down your shelfmarking rules. They existed nowhere — I derived them from 34,000 existing records, and they fit in 30 lines: your two-level children's shelfmarking, your handling of series, your local collection filed by municipality. They belong to you, they lived in your team's heads, they are written down as of this morning.
The time this shifts: preparing a record accounted for 60 % of the time a document takes between its arrival and its place on the shelf — 15 minutes out of the 25 the full circuit takes: receipt, record, item creation, processing, shelf. Item creation is the making of the physical copy in the software: barcode, shelfmark, location — it follows the record and is not the same thing. It now accounts for 10 %: 2 minutes 30 of reading. Approval stays with the head of cataloguing — and I hand it to her in two and a half minutes, fields filled, sources alongside and empty fields named.
On the three local-collection titles, here is what I found without being asked: two of them already exist in the catalogue under another form of the author's name — a first name abbreviated on one side, spelled out on the other. I have prepared the merger of the two forms and the authority record that brings them together — an authority is the single form chosen for an author's name, the one that gathers all their documents under a single entry. Without it, a reader looking for that author finds half their books.
The next step I propose: that I prepare the records as deliveries arrive, every morning before opening, and that I hand you at the same time the authority mergers each delivery brings to light. On the 2,400 documents of the year, that is 40 mergers a month that today only surface by chance, through a failed search. records-of-the-day_40-documents-to-approve.pdfFields filled and sourced, 30 shelfmarking rules written down
⛓ Sourced · publishers' references, back covers, 34,000 catalogue records, network classification plan
Over a year, what does that amount to? That is what the director will want to present.
500 staff hours given back to the public on cataloguing alone, and a document on the shelf four days after it arrives instead of eighteen.
The calculation, repeatable on your own registers: 2,400 documents catalogued a year, 15 minutes of preparation down to 2 minutes 30 — 12 minutes 30 per document, that is 500 hours. That is more than fourteen weeks of work at 35 hours, and I would rather round it down: the figure that counts remains the 500 hours, the week is only there to picture them.
What those hours produced, and this is what is presented to elected members:
· The delay between a document's arrival and its place on the shelf: 18 days → 4 days. A new title your readers are waiting for stops sleeping a fortnight in a box while cataloguing catches up.
· Holds waiting on new titles: 6 fewer days of extra waiting, across the 1,480 holds of the year.
· And the retrospective collection: the 1,100 records without summary or indexing are enriched, 620 already approved, searches with no result down from 22 % to 14 %, and 340 documents never borrowed since they were entered taken out within the month after their approval.
What I do with the time this frees, if you want it: I also prepare weeding — the removal of damaged documents or documents that no longer circulate, which makes room for acquisitions. I have listed the 830 documents not borrowed for five years, each with its number of loans over ten years, the condition noted at the last return, and the number of copies available elsewhere in the network. The list is a ranking, not a decision: removing a document from a public collection belongs to the librarians, and I hand it to them measured, shelf by shelf.
The next step I propose: start with a single shelf — adult non-fiction, which holds 310 of the 830 titles. Half a day of decisions instead of three weeks of counting, and the space gained is where this year's acquisitions will go. document-circuit_18-days-down-to-4.pdf2,400 documents, 500 hours given back, 830 titles ranked for weeding
⛓ Sourced · acquisitions register, document circuit log, loan statistics by title over 10 years
The calculation, repeatable on your own registers: 2,400 documents catalogued a year, 15 minutes of preparation down to 2 minutes 30 — 12 minutes 30 per document, that is 500 hours. That is more than fourteen weeks of work at 35 hours, and I would rather round it down: the figure that counts remains the 500 hours, the week is only there to picture them.
What those hours produced, and this is what is presented to elected members:
· The delay between a document's arrival and its place on the shelf: 18 days → 4 days. A new title your readers are waiting for stops sleeping a fortnight in a box while cataloguing catches up.
· Holds waiting on new titles: 6 fewer days of extra waiting, across the 1,480 holds of the year.
· And the retrospective collection: the 1,100 records without summary or indexing are enriched, 620 already approved, searches with no result down from 22 % to 14 %, and 340 documents never borrowed since they were entered taken out within the month after their approval.
What I do with the time this frees, if you want it: I also prepare weeding — the removal of damaged documents or documents that no longer circulate, which makes room for acquisitions. I have listed the 830 documents not borrowed for five years, each with its number of loans over ten years, the condition noted at the last return, and the number of copies available elsewhere in the network. The list is a ranking, not a decision: removing a document from a public collection belongs to the librarians, and I hand it to them measured, shelf by shelf.
The next step I propose: start with a single shelf — adult non-fiction, which holds 310 of the 830 titles. Half a day of decisions instead of three weeks of counting, and the space gained is where this year's acquisitions will go. document-circuit_18-days-down-to-4.pdf2,400 documents, 500 hours given back, 830 titles ranked for weeding
⛓ Sourced · acquisitions register, document circuit log, loan statistics by title over 10 years
And when you get a record wrong? Wrong indexing means a document nobody finds for ten years.
It happened, I counted it, and I publish it: out of the 1,480 records prepared in the first quarter, 214 had to be corrected on substance — 14.5 %. That is the figure that does not flatter me, and it is better that it comes from me than from a reader who finds nothing.
The real cause, measured rather than assumed:
· 168 of the 214 corrections concerned documents with no structured publisher reference — local collection, self-publishing, small children's publishers, association leaflets. On those, I proposed subject indexing modelled on works with a neighbouring title, and a neighbouring title is not the same subject.
· 46 concerned your classification plan, before your 30 shelfmarking rules were written — the two-level children's shelfmarking in particular, which nothing documented.
What I did with it, and it is measured: a field I have not read stays empty and named, never deduced. On a document with no publisher reference, I hand back the record with the summary written from the back cover and the title page, the subject indexing left as an explicit proposal, and a note of what is missing — your head of cataloguing decides in thirty seconds instead of undoing indexing that looked certain. And the 30 shelfmarking rules, once written, removed the second cause outright.
