ColGrep-Agent: The Best Claude Code Plugin in a Long Time

ColGrep-Agent: the free Claude Code plugin that cuts API tokens by up to 46% on large repos by searching code locally instead of burning context on blind guesses. ColGrep-Agent is LightOn's free, open-source two-billion-parameter code-search model that runs locally on your machine and integrates directly into Claude Code as a plugin. Instead of Claude burning context tokens on repeated guesses about naming conventions, ColGrep-Agent searches your repository semantically, reads candidate files, and follows references on your hardware, then hands back exact file paths, line ranges, and the code itself. In LightOn's benchmark on a real CRM codebase, when asked where "custom attributes" are declared (but the code calls them "fields"), grep-only Claude Code spent six turns and roughly 160,000 tokens guessing names; ColGrep-Agent finished in three turns and roughly 80,000 tokens. Across 52 code-search questions, delegating to the agent cut Claude Code's API tokens by 34% in full agent mode, or roughly 20% in the default hybrid (Balanced) mode while matching grep's accuracy at 83.3%. The tool works through three Claude Code hooks that suggest the agent for behavior questions when exact names are unknown. Token savings scale with repository size: about 19% on repos under 50,000 lines, 34% between 150,000 and 600,000 lines, and up to 46% on codebases over one million lines. Setup takes four terminal commands, grab the binary, run `colgrep install-agent` to download the 2.5 GB model, run `colgrep init` to index your codebase once, then `colgrep install-claude-code` to wire the plugin in. On Mac M3 Pro with GPU, queries take around 20 seconds; on CPU-only setups, expect 75, 90 seconds per question. Everything runs locally with Apache 2.0 licensing, your code never leaves your machine, and you can uninstall cleanly in one command. This is for developers working on large codebases who want to cut exploration costs without writing custom integrations. Chapters: 0:00 Intro 1:35 How It Works 4:20 The Plugin 6:24 Setup 8:54 Trade-offs 11:02 Verdict Tools & resources mentioned: - ColGrep-Agent: https://huggingface.co/lightonai/colgrep-agent-2B - Claude Code: https://code.claude.com/ About The Stack The Stack helps you build with AI. Each video takes one tool, model, or workflow and shows how it works in a few focused minutes, with the real benchmarks and real costs. We go deep on Claude Code and Cursor for AI coding, AI agents and MCP servers, the open-source AI tools and GitHub repos most people miss, RAG and vector search, fine-tuning, and running local LLMs on your own machine with Ollama and LM Studio. We compare models like ChatGPT and Claude, test AI automation with Zapier, Make, and n8n, and flag the tools that actually ship. Subscribe for new breakdowns: https://www.youtube.com/@the-stack-ai?sub_confirmation=1 #Claude Code #Code Search #AI Agents