The new Lat tool creates a structured knowledge graph for codebases using interconnected markdown files, enabling AI agents and developers to maintain context and documentation more efficiently.

Key points

• Lat organizes project documentation into a lat.md/ directory, using wiki-style links to connect markdown files, business logic, and specific source code symbols.
• The system includes a CLI that enforces referential consistency, ensuring that documentation and code do not drift apart as projects evolve.
• Developers can use the lat check command to validate links and ensure test specifications are properly referenced within the codebase.
• The tool supports semantic search via OpenAI or Vercel AI Gateway, allowing agents to retrieve design decisions and constraints without performing manual grep searches.
• Lat requires Node.js 22+ and pnpm, providing hooks for CI tasks, pre-commit checks, and GitHub bot integration to automate knowledge maintenance.

Why it matters: As codebases grow, traditional documentation often becomes outdated or lost, leading to AI hallucinations and developer inefficiency. By embedding a persistent knowledge graph directly into the repository, Lat ensures that both human developers and AI agents maintain a shared, accurate understanding of system architecture and design intent.