Show HN: Semble – Fast code search for agents with near-transformer accuracy (github.com)

Semble, a new code search library designed specifically for agents, has been introduced, providing exceptionally fast and accurate access to code snippets. This tool supports indexing and searching a complete codebase in under a second, achieving approximately 200 times faster indexing and 10 times quicker queries compared to a code-specialized transformer, all while maintaining 99% of its retrieval quality. Importantly, Semble operates solely on CPUs without requiring API keys, GPUs, or external services, making it easily deployable. It can be set up as an MCP (Model-Centric Programming) server, enabling agents like Claude Code and Codex to interact seamlessly with any repository that can be cloned and indexed on demand. The significance of Semble lies in its capability to optimize code search for AI-driven development tools, greatly reducing latency and enhancing the efficiency of programming tasks. It uses a combination of static Model2Vec embeddings and a lexical retriever (BM25) to achieve its performance, integrating innovative ranking signals to boost relevant results more intelligently. The benchmarks show Semble almost matching the performance of significantly larger models while being lightweight and fast, which could transform how developers and AI models interact with large codebases. With installation as simple as a pip command, Semble is poised to become an essential resource for the AI/ML community focused on coding and development efficiency.