Library Systems and Retrieval Agents

A university librarian in Coimbatore had already paid for the databases. What she had not paid for was a public chatbot that answered a student's question by quoting a paywalled review article the campus licensed, then sending the query string to a vendor that keeps prompts. The student thought the library had become generous. The publisher's terms, when the librarian reread them, were not.

This is a field guide for librarians, INFLIBNET node officers, research deans and computer-centre heads. The useful agent sits on top of the catalogue you run — often Koha or another ILS — and the collections you have licensed or to which INFLIBNET, e-ShodhSindhu, N-LIST or Shodhganga already admit you. It retrieves. It cites the record. It does not pretend a large model is a substitute for a database invoice.

Not legal advice and not a copyright opinion on any one publisher contract. Read the licence you signed. We will tell you which questions to take to that page.

What the library already is

A campus library is a set of systems of record: bibliographic records, item status, circulation, fines, reserves, and licensed electronic collections. INFLIBNET's public services — including union catalogues, Shodhganga theses, and consortium access routes such as e-ShodhSindhu and N-LIST — are national infrastructure many campuses already use. Your agent should know those names as sources you configured, not as websites it casually scrapes.

Circulation logs are personal data. Who borrowed a book on caste, sexuality or a medical topic is not a feature for a placement agent. The librarian is the custodian. A retrieval agent for search is not entitled to the circulation table.

Theses in Shodhganga have their own deposit and access terms. An agent that embeds every thesis into a vendor vector host 'for better search' may be a deposit-term problem and a DPDP problem. Do that only with a written purpose, a local index, and a training ban.

Search is not borrowing, and borrowing is not a profile.

Object	Agent may	Agent may not
OPAC / ILS bibliographic records	Retrieve and cite the record, show availability if the ILS allows	Rewrite the catalogue into a chatty essay that hides the call number
Licensed full text	Retrieve inside the rights the licence grants, usually click-through to the official copy	Paste publisher PDF into a public model or a shared embedding host
Open campus repository / open theses where deposit allows	Index locally, cite, link	Train a generative model that emits long passages as if they were new
Circulation, reading-room logs, fine lists	Tell a logged-in student their own holds and fines	Build a reading profile for any other office
Inter-library loan and document delivery	Raise a request in the existing form	Fetch a PDF from a shadow library because the model 'knows a link'

Retrieval, not a replacement collection

Students will ask the agent to 'just give me the paper'. If the campus licensed it, the right answer is a link through the official access route. If the campus did not, the right answer is the inter-library form or a no. A model that quotes three pages from a book you do not hold is not a library. It is an unlicensed collection with a smile.

Citations must be real records. Generated bibliographies that invent volumes are now a standard failure mode. The agent should only emit a citation that matches an identifier in your index — ISBN, DOI, accession, Shodhganga handle. If it cannot, it says it cannot.

Indic scripts and thesis transliteration are ordinary. A Tamil-language query for a record catalogued in English is the job. Do not buy a library agent that only demos in English abstracts of American journals.

Licences, training bans and the publisher letter you do not want

Publisher licences differ. Some forbid systematic downloading. Some forbid text-and-data mining. Some allow research TDM on campus. Almost none allow you to ship their corpus to a foreign model host so a vendor can improve a product. Put a training ban in the processor contract. Put a no-exfiltration rule on the retrieval pipeline.

On-prem or the campus data centre is the default for any index that includes licensed text or student theses with personal acknowledgements. Acknowledgements name families and supervisors. They are not 'just metadata'.

If a publisher asks how you search, you should be able to say: queries stay here, we link to you, we do not train. If you cannot say that, do not switch the agent on.

Staff hours, not chatbot hours

Reference interviews for systematic reviews, copyright questions, and research-data management stay human. The agent handles holdings, hours, fine status, and 'do we have this DOI'. Measure deflection of those, not 'engagement'.

