SigMap — 97% Token Reduction for AI Coding Sessions

Zero dependencies. 21 languages. Built in production. Free forever.

SigMap is an open-source AI context engine that solves one of the most expensive problems in modern software development: your AI coding assistant is reading your entire codebase every single time you ask it a question.

That costs you money. It slows down responses. And it causes hallucinations.

SigMap fixes this at the root — with one command, no configuration, and no dependencies.

The Problem

When you use Claude, GitHub Copilot, Cursor, or ChatGPT for coding, here's what happens behind the scenes:

• Your AI reads every file in your project
• That consumes ~80,000 tokens per session
• Most of that context is irrelevant to your actual question
• The AI gets confused by noise and hallucinates wrong file names, functions that don't exist, and incorrect logic
• You pay for all of it — every single time

At scale, this compounds fast. A team of 10 developers running 20 AI sessions per day is burning millions of tokens weekly on context that doesn't help.

The Solution

npx sigmap

One command. SigMap scans your entire codebase and extracts only the signals that matter — function signatures, class definitions, type declarations, exports, and interfaces — across 21 programming languages.

The result is a compact, structured signature map written automatically to .github/copilot-instructions.md, which is natively picked up by Claude, GitHub Copilot, Cursor, ChatGPT, and any MCP-compatible LLM.

Your codebase (~80,000 tokens)
        │
        ▼
  SigMap scans & extracts
  function signatures
  class definitions
  type declarations
  exports & interfaces
        │
        ▼
Compact context (~2,000 tokens)
        │
        ▼
  AI tools auto-read it ──► Better answers. Lower cost. Zero hallucinations.

No configuration files. No setup. No dependencies. Works in about 10 seconds on any project.

Benchmark Results

Validated across 18 real-world repositories — not toy examples, not synthetic benchmarks.

Full benchmark methodology and raw data: sigmap-benchmark-suite

Key Features

⚡ 97% Token Reduction

Reduces context from ~80,000 tokens to ~2,000 tokens per AI session. Directly translates to lower API costs and faster response times.

🔧 Zero Dependencies

Install and run via npx sigmap — no configuration, no setup, no package bloat. Works on any machine with Node.js.

🌍 21 Programming Languages

Full support for: JavaScript, TypeScript, Python, Java, Go, Rust, C#, C++, C, Ruby, Swift, Kotlin, PHP, Scala, Haskell, Elixir, Dart, Lua, R, Julia, and Shell.

🤖 Built-in MCP Server

Native Model Context Protocol integration. Works directly with Claude, ChatGPT, and any LLM that supports MCP — no middleware, no extra setup.

🔐 SHA-Based Semantic Cache

SigMap caches processed signatures using SHA hashing. Unchanged files are never reprocessed. In production environments, this achieves approximately 70% LLM call reduction on repeated runs.

🛠️ IDE Integration

• VS Code — extension available
• JetBrains — plugin available (IntelliJ, WebStorm, PyCharm, etc.)

📁 Auto-generates AI instruction files

Writes directly to .github/copilot-instructions.md — the file that Claude, Copilot, and Cursor auto-read before every session. No manual steps required.

How to Use It

No install needed:

npx sigmap

Global install:

npm install -g sigmap
sigmap

With MCP (Claude / ChatGPT integration):

{
  "mcpServers": {
    "sigmap": {
      "command": "npx",
      "args": ["-y", "sigmap", "--mcp"]
    }
  }
}

Output example:

✓ Scanned 847 files across 21 language parsers
✓ Extracted 2,341 signatures
✓ Reduced context: 81,420 tokens → 1,987 tokens (97.6% reduction)
✓ Written to .github/copilot-instructions.md
✓ Cached 94% of unchanged files (SHA match)

Done in 8.3s

Production-Proven

SigMap was not built as a side project or hackathon demo. It was developed and validated inside a production banking environment at ING, where:

• LLM pipelines process millions of tokens daily
• The SHA-based cache achieved ~70% LLM call reduction in live systems
• Context accuracy directly impacts compliance and audit workflows
• Zero-tolerance for hallucinations in financial data extraction

The same engine that runs in one of Europe's largest banks is available free, open source, under MIT license.

Who Uses SigMap

Individual developers using Claude, Copilot, Cursor, or ChatGPT who want better answers and lower costs.

Engineering teams running LLM-powered workflows at scale who need to control token spend without sacrificing context quality.

AI engineers building RAG pipelines who need clean, structured codebase context as input.

Solution architects integrating LLMs into existing systems who can't afford hallucinations in production.

Architecture

SigMap operates as a pure extraction layer — it never modifies your code, never sends data to external servers, and never requires API keys.

Project files
    │
    ├── JavaScript/TypeScript parser
    ├── Python parser  
    ├── Java parser
    ├── Go parser
    ├── Rust parser
    ├── ... (21 total)
    │
    ▼
SHA cache layer (skip unchanged files)
    │
    ▼
Signature aggregator
    │
    ├── .github/copilot-instructions.md  ← auto-read by AI tools
    ├── MCP server endpoint              ← Claude / ChatGPT direct
    └── stdout / pipe                   ← CLI / CI integration

Everything runs locally. Your code never leaves your machine.

Traction

• 📦 10,000+ npm downloads
• ⭐ 96 GitHub stars
• 🍴 9 forks
• 💬 Active community discussions
• 🏦 Production-deployed in EU FinTech
• 📅 Actively maintained — v3.3.0, 197 commits

Roadmap

• SigMap Cloud — team dashboard with token analytics
• GitHub Action — run SigMap on every PR automatically
• Enterprise mode — private MCP server for air-gapped environments
• DORA/NIS2 compliance context profiles for regulated industries
• Multi-repo context aggregation

Get Started

# Try it right now — no install needed
npx sigmap

MIT licensed. Free forever. No API key. No account. No telemetry.

Built by Manoj Mallick — Solution Architect, ING Bank Amsterdam. 15+ years EU FinTech. Open source contributor.