MIT open source · Zero npm install

97% fewer tokens.
Better AI suggestions.

SigMap extracts function signatures from your codebase and writes a context file that Copilot, Claude, Cursor, and Windsurf read automatically. Every session starts informed.

terminal
$ node gen-context.js --report
SigMap token report
──────────────────────────────
Raw source files ~80,000 tokens
Repomix compressed ~8,000 tokens
SigMap ~4,000 tokens −97%
──────────────────────────────
$ # Copilot now knows your codebase.
97%Token reduction
21Languages
0npm deps
4KTokens / session
How it works

Three steps. No configuration required.

SigMap scans your source files, strips function bodies and comments, and writes a compact signature index your AI tools read automatically.

Step 01

Extract signatures

Scans every source file, extracting only the public API surface — function names, class definitions, types, interfaces. No bodies. No imports. No comments.

Step 02

Write context files

Writes .github/copilot-instructions.md, CLAUDE.md, .cursorrules, and .windsurfrules automatically. Copilot reads these on every request.

Step 03

Stay current automatically

File watcher updates on every save. Post-commit git hook updates on every commit. Context is always current — zero developer action required.

Approach Tokens / session Reduction Cost (Claude Sonnet)
No context (cold session)
~80,000
baseline ~$0.24 / session
Repomix --compress
~8,000
−90% ~$0.024
SigMap
~4,000
−97% ~$0.012
SigMap + MCP (v0.3)
~200–2K
−99% ~$0.001
Language support

21 languages. Zero Tree-sitter.

Pure regex AST extraction — no compiler required, no binary dependencies, works on any machine with Node.js 18+.

🔷

TypeScript

.ts .tsx
🟡

JavaScript

.js .jsx .mjs
🐍

Python

.py .pyw

Java

.java
🎯

Kotlin

.kt .kts
🐹

Go

.go
🦀

Rust

.rs
🔵

C#

.cs
⚙️

C / C++

.cpp .c .h
💎

Ruby

.rb
🐘

PHP

.php
🍎

Swift

.swift
🎯

Dart

.dart
🔴

Scala

.scala .sc
💚

Vue

.vue
🧡

Svelte

.svelte
🌐

HTML

.html .htm
🎨

CSS / SCSS

.css .scss .sass
📋

YAML

.yml .yaml
🐚

Shell

.sh .bash .zsh
🐳

Dockerfile

Dockerfile.*
Companion tool

Works perfectly with Repomix

SigMap and Repomix are complementary. Use both — they solve different parts of the context problem.

⚡ SigMap

Always-on. Runs automatically on every save and commit. Produces a compact signature index — 4K tokens for daily Copilot use.

  • Auto-updates on save + git commit
  • Signatures only — no function bodies
  • 21 languages, zero npm install
  • Writes copilot-instructions.md, CLAUDE.md, .cursorrules

📦 Repomix

On-demand. Run before a deep refactor or architecture session. Packs full file content — 8K compressed tokens for detailed analysis.

  • Manual run before deep sessions
  • Full file content available
  • 170+ languages via Tree-sitter
  • MCP server for agent use
combined workflow
# Always-on context (SigMap)
$ node gen-context.js --setup
# Before a deep refactor session (Repomix)
$ npx repomix --compress
Quick start

Working in under 60 seconds

No package manager. No configuration. One file, one command.

1

Download gen-context.js

One file, no install required. Drop it in your project root or download globally.

curl -O https://raw.githubusercontent.com/manojmallick/sigmap/main/gen-context.js
2

Generate your first context file

Scans your project and writes .github/copilot-instructions.md automatically.

node gen-context.js
3

Set up auto-update

Installs a post-commit git hook and file watcher. Context stays current without thinking about it.

node gen-context.js --setup
4

Open Copilot and ask about your code

Copilot now knows your class names, function signatures, and patterns — without reading every file on each request.

node gen-context.js --report # see your token reduction
strategy quick picks
full best default · complete context
per-module no context loss · lower per-question tokens
hot-cold smallest always-on · MCP recommended
Read full strategy guide →
Open source · MIT

Built for the community

Free forever. Zero telemetry. Runs locally. Adding a language extractor is 80 lines of code — contributions welcome.

★ Star on GitHub Contribute