Sheaf
Sheaf vs Reducto
Reducto is a capable parsing and splitting API. Sheaf covers the same ground and then does the half of the job that actually holds files up.
The short answer
Reducto turns documents into structure and hands it back. Sheaf turns documents into a decision: filed, checked against every other document in the file, approved or escalated, and recorded. If a person is going to look at the result, Sheaf is the product.
Sheaf vs Reducto at a glance
| Requirement | Sheaf | Reducto |
|---|---|---|
| Split a mixed bundle | Determined boundaries that tile the file end to end | Split, plus an agentic Deep Split mode |
| Setup before first result | None — upload and go | Zero-shot, no training step |
| Order of operations | Preview boundaries, approve scope, then pay to parse | Classify can run before parse |
| Grounding | File, page and span, supplied by the reader | Citations with page and bounding box |
| Missing documents | Named by requirement, recomputed live | Per-document operations |
| Portfolio completeness | One dashboard across every open application | Yours to build |
| Cross-document consistency | Names, dates and amounts checked automatically | Per-document operations |
| Duplicate and superseded versions | Detected and flagged | Yours to build |
| Straight-through processing | Per-requirement thresholds you control | Yours to build |
| Filing and approval | Ships as the product | API endpoints |
| Audit trail | Every outcome walks back to a page and a person | Yours to build |
Where Sheaf wins
- Everything after the split Reducto hands you separated documents with citations. That is the point at which the actual work starts: matching each document to the requirement it satisfies, filing it, routing the exceptions, letting a reviewer approve or withdraw, and recomputing coverage from what is filed right now.
- The completeness question Sheaf names the required document that never arrived — the requirement itself, by name, on a screen someone can act on — and rolls it up across every open application into one live dashboard.
- Documents checked against each other A file is not correct because each document parsed cleanly. Sheaf checks that names, dates and amounts agree across every filed document and raises the disagreement, which is the error class that survives every per-document pipeline.
- Abstention as a recorded answer When a page does not state something, Sheaf records that distinctly from unread and from not applicable. Three states, three different actions, no ambiguous empty field to misread downstream.
- Corrections that compound A corroborated reviewer decision becomes a rule scoped to your organization. Your instance gets measurably better at your documents with no training cycle and no model to retrain.
Where Reducto is strong
- Deep Split An agentic loop aimed at genuinely hard bundles. Serious engineering, pointed at one stage of the pipeline.
- Classify before parse The same instinct behind Sheaf’s preview-then-spend design, and good company to be in.
- A clean API surface Well-designed endpoints, which matter if you are building the application yourself. Sheaf ships an API too, so this is not the deciding factor it once was.
Choose Sheaf if
- People, not services, are the consumers of the output.
- You need to prove a file is complete and internally consistent, not just parse the documents in it.
- Approval, exception routing and audit history are requirements.
Choose Reducto if
- Your only hard problem is the split, and you already have somewhere to put the results.
- You want an endpoint and nothing else.
The bottom line
Reducto is good at the half of the problem it targets. Sheaf covers that half and the half that comes after — which is the half where files actually get stuck.