Sheaf
SheafDocs
Sheaf docs
How to use Sheaf: in the app, over the API, and from an agent.
Sheaf takes a packet of scanned documents exactly as it arrived and turns it into a filed, checked case. These pages say how to use it. They describe what exists today, not what is planned.
What Sheaf does, in order
- Receives the packet. Every file lands as an immutable source file, and pre-flight checks flag pages that will read badly.
- Splits it into documents whose page spans tile the packet end to end.
- Identifies each document by what it says about itself.
- Files each document against the case's expected document list, so what is missing is named.
- Extracts field values, each traceable to the page it was read from.
- Records every step and decision in the case's ledger.
The five words
| Word | Meaning |
|---|---|
| Packet | The files as they arrived: merged, out of order, photographed. |
| Case | One application: its files, its documents, its decisions. The API calls it an application. |
| Expected document list | What a complete case must contain. Versioned, and copied into the case when the case starts. |
| Workflow | A saved recipe of steps and models. Attached at the start of a run, it lets the run go unattended. |
| Ledger | The append-only record of who did what on a case, and under whose authority. |
Where to go
- New here: create an account, then run your first packet.
- Connecting your own software: start at the API overview. The quickstart is three commands.
- Building or instructing an agent: using Sheaf from an agent.