Semantic code search by meaning rather than exact text strings.

• Searches code by intent and concept similarity using embeddings, returning ranked results with relevance scores (0.0–1.0)
• Requires GrepAI initialization, an active index created via grepai watch, and a running embedding provider like Ollama
• Supports natural language queries describing behavior or intent; 3–7 word phrases work best, with results limited via --limit flag
• Interprets scores: 0.90+ excellent match, 0.80–0.89 good match, 0.70–0.79 related, below 0.60 weak match