# text-extract

F∆I module — extract plain text from PDF and DOCX documents.

## Capability

| Field | Value |
|-------|-------|
| Capability | `text.extract@0.1.0` |
| Inputs | `document: bytes` (PDF or DOCX) |
| Outputs | `extracted: json` (ExtractedDocument) |
| Permissions | _none_ — pure in-WASM, no filesystem, no network |
| Status (in store index) | `alpha` |

## Output schema

```json
{
  "engine": "pdf-extract",
  "engine_version": "0.7",
  "pages": [
    { "number": 1, "text": "..." }
  ]
}
```

`confidence` and `bbox` per page are reserved for a future
service-mode variant (host-services-backed pdfium/mupdf engine).
v0.1.0 does not populate them; the schema accepts them as
optional so that v0.2.0 can extend without an interface break.

## Build

```bash
cargo build --release --target wasm32-wasip2
# Artifact: target/wasm32-wasip2/release/text_extract.wasm
```

The build depends on a sibling checkout of `fai/module-sdk`:

```
~/fai/
├── module_sdk/                 # cloned from fai/module-sdk
└── fai_modules/
    └── text-extract/           # this repo
```

This is a temporary path-based dependency. Once `chain-module-sdk`
is published to a registry or a stable git tag, this module's
`Cargo.toml` will switch over. The Forgejo CI workflow checks out
the SDK repo into a sibling directory before building.

## Test

```bash
cargo test                       # 9 tests: 6 unit + 3 integration
cargo test --release             # same, optimized
```

The DOCX path is covered end-to-end by an integration test that
builds a minimal valid DOCX in memory and feeds it through
`invoke`. The PDF path is covered by unit tests on the page
splitter; full PDF coverage will land once the platform's hub
loads this wasm component as part of its store-install
integration test.

## Limitations of v0.1.0

- **Layout-naive PDF parsing.** Multi-column layouts and tables
  may produce reordered text. Single-column documents work
  reliably.
- **DOCX text only.** Tables, footnotes, headers/footers and
  embedded objects are not yet extracted. Plain paragraph text
  with run-level concatenation is the v0.1.0 surface.
- **No OCR.** Image-only PDFs return empty text per page.
  An OCR variant will arrive as a separate module
  (`text.ocr@x.y.z`) once the host-services infrastructure
  for tesseract is in place.

## License

Apache-2.0. See `LICENSE`.

Author: Dr. Stefan Flemming, Flemming.AI <chain@flemming.ai>
Repository: https://git.flemming.ws/chain-modules/text-extract