wisesean · Oct 20, 2026 · Oct 20, 2026 · Oct 20, 2026 · Oct 20, 2026
diff --git a/Cargo.lock b/Cargo.lock
diff --git a/Cargo.toml b/Cargo.toml
@@ -9,7 +9,7 @@ version = "0.2.4"
 edition = "2024"
 # Edition 2024 needs 1.85; zip raises it to 1.88.
 rust-version = "1.88"
-description = "Convert documents (doc, docx, odt, rtf, epub, pdf, presentations, spreadsheets, csv) to GitHub-Flavored Markdown"
+description = "Convert documents (doc, docx, odt, rtf, epub, chm, pdf, presentations, spreadsheets, csv) to GitHub-Flavored Markdown"
 license = "MIT"
 repository = "https://github.com/firecrawl/anydoc"
 readme = "README.md"
@@ -26,6 +26,7 @@ cfb = "0.14.0"
 csv = "1.4.0"
 flate2 = "1"
 encoding_rs = "0.8.35"
+html5ever = "0.40"
 log = "0.4"
 pdf-inspector = "1.14.2"
 quick-xml = "0.41.0"

diff --git a/README.md b/README.md
@@ -6,7 +6,7 @@
 [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)
 [![skills.sh](https://skills.sh/b/firecrawl/anydoc)](https://skills.sh/firecrawl/anydoc)
 
-Fast Rust library that converts documents (Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, CSV, and PDF) into clean GitHub-Flavored Markdown. Includes bindings for [Node.js](node/README.md), [Python](python/README.md), and the [browser](wasm/README.md) (WebAssembly).
+Fast Rust library that converts documents (Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, CHM, CSV, and PDF) into clean GitHub-Flavored Markdown. Includes bindings for [Node.js](node/README.md), [Python](python/README.md), and the [browser](wasm/README.md) (WebAssembly).
 
 Built by [Firecrawl](https://firecrawl.dev) to turn any office document into LLM-ready Markdown in single-digit milliseconds, with one consistent output no matter which format goes in. It powers [Firecrawl Parse](https://firecrawl.dev/parse), so if you'd rather not run it yourself, the hosted API gives you the same conversion plus our OCR models for the scanned pages anydoc can't read on its own.
 
@@ -166,9 +166,20 @@ Only documents that need OCR leave the machine, and the whole document goes, sin
 | OpenDocument     | `.odt`, `.ods`, `.odp`                                     |
 | Rich Text Format | `.rtf`                                                     |
 | EPUB             | `.epub`                                                    |
+| Compiled HTML Help | `.chm`                                                   |
 | CSV              | `.csv`                                                     |
 | PDF              | `.pdf`                                                     |
 
+CHM topics are read in `.hhc` table-of-contents order, with the hierarchy
+preserved as a nested list. Unlisted HTML topics are appended in stable path
+order; the same fallback applies when the TOC is missing, unreadable, or binary.
+Legacy HTML uses BOM/charset declarations, then the book's Windows language
+encoding (including GBK/Big5). Topic links, HTML bases, and stylesheet encodings
+are resolved locally. LZX supports 4/8-byte reset tables and sequential SpanInfo
+streams; the private decoder adaptation and its MIT notice are under
+`src/formats/chm/lzx/`. Embedded images remain available on `Document.assets`,
+as with the other formats.
+
 ## Benchmark
 
 anydoc is measured against six other converters on 100 real-world documents spanning fourteen formats. Scores run from 0 to 100, higher is better; speed is the median time to convert one document.
@@ -209,7 +220,7 @@ Speed is one warm conversion per document on a Ryzen 9 9950X3D (Windows 11, 64 G
 