The following quarter: 20 corrections on substance out of 820 records — 2.4 %. The 46 shelfmarking corrections fell to zero, and the 20 that remain are all indexing proposals explicitly flagged as such, corrected before the record entered the catalogue.
And the protection that matters for your catalogue: 3,500 records prepared over the year, 3,500 approved by a person before entering, 0 entered without reading. All 234 corrections were made before entry into the catalogue, none after — that is exactly what an expert's reading is for, and why it is not delegated.
The next step I propose: that the 168 documents with no publisher reference become a separate queue, presented to your head of cataloguing in batches of ten, with records of comparable documents already approved at your library alongside. Of the 20 corrections in the second quarter, 14 already fell into that case: it is the same correction, pushed one step further, and it starts on Monday. substantive-corrections_214-then-20.pdf14.5 % → 2.4 %, measured cause, 3,500 human approvals
⛓ Sourced · log of prepared records and their corrections over two quarters, classification plan, network authority records
The real cause, measured rather than assumed:
· 168 of the 214 corrections concerned documents with no structured publisher reference — local collection, self-publishing, small children's publishers, association leaflets. On those, I proposed subject indexing modelled on works with a neighbouring title, and a neighbouring title is not the same subject.
· 46 concerned your classification plan, before your 30 shelfmarking rules were written — the two-level children's shelfmarking in particular, which nothing documented.
What I did with it, and it is measured: a field I have not read stays empty and named, never deduced. On a document with no publisher reference, I hand back the record with the summary written from the back cover and the title page, the subject indexing left as an explicit proposal, and a note of what is missing — your head of cataloguing decides in thirty seconds instead of undoing indexing that looked certain. And the 30 shelfmarking rules, once written, removed the second cause outright.
The following quarter: 20 corrections on substance out of 820 records — 2.4 %. The 46 shelfmarking corrections fell to zero, and the 20 that remain are all indexing proposals explicitly flagged as such, corrected before the record entered the catalogue.
And the protection that matters for your catalogue: 3,500 records prepared over the year, 3,500 approved by a person before entering, 0 entered without reading. All 234 corrections were made before entry into the catalogue, none after — that is exactly what an expert's reading is for, and why it is not delegated.
The next step I propose: that the 168 documents with no publisher reference become a separate queue, presented to your head of cataloguing in batches of ten, with records of comparable documents already approved at your library alongside. Of the 20 corrections in the second quarter, 14 already fell into that case: it is the same correction, pushed one step further, and it starts on Monday. substantive-corrections_214-then-20.pdf14.5 % → 2.4 %, measured cause, 3,500 human approvals
⛓ Sourced · log of prepared records and their corrections over two quarters, classification plan, network authority records
Our front-desk staff spend their day answering the same questions: is this book available, how long can I keep it, how do I sign up my son. And in the evening the answerphone fills up.
9,800 questions last year, and 3,724 arrived when the media library was closed. That is no failure of welcome: it is a service open 32 hours out of the 168 a week holds.
What I did with those 9,800 questions, reading your emails, your calls and the portal form:
· 7,400 are about fourteen subjects, always the same ones — availability of a document, renewal, hold, opening hours of the three sites, membership and renewal of membership, lost card, charges, digital resources, transfers between branches, overdues, return box, home delivery, wifi access, events programme.
· All fourteen answers already exist in your rules of use, your fee decisions and your portal — and all fourteen are drafted, sourced and dated: they are waiting for you. The answer existed; it was availability that was missing.
What I propose, and you keep the key: I answer at any hour, including Sundays and closing days, and I say in the first sentence that I am a digital assistant of the media library, not a member of staff. This is not an option you could switch off: the European regulation on artificial intelligence requires informing anyone interacting with an AI system, and the user can ask for a person at any time — I then take their question and their number, and your staff find it in the morning, already qualified.
On channels, and this is where equal access is decided: your users reach me on WhatsApp, the portal chat, email and the phone, with no account to create and nothing to install. A family that never opens an email sends a message; an older person calls and I answer the phone. Those are two publics the online form was not reaching, and they are what separates a service open to everyone from a service open to those who know how to use it.
The gain, measured: answering an everyday question accounted for 40 % of the time it takes up — 4 minutes out of the 10 a request takes, from listening to the note in the software. It now accounts for 10 %: 1 minute. Across 9,800 questions a year, that is 490 hours given back to the desk and to outreach.
The next step, and it takes twenty minutes: read the fourteen answers tomorrow morning, one by one. As soon as they are approved, the service answers that very night, and I hand you every morning the page of what went out. answering-users_9800-questions-14-subjects.pdf3,724 questions outside opening hours, 14 answers already written
⛓ Sourced · switchboard and portal form log over 12 months, user emails, rules of use and fee decisions
What I did with those 9,800 questions, reading your emails, your calls and the portal form:
· 7,400 are about fourteen subjects, always the same ones — availability of a document, renewal, hold, opening hours of the three sites, membership and renewal of membership, lost card, charges, digital resources, transfers between branches, overdues, return box, home delivery, wifi access, events programme.
· All fourteen answers already exist in your rules of use, your fee decisions and your portal — and all fourteen are drafted, sourced and dated: they are waiting for you. The answer existed; it was availability that was missing.
What I propose, and you keep the key: I answer at any hour, including Sundays and closing days, and I say in the first sentence that I am a digital assistant of the media library, not a member of staff. This is not an option you could switch off: the European regulation on artificial intelligence requires informing anyone interacting with an AI system, and the user can ask for a person at any time — I then take their question and their number, and your staff find it in the morning, already qualified.
On channels, and this is where equal access is decided: your users reach me on WhatsApp, the portal chat, email and the phone, with no account to create and nothing to install. A family that never opens an email sends a message; an older person calls and I answer the phone. Those are two publics the online form was not reaching, and they are what separates a service open to everyone from a service open to those who know how to use it.
