Indexify

Make a document—or a directory full of documents—searchable from your terminal. PDF, Markdown, and Word (.docx). Local text extraction, ranked passage search, and a Windows 95 inspired Ratatui interface: teal desktop, silver beveled windows, navy title bars, white document panes, and yellow search highlights. Collections, passage previews, and keyboard navigation stay close at hand. No account, server, or model is required.

Install

Homebrew

With Homebrew installed, run:

brew install arrxy/tap/indexify
indexify --version

Homebrew automatically adds the arrxy/tap tap and makes the indexify command available on your PATH. The formula builds from source and installs the required Rust build tools automatically.

To update an existing installation:

brew update
brew upgrade arrxy/tap/indexify

From source

With a current Rust toolchain installed:

git clone https://github.com/arrxy/indexify.git
cd indexify
cargo install --path . --locked

Quick start

Replace the example paths with your own document or directory:

# Index a directory recursively and open the interface.
indexify ~/Documents/papers --name papers

# Or index one document.
indexify ~/Documents/report.pdf --name report

# Open an existing collection later.
indexify browse papers

Type a few keywords and press Enter to search. F7 / Ctrl+O switches between OR (any word) and AND (all words). Use F2–F6 to jump between panels, Ctrl+N to add a collection, and Ctrl+Q to exit. The full keyboard reference is below.

The first positional argument is the source path. Collection names are optional and default to the filename or directory name. Names accept letters, numbers, hyphens and underscores. Existing names require an explicit refresh; an import never silently replaces a collection. For a path containing spaces, use quotes and $HOME:

indexify "$HOME/Documents/My Papers" --name papers

Screenshots

Screenshots of the actual Ratatui interface, using the bundled example notes.

Wide terminal: collections, ranked results, source locations, and highlighted passage previews in the Windows 95 inspired theme.

Compact terminal layout

The collection sidebar moves into a picker on smaller terminals. F5 opens Collections, and F6 opens Collection info.

To try the same example notes from a source checkout:

indexify ./examples/documents --name research

Search and manage collections

indexify browse research
indexify search research "inverted index compression"
indexify search research "inverted index" --mode and
indexify search research "compression" --json
indexify list
indexify refresh research
indexify remove research

remove deletes only the collection's generated index and text snapshots. It never deletes source documents. Directory imports are recursive. Hidden directories, symlinks, Word lock files, and unsupported extensions are skipped. Extensions are case-insensitive. Individual extraction failures are reported and excluded; other files continue to import. Use w outside the search field to inspect a collection's report.

For scripts or a terminal without an interactive TTY:

indexify ./documents --name research --no-ui
indexify refresh research --no-ui

Keyboard

Key	Action
F1	Open keyboard help from any panel, including search
F2 / Ctrl+F	Jump to Search
F3 / Ctrl+E	Jump to Results
F4 / Ctrl+P	Jump to Preview
F5 / Ctrl+L	Jump to Collections
F6 / Ctrl+G	Jump to Collection info
/	Focus search (outside text fields)
Enter in search	Run query and focus results
Tab / Shift+Tab	Cycle all five panels, including Collection info
Up/Down or j/k	Select result/collection, or scroll focused preview
Enter on a result	Open the original document in the system viewer
Enter on a collection	Load it
Up/Down + Enter in Collection info	Select and run Refresh, Import report, or Delete index
Ctrl+D	Open Delete index confirmation from any panel
d / Delete outside search	Delete the chosen index; Delete key works in Collections and Collection info
F7 / Ctrl+O	Toggle OR (any word) / AND (all words); rerun the current query
c	Copy the selected passage
Ctrl+N or a	Add a document or directory
r	Rebuild the active collection from its original path
w	Show import report
?	Show help
Esc	Leave search or close a dialog
Ctrl+Q / Ctrl+C	Quit; cancel safely during import

Single-letter shortcuts apply outside text fields. On Linux, copying uses wl-copy or xclip; macOS uses pbcopy, Windows uses clip.exe. Opening uses the system default application. Page/line/paragraph references are displayed in the interface; opening a specific location depends on the viewer and is not currently automated.

