Data Management
What the AI can read
An item marked available to the AI is used in Chat & Research. Every available document is considered for each message: sent whole when they fit together, otherwise searched for the passages that best answer you, each cited by document and page. Picking items in the composer narrows it to those. An item marked unavailable is never sent.
Leaving an item unpicked does not keep it out of answers. To keep an old price list, a draft or a private contract out, choose Hide from the AI in its menu.
What it reads
| Pages | Every page in Data Management: the ones you write there, and the ones saved from Chat & Research. |
|---|---|
| Text files | .txt, .md, .csv, .tsv, .json, .html, .xml, .yaml and .log, up to 400,000 characters each. |
| PDF, Word, PowerPoint and Excel | .pdf, .docx, .pptx and .xlsx, up to 60 MB — every page, slide and sheet. |
| Older Office formats | .doc and .xls, up to 4.5 MB. |
| Scanned PDFs | Pages that are pictures of text are read with text recognition (OCR). |
Photos, archives, audio and video are stored and can be downloaded, but have no text to read, so they are not available to the AI.
Not read is not the same as not viewable. Click a video, image, audio file or PDF and it plays or opens right there, with Download for the original.
How a document gets read
Upload a document and it joins a queue. Within about a minute the whole file is read, and its row says Read in full with its page count or length. Reading a PDF, Word, PowerPoint or Excel file is free.
Scanned pages cost a small amount per page, shown in your usage as Reading scanned documents (OCR). Older .doc and .xls files are read by the AI and use credit like any AI request.
Every row says what was read
Under each file's name:
| Read in full | The whole file, with its page count or length. |
|---|---|
| Partly read | How much was read, and a Read the rest button. |
| Not read yet | With a Read it now button. |
| Not readable | The reason — for example, a file larger than the reader accepts. |
| Reading… | In the queue. It updates by itself when it finishes. |
A file the reader could not open says so and offers Try again. The Chat & Research picker marks a partly read file the same way, and the AI is told how much of it it has, so it says so rather than claiming your document does not mention something.
Page numbers
The AI counts pages from the first page of the file. A long report often prints its own page numbers starting after the cover and contents, so page 128 of the file may be printed as 122. An answer may cite either.
Re-read all, and rebuild the search index
At the foot of every folder, under Whole-document reading:
- Re-read all documents queues every PDF, Word, PowerPoint and Excel file in the project that is partly read or not read yet.
- Rebuild search index refreshes search for every document without reading it again.
Each says what it queued, beside the button.
Switching it off
The small dot beside a file in the tree means it is hidden from the AI. Hide a file when you want to keep it in the project without it shaping what the AI writes — an old draft, or a competitor's document. The same switch is on the project's Context screen. What the AI reads for a project
Signed in to VeraGen? Press Guide me on any screen for a walkthrough, or ask Odin.