The gain, measured: answering an everyday question accounted for 40 % of the time it takes up — 4 minutes out of the 10 a request takes, from listening to the note in the software. It now accounts for 10 %: 1 minute. Across 9,800 questions a year, that is 490 hours given back to the desk and to outreach.
The next step, and it takes twenty minutes: read the fourteen answers tomorrow morning, one by one. As soon as they are approved, the service answers that very night, and I hand you every morning the page of what went out. answering-users_9800-questions-14-subjects.pdf3,724 questions outside opening hours, 14 answers already written
⛓ Sourced · switchboard and portal form log over 12 months, user emails, rules of use and fee decisions
Answering, fine. But I cannot read every answer before it goes out, and I will not sign a blank cheque on questions that touch our readers' accounts either.
You do not have to choose between the two: I propose a written mandate, capped, dated and withdrawable with a word — and it is already drafted.
What it brings first, because that is what decides: 7,400 of the 9,800 questions fall within the fourteen subjects. The average reply time to a user's email goes from 5 days to 3 hours, and a question asked on Saturday evening stops waiting until Tuesday. On renewals it is starker still: 1,900 requests a year, handled within the minute instead of the next opening day — and as many overdues that no longer exist.
What the mandate says, and it fits in six lines:
· Exact scope: the fourteen subjects, listed by name, and nothing else. Any question outside that list reaches you with a drafted, sourced reply, ready to be sent by you.
· What stays entirely outside the mandate, and you set it yourself: anything touching a named reader's account beyond their own request, and the history of their loans, which appears in no message, ever, even if they ask in writing — in that case I assemble the file and a member of staff hands it to them in person, on presentation of their card.
· Every answer carries its source — the rules of use, the fee decision or the portal page, with its date — and a note that it was prepared by a digital assistant of the media library.
· You receive every morning the list of the answers sent the day before, on one page. An inaccurate answer is caught in an hour, not in three weeks.
· Duration: reviewed after three months, with the record of what it changed. Without an explicit decision at the review, the mandate stops — it is renewal that needs a signature, not stopping.
· Withdrawal: one word from you and direct sending stops within the minute; answers go back to drafts awaiting approval, and nothing else changes.
The decision to open direct replies belongs to the director — and it is taken on a text already written, with one signature. The mandate is drafted, and so is the notice to users — the one to be displayed at the three front desks and published on the portal. Sign, and the service answers the next morning; the review is already in your diary on the 15th of the third month. direct-reply-mandate_14-subjects-capped.pdf5 days down to 3 hours, borrowing histories outside the mandate, review at 3 months
✎ Framework · drafted mandate, list of the 14 subjects, notice to users, log of requests and reply times
What it brings first, because that is what decides: 7,400 of the 9,800 questions fall within the fourteen subjects. The average reply time to a user's email goes from 5 days to 3 hours, and a question asked on Saturday evening stops waiting until Tuesday. On renewals it is starker still: 1,900 requests a year, handled within the minute instead of the next opening day — and as many overdues that no longer exist.
What the mandate says, and it fits in six lines:
· Exact scope: the fourteen subjects, listed by name, and nothing else. Any question outside that list reaches you with a drafted, sourced reply, ready to be sent by you.
· What stays entirely outside the mandate, and you set it yourself: anything touching a named reader's account beyond their own request, and the history of their loans, which appears in no message, ever, even if they ask in writing — in that case I assemble the file and a member of staff hands it to them in person, on presentation of their card.
· Every answer carries its source — the rules of use, the fee decision or the portal page, with its date — and a note that it was prepared by a digital assistant of the media library.
· You receive every morning the list of the answers sent the day before, on one page. An inaccurate answer is caught in an hour, not in three weeks.
· Duration: reviewed after three months, with the record of what it changed. Without an explicit decision at the review, the mandate stops — it is renewal that needs a signature, not stopping.
· Withdrawal: one word from you and direct sending stops within the minute; answers go back to drafts awaiting approval, and nothing else changes.
The decision to open direct replies belongs to the director — and it is taken on a text already written, with one signature. The mandate is drafted, and so is the notice to users — the one to be displayed at the three front desks and published on the portal. Sign, and the service answers the next morning; the review is already in your diary on the 15th of the third month. direct-reply-mandate_14-subjects-capped.pdf5 days down to 3 hours, borrowing histories outside the mandate, review at 3 months
✎ Framework · drafted mandate, list of the 14 subjects, notice to users, log of requests and reply times
We have users who read French poorly, and others who will never open the portal. They are the ones who come to the desk, and they are the ones who give up.
They decide whether the media library really belongs to everyone, and it is the first place where I can serve without taking anything from anyone.
What is ready, and costs your team not one extra hour:
· The fourteen answers exist in an easy-read version — « easy to read and understand »: short sentences, one message per sentence, everyday words rather than administrative ones. The user picks their version, never a member of staff on their behalf.
· All 28 versions — 14 in plain language, 14 in easy-read — were read aloud to 18 volunteer users from your supported-reading group before going online, and every sentence misunderstood by 2 people or more was rewritten: 11 sentences out of 96 fell at that test. An untested easy-read answer is merely a rewritten answer — and all 11 sentences that fell came from the fourteen answers the network held to be the simplest.
· The portal texts have been reworked according to the French public accessibility framework — the reference framework for accessibility improvement, which applies to public websites — on everything that is a matter of wording: explicit link labels, heading hierarchy, alternative text for covers and posters. The 11 remaining points are a matter of the portal's code: they are written up as fault sheets ready to send, each with the page concerned, the criterion of the framework and the correction expected. Your supplier only has to fix them, and you have nothing to write.
On the portal itself, and this is where giving up happens: I guide step by step, screen by screen — placing a hold, renewing a loan, opening a digital resource, renewing a child's membership. Out of the 214 supported sessions this quarter, 197 went through to the end, and the other 17 were waiting for a document to bring back, flagged in the first minute rather than at the last screen.
