What happened
MinishLab has launched Semble, an open-source code search library engineered to enhance AI agent efficiency by drastically reducing token consumption and latency during code interaction. Semble delivers precise code snippets, utilising approximately 98% fewer tokens than traditional grep+read methods. The library indexes an average repository in ~250 milliseconds and processes queries in ~1.5 milliseconds, achieving ~200x faster indexing and ~10x faster queries than code-specialised transformers, while maintaining 99% of their retrieval quality (NDCG@10 of 0.854). Semble operates entirely on CPU, requiring no API keys, GPUs, or external services, and integrates with agents such as Claude Code, Cursor, Codex, and OpenCode via MCP server or Bash.
Why it matters
This release significantly lowers the operational cost and execution time for AI agents engaged in code tasks. Semble's ability to provide relevant code chunks with substantial token savings and accelerated query speeds directly reduces inference expenses and quickens development cycles for platform engineers and developers. The local, CPU-only solution eliminates reliance on expensive GPU infrastructure and external APIs, offering a self-contained, efficient alternative for agent-driven code analysis. This follows a broader industry trend, exemplified by recent releases like Zerostack's Rust coding agent, towards more efficient and specialised tooling for AI agents.
