AI coding agents like Claude Code traditionally use grep for code exploration, but this text-based approach forces them into iterative keyword guessing that wastes tokens. Benchmarks show grep-based workflows consume 39% more input tokens and require 26% more reasoning turns than semantic search alternatives. grepai is an open-source tool that enables AI agents to search code by intent rather than keywords, using local vector embeddings via Ollama. It combines semantic and text search with automatic path-based scoring, includes call graph tracing for multiple languages, and integrates natively with Claude Code as a sub-agent. The tool runs entirely locally without cloud dependencies, reducing token consumption while improving search precision.