HN Showcase
Back to browse
GitHub Repository

97% token reduction for AI coding sessions — zero deps, 31 languages, MCP server

510 starsJavaScript

SigMap – 81.1% retrieval hit 5, 96.9% token reduce,zero deps

by manoj079·Apr 30, 2026·14 points·4 comments
Visit ProjectView on HN

AI Analysis

●●SolidBig BrainSolve My Problem

TF-IDF on signatures beats vector embeddings for file retrieval without the infra overhead.

Strengths
  • •Zero dependencies means it runs anywhere via npx without installation headaches.
  • •Benchmarks show 80% hit@5 accuracy, significantly beating the 13% baseline.
  • •Includes a judge loop to iteratively boost weights for frequently solved tasks.
Weaknesses
  • •TF-IDF approach may struggle with semantic queries compared to modern embedding models.
  • •Requires maintaining a separate signature index which adds a build step to workflows.
Target Audience

Developers using AI coding assistants like Cursor or Copilot

Similar To

Sourcegraph Cody · Cursor · Continue

Similar Projects