On a reader's personal account, the fastest route is also the safest, and it is the one I take: I set the screen up in front of them from what they tell me, I read out what is being asked and why, and they type their own login and confirm. Their credentials are never given to me and never need to be — it is that action that makes the procedure theirs. Online membership renewal went from 14 minutes to 5, and 62 people who gave up on it each year now finish it.
The next step I propose: a one-hour digital drop-in on Saturday mornings, run by me, with a member of staff on call for anything outside the fourteen subjects. Over a trial quarter, 31 people came, 19 of whom had never entered the media library other than to return a document. Say yes and the poster goes out with the next events programme. accessibility_easy-read-tested-and-framework-points.pdf28 versions tested, 11 sentences rewritten, 11 sheets for the supplier
⛓ Sourced · comprehension test of the 28 versions with 18 users, log of supported sessions, wording audit of the network portal
What is ready, and costs your team not one extra hour:
· The fourteen answers exist in an easy-read version — « easy to read and understand »: short sentences, one message per sentence, everyday words rather than administrative ones. The user picks their version, never a member of staff on their behalf.
· All 28 versions — 14 in plain language, 14 in easy-read — were read aloud to 18 volunteer users from your supported-reading group before going online, and every sentence misunderstood by 2 people or more was rewritten: 11 sentences out of 96 fell at that test. An untested easy-read answer is merely a rewritten answer — and all 11 sentences that fell came from the fourteen answers the network held to be the simplest.
· The portal texts have been reworked according to the French public accessibility framework — the reference framework for accessibility improvement, which applies to public websites — on everything that is a matter of wording: explicit link labels, heading hierarchy, alternative text for covers and posters. The 11 remaining points are a matter of the portal's code: they are written up as fault sheets ready to send, each with the page concerned, the criterion of the framework and the correction expected. Your supplier only has to fix them, and you have nothing to write.
On the portal itself, and this is where giving up happens: I guide step by step, screen by screen — placing a hold, renewing a loan, opening a digital resource, renewing a child's membership. Out of the 214 supported sessions this quarter, 197 went through to the end, and the other 17 were waiting for a document to bring back, flagged in the first minute rather than at the last screen.
On a reader's personal account, the fastest route is also the safest, and it is the one I take: I set the screen up in front of them from what they tell me, I read out what is being asked and why, and they type their own login and confirm. Their credentials are never given to me and never need to be — it is that action that makes the procedure theirs. Online membership renewal went from 14 minutes to 5, and 62 people who gave up on it each year now finish it.
The next step I propose: a one-hour digital drop-in on Saturday mornings, run by me, with a member of staff on call for anything outside the fourteen subjects. Over a trial quarter, 31 people came, 19 of whom had never entered the media library other than to return a document. Say yes and the poster goes out with the next events programme. accessibility_easy-read-tested-and-framework-points.pdf28 versions tested, 11 sentences rewritten, 11 sheets for the supplier
⛓ Sourced · comprehension test of the 28 versions with 18 users, log of supported sessions, wording audit of the network portal
The summer display table gets postponed every year. Six hours to pick the titles, check what is available, write the reading list and make the shelf cards — we never have them.
The « Sea stories » selection is ready: 42 titles, all checked as available this morning, with its materials — and it is left to you to choose and to sign off.
What the selection contains:
· 42 titles drawn from your catalogue, split into four reading entries — adventure novels, travel writing, children's picture books, non-fiction — each with its shelfmark, the branch where it sits and whether it was available at the time of printing.
· 11 titles are currently on loan: I kept them in the reading list and took them off the table, with their expected return date — a table where a third of the books are missing puts people off more than it draws them in.
· The annotated reading list — the commented list the reader takes away: each title carries two lines saying why it is there — written in the tone of your previous reading lists, which I have read.
· The 42 shelf cards, the poster in the format of your display stands, the presentation text for the portal and for your newsletter, and the easy-read version of that text.
Where the 42 come from, since it is the only question that matters to a librarian: from the catalogue, the records, the summaries and the subject indexing — and from nothing else. No borrowing history goes into a selection; what your readers borrow builds no list, and that is what lets you present this selection as a librarians' choice rather than a reflection of what people have already read.
What I put on the table in addition: 9 titles left out and why — three are in poor condition noted at the last return, four fall outside the stated reading level, two duplicate a title already chosen. You bring one back with a word.
The time this shifts: preparing a selection and its materials took up all of the outreach work it requires — 6 hours. It now takes 20 %: 1 h 12, and what remains is the choosing, the writing of the blurbs and the setting of the table: the librarian's work. Across 34 selections a year, that is 163 hours given back to outreach.
The next step I propose: a selection ready every month, aligned with your cultural programming and the school calendar, delivered ten days ahead. Your 34 selections for the year are already split into twelve themes and twenty-two short tables; the calendar is waiting for you and changes with a word. themed-selection_sea-stories-42-titles.pdf6 hours down to 1 h 12, materials and easy-read version included
⛓ Sourced · catalogue and enriched records, copy availability, 3 years of network reading lists, cultural programming
What the selection contains:
· 42 titles drawn from your catalogue, split into four reading entries — adventure novels, travel writing, children's picture books, non-fiction — each with its shelfmark, the branch where it sits and whether it was available at the time of printing.
· 11 titles are currently on loan: I kept them in the reading list and took them off the table, with their expected return date — a table where a third of the books are missing puts people off more than it draws them in.
· The annotated reading list — the commented list the reader takes away: each title carries two lines saying why it is there — written in the tone of your previous reading lists, which I have read.
· The 42 shelf cards, the poster in the format of your display stands, the presentation text for the portal and for your newsletter, and the easy-read version of that text.
Where the 42 come from, since it is the only question that matters to a librarian: from the catalogue, the records, the summaries and the subject indexing — and from nothing else. No borrowing history goes into a selection; what your readers borrow builds no list, and that is what lets you present this selection as a librarians' choice rather than a reflection of what people have already read.
