rgfind is search that can account for missed results. A plain rg search can skip gitignored, hidden, and binary files while it still exits successfully. The rf command recovers these matches and identifies the filter that hid each one.

The problem: a search that fails silently

By default, ripgrep skips files that .gitignore excludes, hidden files, and binary files. That is a good default at a terminal. If a search comes up short, you add -uu and look again.

A program cannot look again. When an agent or a script runs rg, a search that missed most of the matches still exits 0. The caller thinks the search succeeded. The problem gets worse when rg is chained with fd, Git history, or structural search. No stage of the pipe can say what the whole pipeline did not search.

What `rf` shows you

Take a repo with the word timeout in three places: a tracked file, a hidden file, and a gitignored file. Plain ripgrep finds one. rf finds all three and names the filter that hid each:

$ rf content timeout .
content 'timeout' in . [regex]: 3 file(s), 1 by default, 2 hidden by filters
  default      config.py
  hidden       .env.local
  vcs_ignore   cache/build.py
  ! IGNORE_VCS: 1 match(es) hidden by default; add -u (ignore .gitignore/.ignore rules)
  ! HIDDEN_SKIPPED: 1 match(es) hidden by default; add -uu (also search hidden/dotfiles)
  $ 'rg' '-u' '-e' 'timeout' '--' '.'
  $ 'rg' '-uu' '-e' 'timeout' '--' '.'

Each miss has a warning code and the rg command that finds it.

Commands

rf content searches file contents. It removes ripgrep’s default filters one layer at a time — gitignore, hidden, binary, and case — and shows which layer hid each file.
rf find runs staged discovery across file names, file contents, Git history, and structural search with ast-grep. It can recover matches that were deleted from the working tree but remain in Git history.
rf why takes one pattern and one file. It tells you whether the file matches and which filter hid it from a normal search.
rf doctor reports which ignore rules apply to a path.
rf capabilities prints the command contract as JSON.

Built for programs that act on search results

Use rf when a program acts on a search result without a person watching: an agent, a script, or a test. For interactive work, plain rg and fd are fine.

Every rf command returns one envelope with the same seven keys. An empty result is a success with no rows, not an error. Exit codes are typed: 0 for success, 1 for bad input, 3 for an environment error, 5 when the files changed during paging, and 6 for an internal error. Results page at 100 rows with a cursor. One command’s file list can feed the next command.

How it works

rf links ripgrep’s and fd’s own Rust crates, ignore and grep, instead of calling the rg and fd binaries. It reads each file’s filter status while it walks the tree, so every match keeps a record of where it came from. Git and ast-grep run as subprocesses. If either tool is absent, rf skips that stage and reports it.

Start with `rf`

rf is in early development. Expect breaking changes before 1.0. It is licensed under Apache 2.0.

cargo install rf
rf content timeout .

See the rgfind documentation for the quickstart, the JSON contract, and the agent guide.

Open rgfind.dev

rgfind