Show HN: Semble – Code search for agents that uses 98% fewer tokens than grep
98% fewer tokens. That's what Semble delivers for code search in AI agents—and it just went open source.
Why it matters
As Claude Code and other AI agents scale on large codebases, token efficiency in retrieval becomes a competitive moat. Semble's open-source MCP server offers a drop-in solution that cuts token spend dramatically while maintaining retrieval quality, directly addressing a pain point for developers shipping agent-based workflows.
The key facts
8 to know98% token reduction vs grep+read baseline
0.854 NDCG@10 retrieval quality; 99% of 137M transformer performance
~250ms repo indexing on CPU; ~1.5ms per query
~200x faster than transformer-based retrieval
MCP server integration: Claude Code, Cursor, Codex, OpenCode
Zero config: no API keys, no GPU, no external services
Benchmark: 1,250 query/document pairs across 63 repos, 19 languages
Open source release on GitHub + Model2Vec embeddings (potion-code-16M) on Hugging Face
Go to the source
Hacker Newsgithub.com
Publisher excerpt: Hey HN! We (Stephan and Thomas) recently open-sourced Semble. We kept running into the same problem while using Claude Code on large codebases: when the agent can't find something directly, it falls back to grep, reading full files or launching subagents. This uses a lot of tokens, and often still…
