A CLI tool for searching your Claude Code conversation history.
- Hybrid search combining BM25 full-text + semantic vector search
- Tool call indexing - search Bash commands, file reads, edits, and more
- Filtering - by tool name, project, time range, errors
- Session viewer - view full conversations or quick summaries
- JSON output - machine-readable format for agents
- Incremental indexing - only re-processes changed files
- Fast SQLite-based storage with FTS5 and sqlite-vec
cargo install --path .# Build the search index (model auto-downloads on first run)
glhf index
# Search your history
glhf search "rust error handling"
# Compact output for scanning
glhf search "cargo test" --compact
# View a session
glhf session abc123 --summary# Basic search (hybrid: text + semantic)
glhf search "rust error handling"
# Compact single-line output
glhf search "cargo" --compact
# Filter by tool type
glhf search "git" -t Bash
glhf search "main.rs" -t Read
# Filter by project
glhf search "bug" -p myapp
glhf search "error" -p . # current directory
# Filter by time
glhf search "error" --since 1d
glhf search "refactor" --since 1w
# Only errors
glhf search "failed" --errors
# JSON output
glhf search "error" --json| Flag | Description |
|---|---|
-l, --limit <N> |
Maximum results (default: 10) |
-t, --tool <NAME> |
Filter by tool (Bash, Read, Edit, Grep) |
-p, --project <NAME> |
Filter by project (substring match, . for cwd) |
--since <DURATION> |
Time filter (1h, 2d, 1w, or 2024-12-01) |
--errors |
Only show error results |
--compact |
One line per result |
--json |
JSON output |
glhf session abc123 # Full session
glhf session abc123 --summary # Quick overview
glhf session abc123 --limit 30 # First 30 messages
glhf session abc123 --json # JSON outputglhf recent # Last 10 sessions
glhf recent -l 20 # More sessions
glhf recent -p myproject # Filter by projectShows database size, document counts, top projects, content breakdown, and timeline.
glhf index # Incremental update (fast)
glhf index --full # Full rebuild
glhf index --skip-embeddings # Text search onlyglhf combines FTS5 keyword search with static vector embeddings (potion-base-32M) using convex combination fusion (α=0.75). The two search systems find largely disjoint results — FTS catches exact keywords while vectors catch paraphrases — so combining them improves recall without hurting precision.
See ARCHITECTURE.md for detailed design decisions, model evaluation results, and what we tried and rejected.
- Bruch, Gai, Ingber. "An Analysis of Fusion Functions for Hybrid Retrieval." ACM TOIS 42(1), 2023.
- Tulkens & van Dongen. Model2Vec. 2024.
make check # Format, lint, and test
make build # Build debug binary
make release # Build release binary
make install # Install to ~/.cargo/binMIT
