Show HN: Semble – Code search for agents that uses 98% fewer tokens than grep
Semble, a new open-source project, was recently featured on Hacker News as a "Show HN," presenting itself as a specialized code search solution designed for AI agents. The project's core innovation lies in its ability to facilitate code exploration for AI models while drastically minimizing token expenditure. According to its developers, Semble achieves a 98% reduction in tokens compared to traditional `grep`-based search methods. This efficiency is critical for AI agents, which operate under strict token budget constraints and often struggle with managing large code contexts when performing tasks like debugging, refactoring, or feature implementation. Large language models (LLMs) used in agents incur costs per token and have finite context windows, making efficient information retrieval paramount. By providing a more token-efficient way for agents to locate and understand relevant code segments, Semble aims to significantly enhance their performance, reduce operational costs associated with large language model API calls, and enable them to tackle more complex programming tasks within limited context windows. The tool's availability on GitHub, as indicated by the provided URL, suggests it is an accessible resource for developers building AI-powered coding assistants or automated development workflows. This approach could lead to agents that are not only faster and cheaper to run but also more capable of deep code understanding by allowing more relevant information to fit within their processing limits, thereby improving the overall effectiveness of AI in software development. The project's public release on May 17, 2026, marks a step towards more practical and scalable AI agent applications in coding environments.
Developers can build more efficient and cost-effective AI agents for code-related tasks by leveraging Semble's token-saving search capabilities.