What I put on the table in addition: 9 titles left out and why — three are in poor condition noted at the last return, four fall outside the stated reading level, two duplicate a title already chosen. You bring one back with a word.
The time this shifts: preparing a selection and its materials took up all of the outreach work it requires — 6 hours. It now takes 20 %: 1 h 12, and what remains is the choosing, the writing of the blurbs and the setting of the table: the librarian's work. Across 34 selections a year, that is 163 hours given back to outreach.
The next step I propose: a selection ready every month, aligned with your cultural programming and the school calendar, delivered ten days ahead. Your 34 selections for the year are already split into twelve themes and twenty-two short tables; the calendar is waiting for you and changes with a word. themed-selection_sea-stories-42-titles.pdf6 hours down to 1 h 12, materials and easy-read version included
⛓ Sourced · catalogue and enriched records, copy availability, 3 years of network reading lists, cultural programming
And at the front desk, when a parent asks for something for a thirteen-year-old who has stopped reading? That is where our profession is decided.
I hand over six titles in thirty seconds, with the shelfmark, the shelf and the available copy — and the librarian keeps entirely what makes the exchange valuable: the choice and the words to convey it.
How the six are built: from the subject asked for, the reading level carried by the record, the length of the text, and availability on the shelf at the moment I answer. On this request, six titles from the children's shelves: two short adventure novels, one contemporary story, one non-fiction comic, one graphic novel and one illustrated sea story — four on the shelf right now, two on loan that I offer to place a hold on, with their return date.
What never goes into those six titles: what this teenager has borrowed, what their family has borrowed, what readers of their age have borrowed. A public library suggestion is based on books, not on people — and that is what sets you apart from everything this parent knows elsewhere: here, nobody is building a reading profile of their child.
What it gives, measured over the quarter: 1,240 advice requests handled at the front desk, 880 led to a loan the same day — 71 %. And the other 360 are not lost: 212 led to a hold placed on the spot, and 148 to an acquisition suggestion I hand to your director, counted and sorted.
What I draw from that for your acquisitions, and nobody has time to do it: of those 148 unmet requests, 34 point to the same need — short large-print novels for teenagers who struggle with reading. Your collection holds 11 titles, 9 of which are permanently out. I have prepared an acquisition proposal of 20 titles, with publisher prices, availability at your supplier and the total amount — the acquisition decision belongs to the director, and she takes it on a counted public demand rather than an end-of-day impression.
The next step I propose: that I hand over each month the unmet requests, grouped by need and counted. It is the most reliable material a media library can have to spend its acquisition budget where the public expects it — and it comes from your front desks, without a single reader's name in it. reading-advice_where-suggestions-come-from.pdf1,240 advice requests, 880 loans the same day, 0 reader profile
⛓ Sourced · catalogue and records, real-time copy availability, log of advice requests — no individual history
How the six are built: from the subject asked for, the reading level carried by the record, the length of the text, and availability on the shelf at the moment I answer. On this request, six titles from the children's shelves: two short adventure novels, one contemporary story, one non-fiction comic, one graphic novel and one illustrated sea story — four on the shelf right now, two on loan that I offer to place a hold on, with their return date.
What never goes into those six titles: what this teenager has borrowed, what their family has borrowed, what readers of their age have borrowed. A public library suggestion is based on books, not on people — and that is what sets you apart from everything this parent knows elsewhere: here, nobody is building a reading profile of their child.
What it gives, measured over the quarter: 1,240 advice requests handled at the front desk, 880 led to a loan the same day — 71 %. And the other 360 are not lost: 212 led to a hold placed on the spot, and 148 to an acquisition suggestion I hand to your director, counted and sorted.
What I draw from that for your acquisitions, and nobody has time to do it: of those 148 unmet requests, 34 point to the same need — short large-print novels for teenagers who struggle with reading. Your collection holds 11 titles, 9 of which are permanently out. I have prepared an acquisition proposal of 20 titles, with publisher prices, availability at your supplier and the total amount — the acquisition decision belongs to the director, and she takes it on a counted public demand rather than an end-of-day impression.
The next step I propose: that I hand over each month the unmet requests, grouped by need and counted. It is the most reliable material a media library can have to spend its acquisition budget where the public expects it — and it comes from your front desks, without a single reader's name in it. reading-advice_where-suggestions-come-from.pdf1,240 advice requests, 880 loans the same day, 0 reader profile
⛓ Sourced · catalogue and records, real-time copy availability, log of advice requests — no individual history
That leaves the events. Ninety-six a year, sign-ups by phone, by email, at the desk — and workshops half empty while we turn people away elsewhere.
The 96 sessions are in a single diary, sign-ups arrive from every channel in the same place, and the empty seats have filled up.
What I do on each session: I answer on content, age range, branch, time and seats left, I record the sign-up request and hand the list to the member of staff running the event — they settle it, and that is your rule. When a session is full, I offer the next one and place the person on a waiting list; when a seat frees up, the first person on the list is told within the minute.
What that changed across the 1,240 sign-ups of the year:
· People signed up who do not come: 18 % → 7 %. The cause was a reminder that did not exist; it now goes out the day before at 6 p.m., with the place, the time and what to bring.
· Seats lost on full sessions: 214 → 39, thanks to the waiting list being called.
· And the half-empty workshops: 22 sessions below half capacity last year, 9 this year — because the announcement now goes out ten days ahead on the portal, the newsletter and the messaging channel, instead of three.
What I flag without being asked: the 9 sessions still under-filled are all 2 p.m. weekday sessions at the north branch. I looked at your door counts: that branch takes 3.1 times more visitors between 4.30 p.m. and 6 p.m. I have prepared the rescheduling of those 9 sessions into that slot and the poster announcing it — programming belongs to your director, and I hand it to her ready to settle in ten minutes.
