Document library (Sources)

The document library — Sources in your navigation — is where you put the material your content should be built from: a syllabus, a product manual, a policy handbook, a compliance standard, a set of slides, a scanned textbook. You upload a file once, Lurno reads it, summarizes it, and indexes it. From then on it can be used as a knowledge base: attach it to a program so every AI generation inside that program draws on it, or open it and ask questions and get answers with citations back to the page or section they came from.

This is deliberately separate from Resources & media. Media files are things learners see. Source documents are things Lurno reads — reference material that shapes what gets generated, whether or not a learner ever opens the original file.

What this does

Uploading a document starts a background pipeline. Lurno parses the file into text with structure preserved, writes a short summary, splits the content into searchable chunks, and indexes them. When that finishes the document is marked Ready and two things unlock:

Grounded generation. The document can be attached to a program as part of its knowledge base. Every AI generation inside that program — lesson drafts, module outlines, assessment items — is then written from your material rather than general knowledge.
Ask this document. You can ask questions about the document and get an answer grounded strictly in it, with citations pointing back at the part of the document the answer came from.

Nothing else in the app changes. The library is an authoring surface — people without permission to view source documents don’t see it in their navigation at all — and uploading a document does not publish anything.

Before you start

You need permission to view source documents to see Sources in the navigation. This is granted to every role that can edit programs, so org admins and instructors normally have it. See Roles & permissions.
You need permission to manage source documents to upload or delete.
Have the file ready as a PDF, Word document (.docx), PowerPoint deck (.pptx), or an image (PNG, JPEG, WebP).
If you administer more than one organization, pick the right one in the selector at the top of the page. A document belongs to the organization it was uploaded into, and it can only be attached to programs in that same organization.

Upload a document

Open Sources from the navigation, then choose Upload document.
Drop a file, or browse for it. The dialog derives the type from the file itself; anything outside PDF / Word / PowerPoint / image is rejected before it uploads.
Give it a Title. Lurno pre-fills a tidied-up version of the filename — change it to whatever your team will recognise, since this is the name that appears in the program picker later.
For PDFs and images only, you can tick Extract tables & figures (higher quality). This uses layout-aware parsing that pulls out figures, tables, and headings with better structure. It is slower and costs a little more, so leave it off for plain text documents and turn it on for anything diagram- or table-heavy. Word and PowerPoint files are parsed with their figures either way, so the option doesn’t appear for them.
Choose Upload & process. The dialog closes as soon as the file is handed off — processing continues in the background, so you can leave the page.

Watch it process

Each document shows a status pill that moves through the pipeline:

Status	What’s happening
Uploaded	The file is stored and queued.
Parsing	Lurno is reading the file and recovering its structure.
Parsed	Text and structure are extracted; a summary is being written.
Indexing	Content is being split into chunks and indexed for search.
Ready	Finished. The document can be attached to programs and asked questions.
Failed	Something went wrong. The reason is shown on the card and on the document page.

The library refreshes itself every few seconds while anything is mid-pipeline and tells you so above the grid, so you don’t need to reload. A short Word document finishes faster than a long scanned PDF.

Long PDFs are handled asynchronously. A large PDF or image is routed to a batch processing job instead of being read inline, and a background job checks on it every minute and finishes the ingest when the result comes back. During that time the document simply sits at Parsing. If a batch hasn’t completed after 30 minutes it is marked Failed with a timeout message so it never lingers forever.

Read and question a document

Open a document from the library to get its page. The header shows the file type, the page count, the number of searchable chunks once it’s indexed, and its status.

The left panel shows the summary Lurno wrote from the parsed content.
The right panel is Ask this document. It stays disabled with an explanatory note until the document reaches Ready.

Type a question and you get an answer built only from that document’s content, plus citation chips underneath. A chip shows a page number when the excerpt came from a numbered page, and the section heading when it didn’t. Hover a chip to see the start of the quoted passage. If the answer isn’t in the document, Lurno says it couldn’t find it rather than filling the gap from general knowledge.

The conversation is not saved — it’s a working session, and it starts fresh next time you open the document.

Delete removes the document and everything indexed from it, including its attachment to any program. That can’t be undone; re-upload the file if you need it back.

Preview the original, or the text Lurno read

Anywhere a document is previewed — in Sources, and on a document attached to an assistant conversation — you get two views:

Document — the original file. PDFs and images render inline; Word and PowerPoint files are painted from the file itself, so a .pptx shows as slides rather than as a download.
Extracted text — what Lurno actually parsed out and grounds answers on. This is the one to check when an answer cites something you don’t recognise, or when a document reached Ready but answers from it look thin.

Open original downloads the file. The preview opens on Document every time, because that’s the version you recognise.

Who can see a document

A document uploaded to Sources is shared with your organization — anyone holding permission to read documents can open it, ask it questions, and attach it to a program.

