Code Search¶

The CodeSearch tool searches a pre-built trigram index of the current working directory instead of scanning files on every call. It is provided by the codesearch plugin (../plugins/janito-codesearch-plugin/), not the core package.

The index is stored at ./.janito/codesearch.db and is built automatically when the plugin loads. Load the plugin and use it:

janito --install-plugin https://github.com/joaompinto/janito-codesearch-plugin
# or, if already installed in ~/.janito/plugins (autoloaded):
janito
# or load directly from a checkout:
janito --plugin ../plugins/janito-codesearch-plugin

Automatic index creation

When the codesearch plugin loads (on_start), if there is no ./.janito/codesearch.db in the current working directory, the index is created automatically. No separate build step is needed.

Maintaining the index

In the interactive shell the /codesearch command maintains the index:

/codesearch update — incrementally update (added/deleted/changed files)
/codesearch recreate — rebuild the index from scratch

Conditional loading

The CodeSearch tool is only loaded when ./.janito/codesearch.db exists in the working directory. If you haven't built the index yet, the tool is not advertised to the model — load the codesearch plugin (creates it automatically) or run /codesearch recreate.

Automatic refresh (1 day TTL)

When the tool loads, the index is refreshed in place with an incremental Update() if its last recorded update is missing (an index built before last-update tracking) or older than 1 day. The refresh is best-effort: a failure never prevents the tool from loading — it stays usable with the existing index.

Skips gitignored files

Files and directories matched by the working directory's .gitignore are excluded from the index, so search results never surface gitignored files (build artifacts, vendored dependencies, secrets, ...). .janitoignore is always respected, matching the behaviour of the other file tools, and the .janitoignore file itself is always ignored. When a file becomes gitignored after the index was built, the next Update() drops it from the index.

CodeSearch¶

Searches the trigram index for lines containing the given keywords. Keywords are matched as whole words (foo does not match foobar or foo_bar). The index narrows the candidate files, and every matching line is returned as path:lineno: content — the same format used by the other search tools.

Parameter	Type	Required	Description
`keywords`	`array` of `string`	Yes	Keywords to search for, matched as whole words. Keywords shorter than 3 characters cannot be indexed and are matched by scanning candidate files directly
`match`	`string`	No	`"and"` (every keyword must appear on the same line) or `"or"` (any keyword is sufficient). Defaults to `"and"`

Example:

result = CodeSearch(keywords=["hello", "world"], match="and")
# result["matches"] == ["hello.py:5:     print('hello world')"]

For "and", lines are filtered in keyword order: lines containing the first keyword are found first, then narrowed to those also containing the second, third, ... keyword. Files that are in the index but no longer exist on disk are skipped.

The index uses the trigram algorithm described by Russ Cox in Regular Expression Matching with a Trigram Index (Google Code Search), with SQLite as the storage backend. See ../plugins/janito-codesearch-plugin/ for the implementation.