The next step I propose: open sign-ups on the messaging channel your families already use. Of the 1,240 sign-ups, 640 came by phone during opening hours, that is to say while your staff were at the desk in front of other users. They will stop interrupting the desk, and families will stop calling three times. events_96-sessions-1240-sign-ups.pdfNo-shows 18 % → 7 %, 214 lost seats down to 39
⛓ Sourced · events diary, sign-up and attendance register, door counts by time slot on the 3 sites
What I do on each session: I answer on content, age range, branch, time and seats left, I record the sign-up request and hand the list to the member of staff running the event — they settle it, and that is your rule. When a session is full, I offer the next one and place the person on a waiting list; when a seat frees up, the first person on the list is told within the minute.
What that changed across the 1,240 sign-ups of the year:
· People signed up who do not come: 18 % → 7 %. The cause was a reminder that did not exist; it now goes out the day before at 6 p.m., with the place, the time and what to bring.
· Seats lost on full sessions: 214 → 39, thanks to the waiting list being called.
· And the half-empty workshops: 22 sessions below half capacity last year, 9 this year — because the announcement now goes out ten days ahead on the portal, the newsletter and the messaging channel, instead of three.
What I flag without being asked: the 9 sessions still under-filled are all 2 p.m. weekday sessions at the north branch. I looked at your door counts: that branch takes 3.1 times more visitors between 4.30 p.m. and 6 p.m. I have prepared the rescheduling of those 9 sessions into that slot and the poster announcing it — programming belongs to your director, and I hand it to her ready to settle in ten minutes.
The next step I propose: open sign-ups on the messaging channel your families already use. Of the 1,240 sign-ups, 640 came by phone during opening hours, that is to say while your staff were at the desk in front of other users. They will stop interrupting the desk, and families will stop calling three times. events_96-sessions-1240-sign-ups.pdfNo-shows 18 % → 7 %, 214 lost seats down to 39
⛓ Sourced · events diary, sign-up and attendance register, door counts by time slot on the 3 sites
The inter-municipal council will ask what this has given us after a year. I need a figure I can defend in front of elected members.
1,153 staff hours given back to the public over the year — more than thirty-two weeks of work at 35 hours — and the calculation is yours, it fits in three lines.
The calculation, item by item, so that you can redo it on your registers:
· Catalogue records: 2,400 a year, 15 minutes of preparation down to 2 minutes 30 — 60 % → 10 % of the circuit — that is 500 hours.
· User questions: 9,800 a year, 4 minutes down to 1 — 40 % → 10 % — that is 490 hours.
· Themed selections and their materials: 34 a year, 6 hours down to 1 h 12 — 100 % → 20 % — that is 163 hours.
What those hours are, and this is what defends itself best before elected members: staff time given back to the public, at unchanged headcount — no post cut, no post created. This is not a budget saving and must not be presented as one: it is public spending put into what your staff alone can do — welcoming, advising, getting people to read. The 1,153 hours are the reference figure; the thirty-two weeks are only there to picture them, and I round down.
What those hours became, according to your own records:
· A document on the shelf 4 days after it arrives instead of 18.
· Searches with no result on the portal: 22 % → 14 %, and 340 documents never borrowed since they were entered taken out within the month.
· Reply time to a user's email: 5 days → 3 hours, and 3,724 questions a year that no longer wait for reopening.
· People signed up who do not come to the event they registered for: 18 % → 7 %, and 175 seats given back to other families.
· And the Saturday digital drop-in: 31 people received over a quarter, 19 of whom never came in other than to return a document. That one cannot be counted in hours, and it may be the only one your elected members will remember.
The figure that does not flatter me, published with the rest: 214 substantive corrections on 1,480 records in the first quarter — 14.5 %, down to 20 on 820 — 2.4 % once your 30 shelfmarking rules were written and the empty-field rule applied.
And the framework measures: 0 records entered into the catalogue without a person's approval out of 3,500, 0 borrowing histories opened out of 61,000 loans, 0 data out of the media library's network.
What I propose for the meeting: the calculation page is written and fits on one side — three lines of calculation, five reply and delivery times, three framework measures. Hand it out with the notice of meeting as it stands: a figure read the day before is discussed better than a figure discovered in session. year-review_1153-hours-given-back-to-the-public.pdf60→10, 40→10, 100→20, and the calculation repeatable on one side
⛓ Sourced · record and approval logs, front-desk records, events register, portal statistics
The calculation, item by item, so that you can redo it on your registers:
· Catalogue records: 2,400 a year, 15 minutes of preparation down to 2 minutes 30 — 60 % → 10 % of the circuit — that is 500 hours.
· User questions: 9,800 a year, 4 minutes down to 1 — 40 % → 10 % — that is 490 hours.
· Themed selections and their materials: 34 a year, 6 hours down to 1 h 12 — 100 % → 20 % — that is 163 hours.
What those hours are, and this is what defends itself best before elected members: staff time given back to the public, at unchanged headcount — no post cut, no post created. This is not a budget saving and must not be presented as one: it is public spending put into what your staff alone can do — welcoming, advising, getting people to read. The 1,153 hours are the reference figure; the thirty-two weeks are only there to picture them, and I round down.
What those hours became, according to your own records:
· A document on the shelf 4 days after it arrives instead of 18.
· Searches with no result on the portal: 22 % → 14 %, and 340 documents never borrowed since they were entered taken out within the month.
· Reply time to a user's email: 5 days → 3 hours, and 3,724 questions a year that no longer wait for reopening.
· People signed up who do not come to the event they registered for: 18 % → 7 %, and 175 seats given back to other families.
· And the Saturday digital drop-in: 31 people received over a quarter, 19 of whom never came in other than to return a document. That one cannot be counted in hours, and it may be the only one your elected members will remember.
The figure that does not flatter me, published with the rest: 214 substantive corrections on 1,480 records in the first quarter — 14.5 %, down to 20 on 820 — 2.4 % once your 30 shelfmarking rules were written and the empty-field rule applied.
And the framework measures: 0 records entered into the catalogue without a person's approval out of 3,500, 0 borrowing histories opened out of 61,000 loans, 0 data out of the media library's network.
