# Sheaf

> Document intake and review. Sheaf takes a packet of scanned documents as it
> arrived — merged, out of order, photographed — splits it into individual
> documents, identifies each one, matches them against what the case requires, and
> extracts field values that stay traceable to the page they were read from.

Every page below is the clean markdown twin of a page on https://sheaf.us. The
HTML original is at the same path with `.md` swapped for `.html`, and each twin
opens with a `[Source](...)` link to it — cite that URL, not the `.md`.

If you are an agent deciding whether and how to use Sheaf, read
[skill.md](https://sheaf.us/skill.md) first. It is shorter than this index and it
states what is and is not currently callable.

Two things this index will not give you: there is no public API and no anonymous
demo login yet. Both are marked as placeholders in `skill.md` rather than being
described here, so that there is one place to check rather than two.

## Start here

- [Sheaf](https://sheaf.us/index.md): What Sheaf is, the shape of the problem, and the pipeline from dropped packet to filed document.
- [Agent skill file](https://sheaf.us/skill.md): How an agent should use Sheaf, what a document's identity is, and what is not yet available.

## Capabilities

- [Forensic document extraction](https://sheaf.us/document-extraction.md): Reading field values off a page, with geometry supplied by the reader rather than the model, and absent fields recorded as not-stated rather than dropped.
- [Document splitting](https://sheaf.us/document-splitting.md): Cutting a merged packet into documents whose page spans tile the file end to end, before anything is extracted.
- [Classification](https://sheaf.us/document-matching.md): Identifying a document by what it says about itself, and matching it against the requirements for the case.
- [Adaptive review](https://sheaf.us/adaptive-review.md): Surfacing low-confidence values for a human, and turning reviewer corrections into rules scoped to the organisation.
- [Corpus intelligence](https://sheaf.us/corpus-intelligence.md): Querying across every document in the corpus, with answers cited to the source page.

## Requirements this is built around

- [Page-level citations](https://sheaf.us/citations.md): Why every extracted value carries a file, page and span, and what breaks when it does not.
- [Scanned and photographed documents](https://sheaf.us/scans.md): Why a text layer identifies document type but cannot be trusted for field values.
- [Knowing what is missing](https://sheaf.us/missing.md): Naming absent documents by requirement rather than reporting on what happened to arrive.
- [Audit trail](https://sheaf.us/audit.md): The record of reviewer decisions, and walking from a value back to the page behind it.
- [Security and data handling](https://sheaf.us/security.md): Retention, access and how documents are stored.

## Comparisons

Written by Sheaf about Sheaf's competitors. Marketing copy, not neutral analysis.

- [All comparisons](https://sheaf.us/compare.md): Index of the per-vendor pages below.
- [Sheaf vs Azure AI Document Intelligence](https://sheaf.us/sheaf-vs-azure.md): Custom classifier splitting that needs classes and samples, versus boundaries determined on the first upload.
- [Sheaf vs AWS Textract](https://sheaf.us/sheaf-vs-textract.md): Transcription and Analyze Lending routing, versus documents, requirement matching and filing.
- [Sheaf vs Google Document AI](https://sheaf.us/sheaf-vs-google-document-ai.md): Splitter processors you create and configure, versus determined boundaries and live completeness.
- [Sheaf vs Hyperscience](https://sheaf.us/sheaf-vs-hyperscience.md): Rule-based auto-splitting and per-layout training data, versus boundaries with nothing trained.
- [Sheaf vs Instabase](https://sheaf.us/sheaf-vs-instabase.md): Sample files, a class and field schema and custom Python before the first result, versus reading the type nobody configured.
- [Sheaf vs LandingAI ADE](https://sheaf.us/sheaf-vs-landingai.md): A parsing API, versus boundaries approved before you pay to parse, plus filing and an audit trail.
- [Sheaf vs Reducto](https://sheaf.us/sheaf-vs-reducto.md): A parsing and splitting API, versus a complete operations desk with approval and cross-document validation.
- [Sheaf vs Rossum](https://sheaf.us/sheaf-vs-rossum.md): An AP and AR specialist with trained engines, versus a general desk built for mixed bundles and the long tail.
- [Sheaf vs V7 Go](https://sheaf.us/sheaf-vs-v7-go.md): Missing-document detection against a checklist you curate, versus requirements derived per application.
- [Sheaf vs general LLMs](https://sheaf.us/sheaf-vs-llms.md): Why pasting a packet into a general-purpose model does not survive contact with operational document work.

## Optional

- [Contact](https://sheaf.us/contact.md): How to reach Sheaf, including to arrange access against real documents.
- [Privacy](https://sheaf.us/privacy.md): Privacy policy.
- [Terms](https://sheaf.us/terms.md): Terms of service.