Delete confirmation starts on Cancel. Use Tab or the left/right arrows to choose Delete index, then Enter; Esc cancels. Deletion removes the generated index, extracted snapshots and collection metadata. Source PDF, MD and DOCX documents remain in place. In Collections and Collection info, actions apply to the highlighted collection; in the other panels, they apply to the loaded one. Close any other sessions using that collection before deleting it. The app releases its own reader, recovers the current search session if deletion is blocked, and loads the next collection after a successful deletion.

The layout shows a collection sidebar and side-by-side results/preview on wide terminals. Smaller terminals hide the sidebar; narrow terminals stack the panes. Ctrl+L opens a collection selector when the sidebar is hidden; use arrows and Enter to switch, or run indexify browse <name>. Minimum useful terminal size is 50 columns × 16 rows.

What is indexed

Format	Extraction	Source reference
PDF	Selectable text, page by page	Page number
MD	UTF-8 Markdown text, including code blocks	Heading and paragraph-start line
DOCX	Main document paragraphs, runs, and table cells	Heading and paragraph number

Long passages are split into windows of at most 180 words, with 24 words of overlap. Short paragraphs remain separate. Word documents have no stable page numbers without rendering, so paragraph references are used instead.

Scanned/image-only PDFs need OCR before indexing. Image-only pages in mixed PDFs are not searchable. A PDF can also render normally while its embedded text/font mapping is damaged or unsupported. Nonempty extracted text can still be garbled; such a copy needs a repaired text layer or OCR for reliable search. PDF layout extraction is best effort; complex columns, font encodings, and reading order can affect results. DOCX headers, footers, comments, embedded images, and tracked-change semantics are not interpreted. Its main document XML is limited to 64 MiB. Markdown is indexed as source text, without fetching linked content. PDF and DOCX are parsed one document at a time; parsing a very large individual document can still use substantial RAM.

Storage and refreshes

Collections default to ~/.local/share/indexify, or $XDG_DATA_HOME/indexify when that variable is set. Set INDEXIFY_HOME or use the global --home flag to choose another location:

indexify --home ./my-library ./documents --name research --no-ui
indexify --home ./my-library search research "indexing"

Each collection stores an immutable text snapshot, source metadata, and the compressed search index. Previews keep working when originals change or move. Use refresh to reflect edits, additions and deletions in the source directory. Refresh currently rebuilds the collection; it is not incremental or automatic.

Builds stage a new generation, then atomically replace a small CURRENT pointer only after success. A failed or cancelled import preserves the previous collection. Completed refreshes remove previous generations. Active readers hold a shared filesystem lock; close other search sessions before refreshing or removing that collection. The interface releases its own reader before a refresh. Abrupt termination can leave an unpublished temporary directory, which can be removed while the app is closed.

Indexify uses the published Indexly crate for external sorting, parallel construction, compressed postings, block directories, and exact BM25 top-ten retrieval. CPU indexing is enabled without GPU dependencies; the engine source is maintained in the Indexly repository. The dictionary, page table and passage-offset array are resident; full passage text stays on disk. Each query scores all matching candidates, so very common terms can be slower. Search supports any/all word matching; it does not provide phrase, typo-tolerant or semantic search.

Verify and preview

cargo test --workspace --locked
indexify preview research --query "inverted index" --output preview.svg

preview exports the actual Ratatui frame through its TestBackend. A companion .cells.json file records its terminal cells. There is no separate web mockup. Tests exercise real PDF/DOCX extraction, mixed directories, passage references, Unicode, independent snapshots, failed refreshes, locks, cancellation, terminal layouts, and integration with Indexly. The engine correctness suite is maintained and run in the Indexly repository.

License

Licensed under the MIT License.

Copyright (c) 2026 Aritro Roy.