readseek

readseek reads source files and PDFs for scripts, editors, and coding agents. It returns compact JSON with stable line/hash anchors, symbol maps, parse diagnostics, AST matches, references, and rename plans.

The screencast above was recorded with a very modest AMD Ryzen 5 PRO 4650 laptop CPU, and demonstrates the tailored inference engine for Qwen3-VL-2B.

Installation

Install the npm wrapper with a prebuilt binary:

npm install -g @jarkkojs/readseek

Or install the native program from crates.io:

cargo install readseek

Prebuilt binaries are available for macOS ARM64; Linux ARM64 and x64; and Windows x64. The Linux binaries are statically linked with musl.

To build and install from source:

make install

Plugins

Pi extension

The bundled pi-readseek extension adds ReadSeek's anchored file and structural-code tools to Pi:

pi install npm:pi-readseek

OpenCode plugin

The opencode-readseek plugin adds the same tools to OpenCode. Add it to opencode.json:

{
  "plugin": ["opencode-readseek"]
}

Vim plugin

The bundled readseek.vim plugin provides anchored reads, structural navigation, parse diagnostics, search, references, and renames. It requires Vim9 with +job, +channel, +timers, +popupwin, and +textprop; Neovim is not supported.

Install it with a Vim plugin manager, such as Plug 'jarkkojs/readseek', then run :ReadSeekInstall to install the matching prebuilt binary. Downloads are explicit by default.

Common commands

readseek detect src/main.rs
readseek read src/main.rs:10 --end 20
readseek read report.pdf --page 3
readseek view report.pdf --page 3
readseek map src/main.rs
readseek check src/main.rs
readseek symbol src/main.rs:run --name
readseek identify src/main.rs:42 --column 8
readseek def src run --language rust --format plain
readseek refs src main --language rust --format plain
readseek search src 'fn $NAME() { $$$BODY }' --language rust
readseek rename src/main.rs --line 42 --column 8 --to renamed

Global options must precede the command. For example, write JSON to a file with:

readseek --output result.json detect src/main.rs

Prefix a target with stdin: to analyze an unsaved buffer while retaining a path for language detection and cursor addressing. This works with detect, read, map, check, symbol, and identify:

printf '%s\n' 'fn main() {}' | readseek identify stdin:scratch.rs:1 --column 4

Images and PDFs

detect reports image metadata and PDF page counts. read returns bounded base64 images by default; select local analysis with --vision-mode:

readseek read photo.jpg                         # default: bounded base64 image
readseek read photo.jpg --vision-mode caption   # detailed natural-language caption
readseek read photo.jpg --vision-mode objects   # object labels + bounding boxes
readseek read photo.jpg --vision-mode ocr       # extracted text
readseek read photo.jpg --vision-mode all       # caption, objects, and OCR in one pass

Vision analysis uses the low level by default. Start at low, then increase to medium or high only when additional detail is needed:

readseek read photo.jpg --vision-mode caption --vision-level low
readseek read scan.png --vision-mode ocr --vision-level high

Set RUST_LOG=tracing to emit vision cache and inference traces on standard error.

Run cargo bench --features vision-bench --bench vision for data-driven comparisons defined in benches/vision.txt.

Set READSEEK_VISION_THREADS to a positive integer to override Rayon's worker count. More workers may improve throughput at the cost of CPU and memory.

PDF reads return page-tagged Markdown and embedded images. Use --page to select a page; --vision-mode applies to each embedded image. After readseek init, view creates or reuses a structural PDF index that can be narrowed by page, node, kind, or depth. Line/hash suffixes, --end, --limit, and --language do not apply to visual files.

The Qwen3-VL-2B Q4_K_M model and Q8_0 multimodal projector download lazily to the user cache and are checksum-verified. Inference uses ReadSeek's built-in CPU engine and can be slow.

Cache

readseek init [path] creates .readseek/maps/ and .readseek/def-index/. Map-dependent commands update them on demand and find .readseek/ by walking up from the target; --readseek-dir selects one explicitly. PDF indexes and extracted assets live in .readseek/documents/, while image analysis results are cached under .readseek/vision/. Image cache entries are level-specific.

Documentation

For the complete CLI reference:

man ./man/man1/readseek.1

Pass --help to any command for command-specific usage.

Licensing

The Rust program is under LGPL 2.1+ license. The NPM distribution also contains Apache 2.0 licensed components.

Third Party Attribution

The downloaded Qwen model under Apache 2.0 license.

Q4_K and Q6_K block decoding and dot-product implementation are derived from Dwarf Seek 4 under MIT license.