 ## Format detection
 
-The format is read from the file content, using the marker its specification designates: the PDF header, the RTF open group, OLE stream names, the ZIP package mimetype and content types. CSV has no such marker, so the extension or an explicit format names it instead.
+The format is read from the file content, using the marker its specification designates: the CHM `ITSF` signature, the PDF header, the RTF open group, OLE stream names, the ZIP package mimetype and content types. CSV has no such marker, so the extension or an explicit format names it instead.
 
 ```rust
 Format::from_bytes(&bytes); // Some(Format::Docx), or None when nothing matches
@@ -261,7 +272,7 @@ document bytes
   ├─► format detection      → content markers, not the extension
   │
   ├─► format parser          → one per format (doc, docx, ppt, pptx, xls,
-  │                            xlsx, odt/ods/odp, rtf, epub, csv)
+  │                            xlsx, odt/ods/odp, rtf, epub, chm, csv)
   │         │
   │         └─► Document     → shared model: blocks, inlines, tables,
   │                            footnotes, assets
@@ -284,6 +295,13 @@ wasm-pack build wasm --release --target web --scope firecrawl && node --test was
 
 A committed fixture corpus under `tests/fixtures/` is snapshot-tested, `tests/robustness.rs` mutation-tests every fixture, and `fuzz/` carries cargo-fuzz targets per format. The speed and quality benchmark lives in [`bench/`](bench/README.md).
 
+The small CHM fixtures can be rebuilt with `python3 tests/fixture-src/chm.py`.
+To run the local ZXR10 manual regression without adding the manual to the repo:
+
+```bash
+ANYDOC_CHM_SAMPLE=/path/to/manual.chm cargo test --test chm local_chm_manual -- --ignored --nocapture
+```
+
 Releases are tagged `v<version>`, which publishes the crate, the npm package, and the PyPI wheels from [`.github/workflows/release.yml`](.github/workflows/release.yml). The version lives in three places, bumped together for a release:
 
 - [`Cargo.toml`](Cargo.toml): the crate

diff --git a/bench/README.md b/bench/README.md
@@ -2,7 +2,8 @@
 
 Benchmarks anydoc against well-known document-to-markdown converters on the
 `samples/` corpus, for both speed and quality (deterministic metrics + LLM judge).
-Office, text, and presentation documents; PDFs are out of scope.
+Office, text, presentation, and CHM documents; PDFs are out of scope. CHM runs
+only against anydoc; the competitor format lists are unchanged.
 
 The corpus is not redistributable and is not in the repo. The harness reads
 whatever documents are in `samples/`.
@@ -48,6 +49,33 @@ python report.py                       # aggregates everything into out/report.m
 All artifacts land in `out/` (gitignored). `convert.py` appends to
 `out/timings.jsonl`; `report.py` keeps only the latest row per (tool, file).
 
+## CHM time and memory
+
+The standard harness accepts `--formats chm`. For per-process peak memory and
+repeatable comparisons with an external CHM, use the dependency-free macOS/Linux
+runner from the repository root:
+
+```
+cargo build --release --example convert
+python3 -m unittest discover -s bench -p test_chm.py
+python3 bench/chm.py /path/to/book.chm --warmup 1 --iters 3
+python3 bench/chm.py /path/to/book.chm --binary /path/to/baseline/convert
+```
+
+`chm.py` prints one JSONL record with every run and their medians. Redirect stdout
+to retain a local result. It never copies the input or saves converted contents.
+Each warmup and measured run starts a fresh converter process; warmups warm OS
+caches without retaining converter state. Run comparisons serially on the same
+machine/build settings.
+`conversion_ms` is the CLI's rounded API time (CHM to Markdown, including the
+document model); `wall_ms` includes process startup, input I/O, conversion, writing
+Markdown to the null device, and process teardown. Assets are not extracted.
+`peak_rss_bytes` is the converter process's peak resident set from `wait4`, not
+total allocated memory or the Python runner's memory: macOS reports bytes, Linux
+reports KiB, normalized here to bytes. Results include binary path, platform, and
+input size; record the tested commit/build separately. No wall-clock thresholds
+are enforced in CI. Use a local private CHM path without adding it to the corpus.
+
 ## Methodology notes
 
 - **Speed**: min of `--iters` warm runs, and the published numbers use `--iters 1`,