grep for documents, image, code
Developer who sees beyond the hype
Providing context to LLMs is more important than ever. It is not just what you provide (just text), but also how which makes Agentic coding either a wasteful and costly run vs cheap and fast.
Found a tool that solves the retrieval process: mgrep
It is an enhanced `grep`. Whilst, grep does text-based searching, it was built around 50 years ago in the UNIX era. Philosophy of being simple is commendable, but in this age of multi-model; we need something better
mgrep is multimodal with semantic search. It can search in human language across PDFS, text, audio, video and codebases.
Now for numbers:
- ๐ง๐ผ๐ธ๐ฒ๐ป ๐๐ณ๐ณ๐ถ๐ฐ๐ถ๐ฒ๐ป๐ฐ๐: Reduces token usage by 53% when utilized by coding agents like Claude Code.
- ๐๐ฎ๐๐ฒ๐ป๐ฐ๐: Improves response speeds by 48%
- ๐ฅ๐ฒ๐๐ฝ๐ผ๐ป๐๐ฒ ๐ค๐๐ฎ๐น๐ถ๐๐: Yields 3.2x better responses based on an LLM-as-a-judge evaluation (GPT-4 / Gemini) on 50 coding tasks
mgrep is a CLI which can be used simply with ๐ฎ๐จ๐ณ๐ฆ๐ฑ "๐ฉ๐ฐ๐ธ ๐ด๐ต๐ณ๐ฆ๐ข๐ฎ๐ช๐ฏ๐จ ๐ช๐ด ๐ช๐ฎ๐ฑ๐ญ๐ฆ๐ฎ๐ฆ๐ฏ๐ต๐ฆ๐ฅ". It can be leveraged with tools like Claude Code, Codex, Opencode as simple as ๐ฎ๐จ๐ณ๐ฆ๐ฑ ๐ช๐ฏ๐ด๐ต๐ข๐ญ๐ญ-๐ค๐ญ๐ข๐ถ๐ฅ๐ฆ-๐ค๐ฐ๐ฅ๐ฆ.
I've been using it in Codex and it has made me use the web-based LLM sites even less.
Check it out here: https://www.mgrep.dev/ย
#LLM #AgenticCoding #SofwareEngineering #RAG