There is one exception, and it matters. A document uploaded from inside an assistant conversation is private to you. It’s labelled private to you, and nobody else in the organization can read it, read its text, search its contents, or see the figures pulled out of it — not even someone who holds permission to manage the whole document library. That privacy covers the document’s parsed versions, its blocks, its searchable chunks, and any images re-hosted from it, because a document whose text is protected but whose figures are not is not protected.

Attaching an already-shared document to a conversation doesn’t make it private, and doesn’t change who else can see it. Only where it was uploaded decides that.

The assistant can only search the documents attached to the conversation you’re in — it can’t be talked into reaching a document you haven’t attached.

What Lurno refuses to ingest

Uploads are checked against the actual bytes of the file, not the extension or the content type the browser declared, because both of those can be set by whoever made the file.

A file whose real type disagrees with what it claims to be is refused, and the message names what it actually looks like.
Executables are refused outright whatever they claim to be — Windows, Linux and macOS binaries, Java class files, and anything that starts with a script shebang.

This is a cheap structural check, not antivirus. Treat it as a floor.

Attach documents to a program

A document only affects AI generation once it’s attached to a program.

Open the program and choose Settings in its side rail.
In the Source documents card, choose Add sources.
Pick the documents to attach. Only documents that have reached Ready appear in this picker — a document still parsing or indexing cannot be attached.
The card lists what’s attached with each document’s status and page count. Remove one with the × on its row.

You can attach up to 5 source documents per program. Attachment order is preserved — the documents are handed to the AI numbered in the order you attached them.

Editing this card needs permission to edit the program, not just permission to manage documents — so someone can curate the library without being able to change what a given program is grounded in.

Once sources are attached, every AI generation inside that program inherits them by default. In a generation dialog you’ll see a line like “Grounded in this program’s 3 source documents” with a Change control, which lets you override the set for that one generation or opt out of grounding entirely. Attaching a document to one program never affects another. See AI tools for how grounding behaves during generation.

Reference

Term	What it is
Source document	A file uploaded to the library for Lurno to read, parse, and index.
Sources	The navigation label for the document library.
Ready / indexed	The finished state — the only state a document can be attached or asked questions in.
Searchable chunks	The indexed pieces a document is split into; shown as a count on the document page.
Extract tables & figures	Opt-in layout-aware parsing for PDFs and images. Slower, better structure.
Knowledge base	The set of source documents attached to a program in its Settings.
Citation	A pointer back to the page or section an answer came from.

Hard limits worth knowing: 300 pages per document, 5 attached sources per program, and a file-size cap that depends on the type — by default 250 MB for PDFs and Word documents and 50 MB for PowerPoint decks. Those size caps are stored per organization, so yours may differ from the defaults; there is no self-serve screen for changing them, so ask Lurno if you need a different limit.

Troubleshooting

A document is stuck on Parsing. Large PDFs go through asynchronous batch processing, which is checked once a minute — several minutes at Parsing is normal for a long scanned file. If it’s still there after 30 minutes it will flip to Failed with a timeout message on its own; delete it and upload again.

A document says Failed with a generic “processing failed” message. That message is deliberately non-specific, because the underlying detail can include infrastructure diagnostics. Delete the document and retry the upload. If the same file fails repeatedly — especially a PDF or image, which depend on Lurno’s document-reading service — contact Lurno; Word and PowerPoint files use a different path and will keep working meanwhile.

A document doesn’t appear in the program’s Source documents picker. The picker only lists documents that reached Ready. Open Sources and check the status: anything Uploaded, Parsing, Parsed, Indexing, or Failed is excluded, because there is nothing indexed to ground a generation in yet.

The upload was rejected for size even though the dialog said 250 MB. The dialog shows a single generous figure, but the real limit is applied per file type on the server — PowerPoint decks are capped much lower than PDFs by default, and your organization may have different caps stored. The error message names the actual limit for that type.

A document failed with a page-count message. Documents are capped at 300 pages, and the count is only known once the file has been read — so this shows up as a Failed document rather than a rejected upload. Split a longer book into parts and upload each as its own source; you can attach several to the same program, up to the limit of 5.

Ask this document is greyed out, or asking returns an AI error. The panel is disabled until the document is Ready. If it’s Ready and questions still fail, check that AI features are switched on for your organization — the answer is generated by AI. See AI tools.

A colleague can’t see a document I uploaded. You uploaded it from inside an assistant conversation, which makes it private to you. Upload it to Sources instead if the organization should have it.

An upload was refused as the wrong type. The file’s actual bytes disagree with its name or extension. Re-export it properly from the application that made it.

Documents fail during indexing across the board. Indexing needs AI switched on for the organization too, not just Q&A. If AI was turned off, documents will not reach Ready. See AI tools.