# Intake

> Upload company documents, ask questions with line-level citations, and review the facts and entities Intake extracts.

Intake makes a set of business documents searchable and question-answerable. Use it for contracts, filings, spreadsheets, decks, reports, screenshots, and other documents you need to quote or compare. Answers link back to numbered lines in the original document, so you can verify the wording before relying on it.

Intake can also identify people, companies, places, and facts across the upload, then surface contradictions, changes, and missing information for review.

*Interactive Intake walkthrough available on this page.*

## Upload documents

### Choose a collection

Open **Intake**. New uploads go into the selected collection, or into **Unsorted** if you have not chosen one. To separate a set of documents, choose **Upload → Into a new collection…**, name it, and choose the files.

### Add files or a folder

Choose **Upload → Files** or **Upload → A folder**. You can also drop files anywhere on the page. Uploading another batch adds to the documents already present; it does not replace the previous batch.

Accepted uploads include PDF, Word, Excel, PowerPoint, CSV, text, images, and `.zip` archives. Empty files and common operating-system junk files are skipped. HEIC and HEIF photos need to be exported as JPEG or PNG first.

### Wait for Ready

Open **Documents** to track each file. A document moves through **queued**, **parsing**, **indexing**, and **extracting** before it is **ready**. **failed** means it could not be processed; **paused** means processing stopped and the document gives a reason.

Filter by status or collection, search by name, or open a document to read its contents and processing details. An unread page is identified as unread, not presented as an empty page.

## Ask and verify an answer

### Ask across the right collection

Open **Intake → Ask**, enter a question, and select one collection or **All collections**. Keep the scope narrow when you are comparing similar documents, such as different years of the same report.

### Open the citations

Numbered markers such as `[1]` connect claims to source passages. Open one to see the document page with line numbers and the cited lines highlighted. The same document title is used in the answer and in **Documents**, so you can find it again.

### Handle an ungrounded result

If the answer is marked **Not grounded**, treat it as no answer. Intake shows the closest passages so you can inspect them, but those passages do not validate the response.

## Review extracted information

Use the other tabs when you need more than a direct answer:

| Tab | What to do there |
|---|---|
| **Entities** | Explore people, organisations, and other names found in the documents. Select an entity to inspect its mentions and supporting passages. |
| **Review** | Accept or reject uncertain classifications, entity merges, names, facts, and extraction results. Select multiple items to settle up to 100 at a time. |
| **Insights** | Compare contradictory values, changes over time, and gaps. Disagreements without a stated period are listed separately because they may describe different times. |
| **Requests** | Inspect questions received from Slack, their status timeline, and the answer returned to the thread. |
| **Storage** | See space used by original files and their searchable index, review the largest documents, and delete documents when necessary. |

Review items need different decisions:

- **Document type unsure**: accept the suggested type, enter a better one, or reject it to leave the document unclassified.
- **Possibly the same entity**: accept to merge the names or reject to keep them separate.
- **Name not found in the text**: accept to keep the suggested entity anyway, or reject it.
- **Fact whose quote was not found**: reject it. The application refuses to accept a fact whose supporting quote is absent from the document.
- **Extraction could not be read**: accept to try extraction again. The document remains searchable either way.

To turn fact and entity extraction on or off, use **Settings → Intake → Find facts and entities in each document**. Turning it off still lets Intake read, index, and search documents.

## Reading limitations

- PDF text is read page by page. A scanned page without a usable text layer is read with OCR and labelled accordingly. OCR works best on clean print and is less reliable on handwriting and skewed scans.
- Page text is read from top to bottom, so a two-column layout can interleave lines from the columns.
- Tables with ruling lines can retain their columns; borderless tables can arrive as plain lines.
- The **Requests** tab currently covers Slack requests.

## Upload limits

| Limit | Value |
|---|---|
| Single file | 40 MB |
| Folder upload | 5,000 files |
| PDF length | 1,000 pages per file |
| OCR | Up to 60 scanned pages per document; later scanned pages are marked unread |
| `.zip` upload | 40 MB, 500 entries, and 400 MB expanded; each entry must also fit the 40 MB file limit |

Split an over-length PDF or an over-sized archive before uploading it.

## Related

- [Channels](https://docs.fatelabs.ca/channels): Send document questions to Intake from Slack.
- [Atlas](https://docs.fatelabs.ca/atlas): Create a cited guide to a codebase instead of a document collection.