Summary

SigMap addresses the challenge of providing relevant context to AI coding assistants by intelligently extracting function and class signatures from codebases. It targets developers using LLMs for code, offering significant token reduction, improved AI accuracy, and faster task completion without requiring any infrastructure setup or external dependencies.

How It Works

SigMap operates by first ranking files within a codebase against a user's query using TF-IDF scores. It then generates compact signature files containing only the most relevant code elements, which are fed into the AI's context. Additional commands like validate and judge help confirm the relevance of files and score the AI's answer groundedness, creating a feedback loop for learning and improving future context selection.

Quick Start & Requirements

Installation is flexible: run directly via npx sigmap for zero-install usage, or install globally (npm install -g sigmap) or per-project (npm install --save-dev sigmap). Standalone binaries are available for macOS, Linux, and Windows, eliminating Node.js dependency. No external dependencies are required. Official documentation is available at manojmallick.github.io/sigmap.

Highlighted Details

• Achieves an 80.0% hit@5 rate for finding relevant files (a 5.9x improvement over baseline) and reduces token usage by an average of 96.8%, significantly enhancing AI coding session efficiency.
• Supports 29 programming languages and integrates seamlessly with AI assistants like GitHub Copilot, Claude, Cursor, and OpenAI models, as well as popular IDEs via extensions.
• Presents a novel alternative to embedding-based context retrieval, requiring no vector databases or cloud infrastructure, offering deterministic results and offline capability.
• Features an MCP server providing 9 on-demand tools for specific AI assistants like Claude Code and Cursor, streamlining advanced workflows.

Maintenance & Community

The project is actively maintained, with contributions credited in the changelog. A roadmap is available at https://manojmallick.github.io/sigmap/roadmap.html, and community discussions are encouraged for workflow setup.

Licensing & Compatibility

Released under the permissive MIT license, SigMap is fully compatible with commercial use and integration into closed-source projects.

Limitations & Caveats

The project focuses on signature extraction and TF-IDF ranking; its effectiveness may vary with highly dynamic or unconventional code structures. Specific MCP server features are tailored for certain AI assistants.