sigmap

Why we included this project

SigMap targets a problem that shows up quickly once an AI assistant starts working on a real codebase: the assistant burns tokens re-reading files and occasionally invents symbols or paths that don't exist. The tool builds a deterministic, byte-stable signature map of the repository, with no LLM calls, no embeddings, and no vector database, so the same repo always yields the same map. Agents can consume that map as compact context, and a verify command checks AI output against real line anchors, catching fabricated files, imports, and symbols before they cause damage. That makes it a practical token-reduction layer for Claude, Copilot, Cursor, or local models like Ollama, cutting context size by roughly 97%, and it can also run as a CI gate. It works offline with zero dependencies across 33 languages, so it slots into existing workflows without new infrastructure. The token savings are a nice side effect, but the real value is the verifiable grounding it gives you before you trust an AI answer.