Semble
Code search for agents that uses 98% fewer tokens than grep
Details
- External ID
- 48169874
- Source
- HN
- Company
- —
- Product
- Semble
- Website domain
- github.com
- Launched
- May 17, 2026
- Cohort
- —
- Upvotes
- 445
- Upvotes percentile
- 0.9903069466882067
- Tags
- —
- Fetched at
- Sept. 7, 2026, 9:26 p.m.
- Updated at
- Sept. 7, 2026, 9:26 p.m.
Description
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 misses the relevant code. There are existing tools for this, but they were either too slow to index on demand, needed API keys, or had poor retrieval quality.Semble is our solution for this. It combines static Model2Vec embeddings (using our latest static model: potion-code-16M) with BM25, fused via RRF and reranked with code-aware signals. Everything runs on CPU since there's no transformers involved. On our benchmark of ~1250 query/document pairs across 63 repos and 19 languages, it uses 98% fewer tokens than grep+read and reaches 99% of the retrieval quality of a 137M-parameter code-trained transformer, while being ~200x faster.Main features:- Token-efficient: 98% fewer tokens than grep+read- Fast: ~250ms to index a typical repo on our benchmark, ~1.5ms per query on CPU (very large repos may take longer)- Accurate: 0.854 NDCG@10, 99% of the best transformer setup we tested- MCP server: drop-in for Claude Code, Cursor, Codex, OpenCode- Zero config: no API keys, no GPU, no external servicesInstall in Claude Code with: claude mcp add semble -s user -- uvx --from "semble[mcp]" sembleOr check our README for other installation instructions, benchmarks, and methodology:Semble: https://github.com/MinishLab/sembleBenchmarks: https://github.com/MinishLab/semble/tree/main/benchmarksModel: https://huggingface.co/minishlab/potion-code-16MLet us know if you have any feedback or questions!
Enrichment
- Theme
- Claude integrations and coding agents
- Vertical
- Horizontal
- Function
- Search & retrieval
- Audience
- Developer
- AI stance
- AI feature
- Project type
- Commercial product
- Normalized one-liner
- code search optimized for ai agents
- Manually corrected
- False
Could you build this?
Yes It is a developer CLI/tool that creates a lightweight AST or symbol index (using tree-sitter or ctags) over local code so LLM agents can query definitions and references with minimal token usage.
Discussion
20 comments analyzed.
Competitors mentioned: context-mode, morph-mcp / WarpGrep, pgr (Entire.io), codemogger, cs
Concerns raised: Difficulty measuring end-to-end agent performance improvements objectively, Claims of extreme optimization metrics (99x tokens, 88x better results) lack transparent methodology and margin of error, Unclear magnitude of token savings across different repository sizes, Output code quality improvements hard to quantify, Risk that tool makes agents worse rather than better despite claims
Feature requests: Show plots of average/median/max token savings per query as function of repo size, Publish comprehensive methodology and measurement details for performance claims, Improve relevance ranking for non-code documentation search
Competitors
Other products that read as similar to this one — 262 launches clear the similarity bar, closest 8 shown.
Attention rank: #4 of 263 (itself plus its competitors, highest first — normalized so YC and Product Hunt are compared fairly).
Launched 192 days after the earliest competitor.
- CodeRLM · hn · 2026-02-11 · 81 upvotes · similarity 0.53
- Graft · hn · 2026-08-14 · 39 upvotes · similarity 0.51
- WarpGrep: A 20x faster subagent to grep for code · yc · 2025-11-28 · 17 upvotes · similarity 0.50
- Maxxwell · hn · 2026-09-09 · 11 upvotes · similarity 0.47
- Edgee Claude Code Compressor V2 · ph · 2026-07-06 · 179 upvotes · similarity 0.47
- Abralo · hn · 2026-07-08 · 37 upvotes · similarity 0.45
- Agentmemory · ph · 2026-05-16 · 322 upvotes · similarity 0.45
- I nerfed our coding agents on purpose · hn · 2026-06-05 · 27 upvotes · similarity 0.44