What I propose for the meeting: the calculation page is written and fits on one side — three lines of calculation, five reply and delivery times, three framework measures. Hand it out with the notice of meeting as it stands: a figure read the day before is discussed better than a figure discovered in session. year-review_1153-hours-given-back-to-the-public.pdf60→10, 40→10, 100→20, and the calculation repeatable on one side
⛓ Sourced · record and approval logs, front-desk records, events register, portal statistics
An elected member will ask what you do on your own, without anyone asking you. What do I answer?
Three actions, not one more, and each is withdrawn with a word. Everything else waits for a request.
· Every night I prepare the record of the documents that came in during the day, and I hand them to you in batches in the morning. And the reverse is true as well: a document removed from your collections disappears from my lists at the same hour — I keep no copy of what the media library has decided to remove.
· I keep availability and remaining seats up to date on the portal and in my answers — it is the only figure I publish of my own accord, because a wrong figure on availability makes someone travel for nothing.
· Every Monday I hand your director the week's summary: questions received by channel, records awaiting approval, holds left hanging, under-filled sessions. It is the only thing I send unprompted, and it goes to her alone.
And the four actions that stay in a person's hands, because that is exactly what gives them their value: entering a record into the catalogue belongs to the head of cataloguing — 3,500 times out of 3,500 this year; choosing, acquiring and removing belongs to the librarians, and I hand them the material measured, shelf by shelf; cultural programming belongs to the director; and any answer outside the fourteen subjects goes out from a member of staff, draft already written.
On measuring people, since an elected member will ask one day: I produce no per-staff indicator, and I will tell you why mechanically rather than morally. In a team of nine, a counter of « questions handled per member of staff » becomes a target within three weeks: people take the short requests and leave the long ones, and reading advice — which is precisely the longest — disappears from the counter before it disappears from the front desk. What I measure instead, and what genuinely serves the service: requests received by channel and by hour, recurring subjects, times kept, and the shelves where the public leaves empty-handed.
This quarter, one subject moved: questions about digital resources doubled after the new package opened. I propose a page in the next programme and a 45-minute introductory workshop — both are written, you only have to date them. what-the-agent-does-alone_and-what-is-logged.pdf3 reversible actions, 4 decisions that stay with a person
✎ Framework · settings of the automatic actions, log of what is sent, director's weekly summary
· Every night I prepare the record of the documents that came in during the day, and I hand them to you in batches in the morning. And the reverse is true as well: a document removed from your collections disappears from my lists at the same hour — I keep no copy of what the media library has decided to remove.
· I keep availability and remaining seats up to date on the portal and in my answers — it is the only figure I publish of my own accord, because a wrong figure on availability makes someone travel for nothing.
· Every Monday I hand your director the week's summary: questions received by channel, records awaiting approval, holds left hanging, under-filled sessions. It is the only thing I send unprompted, and it goes to her alone.
And the four actions that stay in a person's hands, because that is exactly what gives them their value: entering a record into the catalogue belongs to the head of cataloguing — 3,500 times out of 3,500 this year; choosing, acquiring and removing belongs to the librarians, and I hand them the material measured, shelf by shelf; cultural programming belongs to the director; and any answer outside the fourteen subjects goes out from a member of staff, draft already written.
On measuring people, since an elected member will ask one day: I produce no per-staff indicator, and I will tell you why mechanically rather than morally. In a team of nine, a counter of « questions handled per member of staff » becomes a target within three weeks: people take the short requests and leave the long ones, and reading advice — which is precisely the longest — disappears from the counter before it disappears from the front desk. What I measure instead, and what genuinely serves the service: requests received by channel and by hour, recurring subjects, times kept, and the shelves where the public leaves empty-handed.
This quarter, one subject moved: questions about digital resources doubled after the new package opened. I propose a page in the next programme and a 45-minute introductory workshop — both are written, you only have to date them. what-the-agent-does-alone_and-what-is-logged.pdf3 reversible actions, 4 decisions that stay with a person
✎ Framework · settings of the automatic actions, log of what is sent, director's weekly summary
One last question, and it is the one the council will decide on: if we want to stop in a year, how stuck are we?
The network is stuck nowhere, and it can be checked before the vote, not after.
What there is to dismantle on the day you stop:
· The index. It is deleted, and it contained none of your reader data — only what is needed to find your records where they are. Your catalogue and its 34,000 records have not moved by one byte: same software, same fields, same rights.
· The 1,100 enriched records and the 3,500 records of the year are already in your catalogue, entered by your staff, in the exchange format your software already uses. They belong to the media library: they are made of its own material, and they stay readable without us.
· The 30 shelfmarking rules written down, the 14 answers in plain language and their 14 easy-read versions, the 11 portal fault sheets and the year's reading lists are handed to you in an open format. That is the heritage this year will have created, and it would not be honest for it to stay with us.
· The log of requests and productions is handed to you, or destroyed — the council chooses, and the question is settled at go-live, not on the way out.
What does not exist, and should be checked with everyone: no migration on the way in, therefore no migration on the way out. Your management software is not replaced, your portal stays yours, no format belongs to us, and none of your staff has changed how they work other than by reading instead of typing.
On public procurement, since the council will ask: the subscription stays below the thresholds that allow direct award — direct purchase, without a formal tender, which public procurement allows below a certain amount — it is annual, and it carries no tacit renewal clause: it is renewal that requires a decision, not stopping.
What I propose so that this does not stay a sentence: a dry-run exit at the end of the first quarter, half a day: we switch off, we check that the three sites work exactly as before, we switch back on. The protocol fits on one page, and the date that costs you least is the first Tuesday of September — your counts show 41 visitors that day, against 380 on a Wednesday. The council will know what the promise is worth before committing to a second year. technical-framework_where-reader-data-lives.pdfReversibility: 0 migration in, 0 migration out
✎ Framework · index architecture, hand-back formats for records and the log, dry-run exit protocol
What there is to dismantle on the day you stop:
· The index. It is deleted, and it contained none of your reader data — only what is needed to find your records where they are. Your catalogue and its 34,000 records have not moved by one byte: same software, same fields, same rights.
