fastgrep

A drop-in replacement for GNU grep that is parallel by default, builds a lazy trigram index, and is designed from the ground up to be AI-native first.

Why

LLM agents and AI-powered dev tools run grep thousands of times per session. Every millisecond matters at that scale. fastgrep combines SIMD-accelerated literal search, multi-threaded parallelism, and a lazy trigram index to be 2–12x faster than GNU grep across common workloads. No upfront indexing required — the trigram index warms up on the first run and is invalidated automatically when files change (mtime + size check).

Install

With Cargo

cargo install fastgrep

The installed binary is called grep. To use it as your default grep:

# option 1: alias (add to .bashrc / .zshrc)
alias grep="$(cargo bin-dir 2>/dev/null || echo ~/.cargo/bin)/grep"

# option 2: ensure ~/.cargo/bin is before /usr/bin in PATH
export PATH="$HOME/.cargo/bin:$PATH"

From binary releases

Linux (static musl binary, works everywhere including Docker):

curl -fsSL --retry 3 https://github.com/awnion/fastgrep/releases/latest/download/grep-x86_64-unknown-linux-musl.tar.gz | tar xz -C /usr/local/bin

macOS (Apple Silicon):

curl -fsSL --retry 3 https://github.com/awnion/fastgrep/releases/latest/download/grep-aarch64-apple-darwin.tar.gz | tar xz -C /usr/local/bin

All binaries are available on the GitHub releases page.

Usage

# exactly like GNU grep
grep -rn 'TODO' src/

# search only in specific file types
grep -rn 'class User' --include='*.py' .

# case-insensitive search
grep -rni 'error' src/

# fixed string (no regex interpretation)
grep -rFn 'Vec<Box<dyn Error>>' --include='*.rs' .

# list files containing matches
grep -rl 'migration' src/

# count matches per file
grep -rc 'unwrap()' --include='*.rs' .

# second run is faster (trigram index cache hit)
grep -rn 'TODO' src/

# disable trigram index
grep --no-index -rn 'pattern' src/

# control parallelism
grep -j4 -r 'error' .

# pipe from stdin
cat log.txt | grep 'FATAL'

Benchmarks

Criterion benchmarks on a generated corpus (200 files × 5000 lines each).

Scaling with file count:

fastgrep scales ~2x better than GNU grep as file count grows.

Differences from GNU grep

fastgrep intentionally departs from GNU grep behaviour in several places. Every deviation is motivated by the same goal: make recursive search safe and fast for AI agents that can't babysit a hung process.

File size limit (default 100 MiB)

WARNING: 1 file(s) skipped due to size limit:
  - ./data/model.bin (2300.0 MB)

These files may cause grep to hang. To search them anyway, re-run with:
  FASTGREP_NO_LIMIT=1 grep ...
Or adjust the threshold: --max-file-size=<BYTES> (current: 100 MiB)

GNU grep will happily read a 2 GB binary blob line by line, taking minutes or effectively hanging. This is the single biggest pain point for AI agents — the agent is blocked, the user is waiting, and the tool has no way to know it should stop.

fastgrep skips files larger than 100 MiB by default and reports them to stderr. The warning is machine-readable: an agent can parse it, add --exclude, and retry. Override with FASTGREP_NO_LIMIT=1 or --max-file-size=<BYTES>.

Line truncation (default 15000 bytes)

GNU grep outputs lines of any length. In practice, minified JS bundles or serialized data can produce single lines of 10+ MB that flood an agent's context window with noise.

fastgrep truncates lines beyond --max-line-len (default 15000). Set to 0 to disable.

Parallel by default

GNU grep is single-threaded. fastgrep uses all available CPU threads by default (-j0). This changes the output order — files are printed in whichever order workers finish, not in filesystem walk order. For AI agents this doesn't matter (they parse file:line:content tuples), but it means diff <(grep ...) <(fastgrep ...) may differ in line order.

Trigram index

GNU grep has no indexing. fastgrep lazily builds a trigram index on first recursive search and caches it at ~/.cache/fastgrep/trigram/. Subsequent searches skip files that provably can't match. The index is invalidated automatically (mtime + size check). Disable with --no-index.

No -q, -m, -A, -B, -C, -o, -h, -H (yet)

fastgrep implements the flags most commonly used by AI agents. Less common flags are not yet supported. If you hit a missing flag, fastgrep will print an error rather than silently ignoring it.

Build

cargo build --release

The binary is at target/release/grep.

Test

# integration tests (compared against GNU grep)
cargo test

# benchmarks (fastgrep vs GNU grep via criterion)
cargo bench

Environment variables

See ENVIRONMENT.md for the full list of environment variables and CLI flags.

License

Licensed under either of

• Apache License, Version 2.0 (LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0)
• MIT license (LICENSE-MIT or http://opensource.org/licenses/MIT)

at your option.