manojmallick/sigmap

What it solves

SigMap provides a deterministic and verifiable grounding layer for AI coding agents. It solves the problem of "hallucinations" and non-reproducible context in AI-assisted development by ensuring that the files, symbols, and line numbers an AI suggests are actually present in the codebase, rather than relying on probabilistic guesses or unstable embeddings.

How it works

Instead of using vector databases or LLM calls, SigMap builds a byte-stable, auditable map of signatures and evidence from your codebase. It uses TF-IDF for ranking files against queries and extracts signatures (via AST or anchored regex) across 32 languages. This map can be consumed as an "Evidence Pack" (JSON/Markdown) or via an MCP server that provides 21 on-demand tools for agents to verify symbols and blast radius.

Who it’s for

Developers using AI coding assistants (such as Claude Code, Cursor, Aider, or local LLMs via Ollama) who need a reliable way to ground AI answers in real code and integrate verification gates into their CI/CD pipelines.

Highlights

• Deterministic Grounding: Same codebase always produces the same map, allowing for caching and CI gating.
• sigmap verify: A flagship tool that flags fabricated files, imports, and symbols in AI-generated answers.
• Zero Dependencies: Runs fully offline with no vector DB, no embeddings, and no API costs.
• Broad Integration: Supports a wide range of AI assistants through dedicated adapters and an MCP server.
• Multilingual Support: Extracts signatures from 32 different programming languages.