Release v0.5.0: Hybrid Search, Re-ranking, and Context Savings

[dbinky]

Release v0.5.0

This release introduces major improvements to search quality and adds context savings metrics.

New Features

Hybrid Search

• Combines vector similarity search with FTS5 keyword search
• Uses Reciprocal Rank Fusion (RRF, k=60) for optimal result merging
• Configurable via hybrid_search settings in config.yaml

Intelligent Re-ranking

• Heuristic signals boost results based on:
  • Name matches (query terms in function/class names)
  • Exact phrase matches
  • File path relevance
  • Recency (recently modified files)
  • Code structure awareness

Enhanced Output

• --verbose flag: Shows detailed match reasons and score breakdown
• --metrics flag: Shows context savings vs grep baseline

New CLI Flags

• --verbose / -v: Detailed match information
• --metrics: Context savings metrics
• --no-hybrid: Disable hybrid search (vector-only)
• --no-rerank: Disable re-ranking stage

Performance

Token savings benchmark on Pommel codebase (136 files):

• Pommel: 4,374 tokens for 10 searches
• Traditional exploration: ~107,000 tokens
• Savings: 95.9% reduction (24x fewer tokens)

Platform Support

• macOS (Intel + Apple Silicon)
• Linux (x64 + ARM64)
• Windows (x64 + ARM64)

CI Status

All builds and tests pass on all platforms (Go 1.21 + 1.22)

---

Generated with Claude Code