GET STARTED · OVERVIEW

kfind, a Korean inflection search engine

printf '길을 걸었다.\n' | kfind --pos verb 걷다

Use the playground to compare inputs and results without installing anything.

First search and documentation

Choose your runtime, then follow the first-search guide. Usage pages explain how to search; reference pages specify exact fields and values.

• Installation and CLI selection compares the native CLI with the lightweight npm CLI.
• Installation and first search walks through a command and its results.
• Recipes provide queries for code exploration and documentation searches.
• Results and diagnostics explain no matches, errors, and search limits.

Usage environments

Use the native CLI for file and repository searches with ignore rules, context output, and streaming input. The lightweight npm CLI searches UTF-8 text in Node.js with different traversal rules and JSON output. Their outputs are not interchangeable.

Coding agents follow the search workflow to inspect the context of broadly collected candidates. Before registering instructions and hooks automatically, read about integration behavior and removal.

Use the JavaScript API or Rust API to search strings in an application. JSON Lines specifies CLI output and its coordinate systems.

Search scope and limits

kfind finds nouns with particles, verb and adjective endings, irregular inflections, and ordered phrases with explicit parts of speech.

걷다 and 걸다 share the inflected form 걸어. A search for 걷다 therefore returns both 길을 걸어 (walk along the road) and 전화를 걸어 (make a phone call) as candidates. kfind does not determine which lemma’s meaning the shared form expresses in context. Read the surrounding text to decide which results have the intended meaning.

The default smart boundary uses structural evidence to reduce unwanted candidates, but it does not disambiguate the meaning of morphologically valid candidates such as 전화를 걸어. The any boundary collects more word-internal candidates. Boundary policies offer a precision–recall choice, not typo correction or guaranteed complete recall.

Unsupported forms and structural limits can cause omissions. Read standard-text and error-containing results separately in the benchmarks, and verify queries on the documents you intend to search.

Search engine structure

kfind analyzes a short query into a search plan, locates fixed strings in the source, and verifies only nearby candidates. It does not morphologically analyze the entire source. The search pipeline explains compilation, scanning, and verification.

Total execution cost includes query analysis, resource initialization, and structural verification. Interpret performance measurements for their inputs, settings, and quality constraints. Fixed-string scanning speed alone cannot predict total morphological search time.