A 2 a.m. chat that answers a holdings question is a win. A 2 a.m. chat that writes a literature review is an integrity incident the librarian will be blamed for.

Hours, fines and the night desk

Opening hours and holiday lists are the easiest retrieval win and the easiest stale fact. Pin this week's notice. A fluent 'we are open till 8' from last semester is how a first-year waits outside a locked door in the rain. Treat hours like a fee amount: pass through, do not paraphrase.

Fines are money. The agent may show a logged-in student their own fine and the rule that produced it. It may not waive, and it may not discuss another student's overdue. Waivers sit with the librarian under the library committee's note. Generated kindness is an unofficial write-off.

If you offer a night desk, staff it or say you do not have one. An agent that offers '24x7 reference' and then writes a literature review at 2 a.m. has expanded the library's duty without expanding its people. The integrity policy in this cluster is the other half of that sentence.

Objections you will hear — and what to do with them

These are the lines that stall the file. Answer them in the room, then put the answer in the note. A spoken answer without paper will be forgotten by the next officer.

Students already paste papers into ChatGPT.

They do. That is their integrity problem and their account. It is not a reason for the library to become the pasteboard. Institutional processing is the fiduciary's act.

We will be left behind if we only do OPAC.

Then add licensed retrieval where the licence allows, locally. Being left behind is not a defence to a publisher letter or a DPDP complaint.

Embedding is not copying.

Embeddings of identifiable or licensed text are still processing, and may still be a licence problem. Do not take a vendor's metaphysics over counsel and the contract. If you embed, embed locally, and do not train the foundation model on the result.

Shodhganga is public so we can do anything.

Public access is not a training licence and not a transfer licence. Read the deposit terms and the personal data in acknowledgements. Public is a starting fact, not a holiday.

A four-week library retrieval pilot

Catalogue first. Licensed full text only after the librarian and counsel read the licence.

1. Week 1: inventory ILS, repositories, consortium access, and publisher licences that mention TDM or systematic download.
2. Week 2: stand up on-prem retrieval over bibliographic records only. Citation equals identifier.
3. Week 3: staff reference desk uses it beside the OPAC. Log failures: script, thesis titles, conference papers.
4. Week 4: decide whether any licensed full text may be indexed locally. If the licence is silent or hostile, stop at the link.

How this shows up in the file

Library committee note: corpus list, licence row, on-prem index, no circulation profiles, no training, no invented citations, reference interviews remain human.

Attach the publisher clauses you relied on. A clause you did not read is not a clause you can cite later.

This article is informational field guidance for Indian public universities and colleges, not legal, regulatory, academic, labour or procurement advice. Confirm against the live AICTE / UGC / NAAC notice, the DPDP Act and Rules as commenced, your Act and statutes, reservation orders, labour counsel and the syndicate before you file it.

How this works on an Indian campus

A P3 University should be able to run “Library Systems and Retrieval Agents” without importing a US playbook. “library AI search university” hits UGC/AICTE/NAAC clocks, exam secrecy, reservation rules, and students who may be minors.

Library AI is retrieval over what you already licensed, plus the OPAC you already run. It is not a free replacement for a database you did not buy. DPDP applies to student personal data. Chatbots are not a strategy. Exam and admissions writes stay with officers. Affiliated colleges need isolation, not one shared index.

• No production student data in a vendor SaaS sandbox.
• Write the academic integrity policy before the tool.
• Consent and purpose tags on student-facing agents.
• Budget for staff training, not only licences.

Close this loop before the next CAB

Put “Library Systems and Retrieval Agents” on the next change-advisory or bid-opening agenda as a single line item with an owner. If it cannot earn a line item, it will not earn a control. The owner should be a P3 University, not “the vendor.”

Revisit the item when the model, the GeM term, the region, or the SI changes. “library AI search university” is not a one-time workshop. It is a watch item. Date the last check. Unsigned watch items are souvenirs.