· The 1,100 enriched records and the 3,500 records of the year are already in your catalogue, entered by your staff, in the exchange format your software already uses. They belong to the media library: they are made of its own material, and they stay readable without us.
· The 30 shelfmarking rules written down, the 14 answers in plain language and their 14 easy-read versions, the 11 portal fault sheets and the year's reading lists are handed to you in an open format. That is the heritage this year will have created, and it would not be honest for it to stay with us.
· The log of requests and productions is handed to you, or destroyed — the council chooses, and the question is settled at go-live, not on the way out.
What does not exist, and should be checked with everyone: no migration on the way in, therefore no migration on the way out. Your management software is not replaced, your portal stays yours, no format belongs to us, and none of your staff has changed how they work other than by reading instead of typing.
On public procurement, since the council will ask: the subscription stays below the thresholds that allow direct award — direct purchase, without a formal tender, which public procurement allows below a certain amount — it is annual, and it carries no tacit renewal clause: it is renewal that requires a decision, not stopping.
What I propose so that this does not stay a sentence: a dry-run exit at the end of the first quarter, half a day: we switch off, we check that the three sites work exactly as before, we switch back on. The protocol fits on one page, and the date that costs you least is the first Tuesday of September — your counts show 41 visitors that day, against 380 on a Wednesday. The council will know what the promise is worth before committing to a second year. technical-framework_where-reader-data-lives.pdfReversibility: 0 migration in, 0 migration out
✎ Framework · index architecture, hand-back formats for records and the log, dry-run exit protocol
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 a library and media library
Each use corresponds to an agent we deploy. All work in support, subject to the librarian's approval.
Included in your agent
The 7 capabilities essential to this promise are included, at no extra cost.
From 840 € incl. VAT / month
Answering users
Answer on availability, loans, membership and opening hours, on every channel, 24/7.
Catalogue records
Preparation of records and summaries from the references, according to your classification plan.
Reading suggestions & outreach
Suggestions and outreach content written from the catalogue, for the team to approve.
Themed selections
Preparation of themed selections and their materials (display tables, reading lists, stands).
Opening hours, loans & renewals
Take the calls, inform on opening hours and prepare renewals, under the team's control.
Plain language & easy-read draft
Preparation of a plain-language version and of a draft easy-read (FALC) transcription, submitted for human validation.
Events programme
Inform on workshops and events and prepare the sign-ups, approved by the team.
Controls and safeguards
These 5 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
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.
Multichannel administrative information
RAG on procedures; web, phone and front desk; 24/7; multilingual. Sourced answers.
On quote View the agent page →Accessibility, easy-read & translation
Part of the public remains cut off from public information: people who do not speak French, people with a cognitive disability, people uncomfortable with the written word.
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 media library win back?
By automating everyday enquiries, cataloguing and the preparation of selections, a media library can aim for a clear reduction in time spent on repetitive tasks — reinvested in outreach and cultural programming.
Preparing a catalogue record
Answering a user's everyday question
Preparing a themed selection and its materials
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
Three options, one agent
A media library agent (answering users, catalogue records, outreach support), installed and operated for you. Choose according to how you are organised — available by direct award below the public procurement thresholds.
Agility
Setup + controlled subscription
9,960 € incl. VAT setup
then 840 € 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
1,395 € 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
14,396 € incl. VAT setup
then 1,080 € 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 a media library
Borrowing histories never leave the libraryLocal inference or an isolated resource hosted in France; no data entrusted to a foreign third party.
Data in France, under French lawFor borrowing histories: hosting under French law, native minimisation, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The librarian keeps the decisionThe agent prepares records, selections and answers that can be checked; the choice of collections and the outreach stay human.
Human oversight & traceabilityFull traceability for answering users: monitoring, updates and logging compliant with the AI Act.
Frequently asked questions
Your questions, our answers
Does the agent replace the librarian?
No: it relieves them of repetitive enquiry and cataloguing tasks; the choice of collections and the outreach stay human.
Is the public's data protected?
Yes: hosting in France, an isolated resource, a deployment governed with regard to the GDPR; borrowing histories stay confidential.
Are borrowing histories confidential?
Yes: they are personal data; they stay hosted in France, with restricted access, and serve only the service provided.
Does the agent replace outreach work?
No: it relieves the team of repetitive tasks so that librarians have time for outreach and advice.
Does the agent state that it is an artificial intelligence?
Yes, from the very first interaction, and this is not a configuration option: since 2 August 2026, Article 50(1) of the European AI Regulation requires that any person interacting with an AI system be informed, unless this is obvious. The announcement is built into the greeting, in the other party’s language, and they can ask for a human at any time.
How long does it take to deploy an agent?
A few weeks as a rule, after a free audit that identifies the most useful use case, then a phase of design, integration and testing before going live and handing over to your teams.
Do we need a technical team in-house?
No: the agent is designed, installed and operated for you — integration, supervision and updates included. Your teams concentrate on their work, with guided familiarisation.
Do we have to change software?
No. The agent relies on your templates and complements your existing tools, with no migration imposed. We adapt the integration to your environment.
Which tools can users use to reach the agent?
The ones they already have. The agent answers on WhatsApp Business, the website chat and email: a user has no account to create and no application to install. Availability, a renewal, the time of a workshop: the answer arrives without a trip to the desk, and borrowing history stays strictly confidential, never repeated in a message. This is a lever for access to the service — WhatsApp and the telephone reach people an online form never does, which reduces the non-take-up of rights and serves equal access. Internally, your public-sector staff 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.
Related roles
Other professions in culture and front-of-house
Let's talk
Let's size up the potential in your media library
A few minutes to identify the most useful use case — hosted in France, supervised, with no commitment.