frg-marketplace

---

frg (fastripgrep) builds a sparse n-gram index of your codebase, then searches in 10ms instead of scanning every file. Inspired by Cursor's fast regex search.

$ frg index .                     # one-time: ~20s for 9K files
$ frg "renderUsage"               # 10ms (ripgrep: 53ms)

Benchmarks

Tested on openclaw (9,000 files), case-sensitive, warm cache, output to /dev/null:

Pattern	Type	grep	ripgrep	frg	vs rg
renderUsage	selective	1,553ms	53ms	10ms	5.3x
useState|useEffect	alternation	962ms	54ms	13ms	4.2x
.*	wildcard	—	197ms	74ms	2.7x
TODO|FIXME|HACK|XXX|BUG	multi-alt	—	53ms	21ms	2.5x
import.*from.*react	regex	1,080ms	84ms	50ms	1.7x
x	single byte	—	102ms	59ms	1.7x
function (24K hits)	broad	398ms	75ms	51ms	1.5x
^\s*export\s	anchored	—	72ms	68ms	1.06x

frg wins on all 8 pattern types. 100% match count parity with ripgrep.

How

• Indexed patterns (6/8): sparse n-gram index pre-filters to ~5 candidate files out of 9,000 before regex verification
• ScanAll patterns (2/8): parallel file I/O with rayon, SIMD-accelerated byte search via memchr, zero-copy streaming output (no Match struct allocation for millions of results)
• Index build: ~23s one-time cost for 9K files, pays for itself after a few searches

Install

Homebrew (macOS / Linux)

brew install qhkm/tap/frg

Install script

curl -fsSL https://raw.githubusercontent.com/qhkm/fastripgrep/main/install.sh | sh

Cargo

cargo install fastripgrep

Build from source

git clone https://github.com/qhkm/fastripgrep
cd fastripgrep
cargo build --release
# Binary at target/release/frg

Usage

# Build the index (one-time, re-run after major changes)
frg index [path]

# Search (uses index if available)
frg search <pattern> [path]

# Rebuild index
frg update [path]

# Show index stats
frg status [path]

Examples

# Find a specific function
frg search "handleExport" .

# Regex with alternation
frg search "useState|useEffect" .

# Case-insensitive
frg search -i "config" .

# Smart case (lowercase = insensitive, mixed = sensitive)
frg search -S "config" .     # matches Config, CONFIG, config
frg search -S "Config" .     # matches Config only

# Fixed string (no regex interpretation)
frg search -F '${variable}' .

# Files only
frg search -l "TODO" .

# Count matches per file
frg search -c "function" .

# Context lines
frg search -C 3 "error" .

# Filter by file type
frg search --type ts "import" .

# JSON output
frg search --json "pattern" .

# Brute force (skip index, like ripgrep)
frg search -n "pattern" .

Search Flags

Flag	Description
-i	Case-insensitive
-S / --smart-case	Case-insensitive if pattern is all lowercase
-F / --literal	Treat pattern as fixed string
-l	Files-with-matches only
-c / --count	Show match count per file
-C <n> / --context <n>	Show surrounding lines
-m <n> / --max-count <n>	Limit results per file
-q / --quiet	Suppress output, exit code only
-n / --no-index	Brute-force scan (no index)
--glob <pattern>	Filter files by glob
--type <ext>	Filter by file extension
--json	Output as JSON

Configuration

frg supports a config file with default arguments, like ripgrep.

Location: ~/.frgrc or $FRG_CONFIG_PATH

Format: One argument per line. # comments. Empty lines ignored.

# ~/.frgrc

# Smart case by default
--smart-case

How It Works

The Core Idea

grep and ripgrep scan every file on every search. frg builds an index once and uses it to skip 99% of files:

Approach	Strategy	Time on 9K files
grep	Scan all files, match regex	~1,000ms
ripgrep	Scan all files, SIMD regex	~50-100ms
frg	Index lookup + read ~5 files	~10-70ms

Sparse N-gram Indexing

frg uses sparse n-grams — variable-length byte substrings selected by a deterministic weight function. This is the approach described in Cursor's blog post.