The developer introduced a local tool for searching the codebase

An independent developer has introduced a local tool for indexing and semantic search of large codebases. The solution is created as an alternative to cloud services and specialized agents, such as RooCode, which often face memory shortages or high costs for creating embeddings when working with large projects.

The tool is built on the basis of a semantic search engine using Ollama and a vector database Qdrant. The main development task is to ensure complete data privacy: the source code, call graph, and vector indices remain on the user's local machine. This eliminates the need to transfer proprietary code to third-party cloud APIs.

The system has been tested on a Ruby on Rails monolith with over 3.5 million lines of code. Unlike existing solutions, which either require significant indexing time or have limited Ruby language support, this tool is optimized for working with enterprise projects of this scale without the risk of out-of-memory (OOM) errors.

The development took six months. The project author published details of the implementation of the architecture, which allows integrating local semantic search into AI agent workflows. The tool is aimed at developers who require deep context of the entire codebase when working with LLM without using external paid services.