The Athena Guide · 04

The Library

Your documents, photos, videos and sounds, where Athena searches and what she quotes from

What goes in

In Files → RAG, at the top, there is the upload area: drag your files or a whole folder there (or anywhere on the page), subfolders included, or click to choose them, or choose a folder; next to it you paste the address of a web page. Below it you see what it reads: PDF (scanned too) and Word, EPUB books, spreadsheets, presentations, photos, audio and video; files of other types in a folder are left out, and Athena says so. In the open area, Put in chooses the collection the uploaded documents go into (web pages and those already there too): one you have, a new one, or none, and then Athena proposes. If you are looking at a collection, that one is already chosen: even dragging files onto the page with the area folded they go there, and next to "Drag here" you see it written. With the arrow on the right the area shrinks to one line, to leave more room for the list (and opens again the same way); Athena remembers how you left it. Also from a chat's attachments, with + RAG (documents and images).

From an EPUB book Athena takes the title and author and reads the chapters in order, each with its name, so she tells you which chapter something was found in. A protected EPUB (DRM, like those bought on Kindle or with Adobe) cannot be read: Athena says so, and you need a copy without protection.

You can upload thousands at once: they go three at a time, with a single bar ("Uploading 120 of 2000", and how many GB), and Athena reads them two at a time, in a queue: below the area you see how many she is reading, how many are waiting and roughly how long is left. A file that is already there (the same one, even under another name) is not uploaded again: Athena tells you how many were duplicates.

Sounds and audio recordings go in too (mp3, m4a, wav, flac, ogg, opus, aiff, up to 500 MB): Athena transcribes the speech with its minutes, and every 30 seconds takes the sound's «fingerprint», to find music and noises as well.

And images: photos (iPhone HEIC too), screenshots, photographed documents (jpg, png, webp, gif, bmp, tiff, heic, up to 50 MB). For each one Athena:

• straightens it the way the phone shows it and makes a thumbnail;
• reads when and with what it was taken (not the place);
• has the chat model describe it, if that model sees images;
• reads the text inside it (a sign, a page, a receipt);
• takes its «fingerprint», to find it by how it looks too.

And videos (mp4, mov, m4v, up to 2 GB): Athena takes a frame every 10 seconds for the cover and the description, transcribes the speech with its minutes as for a sound, and every 30 seconds takes the images' «fingerprint», to find the right moment too. A long video takes a few minutes.

How Athena searches

When you ask her something «in your documents», she searches by words and by meaning, reorders the best passages and answers quoting document and page, or the minute for a sound. The links in her answer open the document, play the sound or the video from that point, or show the image.

She finds photos and videos by their description, the text inside and the speech; similar images and similar video clips (found only by their fingerprint) she doesn't see: she points you to them, but can't say what they show. The description and the text read are automatic and can be wrong (accents, in the text).

Search by meaning

The administrator chooses it in Settings → Search, one model at a time:

• Gemma 2 (recommended): text, and images, videos and sounds in the same search;
• Qwen3: text only, slightly more accurate on text; sounds are found only by their speech, images and videos by their description, the text inside and the speech;
• Off: by words only.

When the model changes, Athena rebuilds the vectors in the background: a few seconds per thousand chunks.

Many documents: collections, filters, search

Collections are how you order the Library, and you decide them: a place, a house, a person, a job («Bevagna», «House in Rome», «Thesis on Cicero»). A document is in one collection only: putting it in another moves it. The round + at the end of the collections row creates one. When there are many, the row scrolls sideways and the magnifier searches them by name. With a collection chosen, next to it appear the pencil to rename it and the bin to remove it: its documents stay in the Library, or you choose another collection to move them to. No collection shows those still to sort out; there Ask Athena where to put them makes her look at their cards again with the collections you have, and the proposals appear above the list (even for a single document). What a document is (a bill, a contract, a book) Athena understands by herself and writes it in the card: there is nothing to choose.

Above the list:

• search: as you type it filters by name; with Enter (or in the content) it searches inside the documents the way Athena does, and shows the passage found for each one, closest first. First you see the documents close to the best one; the others with show the less close ones too;
• the kinds (documents, spreadsheets, presentations, photos, audio, video, web pages) and the period when you added them;
• the view: table (sort by name, collection, date or size by tapping the column) or grid, with thumbnails.

With the checkboxes you select several documents at once (the one at the top selects them all, filters included, also on the other pages): then Move to a collection…, remove from the collection, re-read (the icon with the arrow: text, chunks and card from scratch) or delete. The list goes in pages, with the numbers below: you choose how many documents per page, next to the numbers (10, 25, 50 or 100; Athena remembers it); changing filters or search goes back to the first.

Athena puts things in order

Once she has read a document, Athena writes its card: what it is about in one or two sentences, what it is (it becomes the first hashtag: #bill, #contract), the subject that links it to other documents, the document's date (not the day you uploaded it), the people and companies mentioned and a few tags. For the subject she looks first at the collections you already have: if the document fits one of them, she uses it. She does it by herself, one document at a time, when she is not answering; until she has finished, above the list she says how far she has got. When the card instructions improve, Athena redoes the old cards by herself (only the card: she does not reread the files).

In the table each row is a file, with the name, its collections and three icons: download the original file (a web page opens), re-read and remove. A click on the row opens the details: the summary, the hashtags, the collections, people and companies, the date, the source, the text Athena reads and the chunk map (one cell per chunk, lit when it has its vector). Date sorts by the document's date (in italics the upload date, if Athena did not find one), and the name search also looks at summary, tags, people and companies. Athena proposes collections: a new one when at least three documents share a subject (you create it with one tap, with its documents inside), or one you already have, for the documents she sees there that you have not put in any collection yet (for example "2 into Bevagna"). You accept them one by one, all together with accept them all, or dismiss them. Re-read also redoes the card.

Your files in order

Athena also keeps the Library's files in order on disk, in the Libreria folder of her data: one folder per user, inside one per collection (and Senza raccolta for the others), and the files with their title. When you move a document, the file changes folder; when you remove it from the Library, it disappears from there too. It is a folder to read and copy: a file changed there does not change what Athena reads (for that, upload it again). Download, among the tools above the list, gives you what you are looking at in one .zip, with the same folders: the chosen collection, the ones without a collection, or everything. The state of the Library (the lights, what it is doing, the chunk map) is at the bottom of the page, below the documents.

Collections in chat and in projects

In chat you can ask Athena to look at and organise the Library: «what collections do I have?», «which 2025 contracts do I have?», «search the Manuals collection for how to clean the filter», «put the A2A invoices in a Bills collection» (she creates it if it isn't there), «remove the duplicates from Contracts» (they stay in the Library). For lists she uses the cards (type, date, summary, people and companies).

In a project you can choose its collections (in the project window, Library collections): in the project's chats Athena searches only there, so answers come from the right documents; if you ask for another collection, she searches that one too. With none chosen, she searches the whole Library.

Each document

On each tile: how it was read, pages or duration, chunks (the pieces Athena splits the text into) and vectors; text read (or transcription for a sound, description for an image) shows what Athena reads; re-read starts it again. A sound has its player; an image its thumbnail, which opens the large image, and original for the file as it was; a video its cover, which plays it. If something went wrong, the tile says so in orange (for example which minutes of speech are missing).

An image or video uploaded while the chat model doesn't see images (DeepSeek) stays without a description: when you're back on a model that sees them, re-read.

Good to know

• The Library is yours: other users don't see it.
• Transcribed speech, the description and text of images and videos, and saved web pages count as content from outside: after reading them, actions towards the outside ask for confirmation (see Permissions and security).