Semble is a code search library built for AI agents, providing fast and accurate retrieval of relevant code snippets from local or remote repositories, running efficiently on CPU.

How fast and accurate is Semble?

Semble indexes repositories in ~250 ms and answers queries in ~1.5 ms. It achieves 0.854 NDCG@10, matching the quality of code-specialized transformer models but with significantly higher speed on CPU.

Does Semble require a GPU or external services?

No, Semble requires zero setup, runs entirely on CPU, and does not need a GPU, API keys, or any external services, making it cost-effective and easy to deploy.

How do AI agents integrate with Semble?

Semble functions as an MCP (Model Context Protocol) server, enabling seamless integration with MCP-compatible AI agents like Claude Code, Cursor, Codex, and OpenCode for direct codebase access and search.

What types of codebases can Semble index?

Semble is highly versatile, supporting the indexing of both local directory paths and remote Git URLs, allowing agents to search various project sources on demand.

Semble

Name: Semble
Author: MinishLab

•

Provides fast and accurate code search capabilities tailored for AI agents, retrieving relevant code snippets instantly.

Semble is a powerful code search library specifically engineered for AI agents. It delivers precise code snippets rapidly, significantly reducing token usage and agent waiting times. With indexing and searching a complete codebase typically taking less than a second, Semble offers dramatically faster performance than traditional code-specialized transformers while maintaining high retrieval quality. It operates entirely on CPU, requiring no API keys, GPU, or external services. Semble can also function as an MCP server, enabling various agent platforms like Claude Code, Cursor, Codex, and OpenCode to access and index repositories on demand.

主要功能

01Indexes repositories in ~250 ms and answers queries in ~1.5 ms, all on CPU.

02Achieves 0.854 NDCG@10, on par with code-specialized transformer models.

03Supports both local directory paths and remote Git URLs for indexing.

04Functions as an MCP server for seamless integration with AI agents.

05Requires zero setup, running on CPU without external services or GPUs.

064 GitHub stars

使用案例

01Enable AI agents to quickly find relevant code snippets within a codebase.

02Integrate code search functionality into MCP-compatible agent environments like Claude Code or Cursor.

03Develop applications requiring fast, local, and accurate code retrieval for development or analysis.