getnao/nao

Supports data warehouse-agnostic agents through a filesystem-based context system that ingests data, metadata, documentation, and Model Context Protocols (MCPs), while using LLMs as the reasoning engine. Built with Fastify, Drizzle ORM, and tRPC on the backend, it includes built-in evaluation tooling to unit test agent performance and track reliability metrics before production deployment. The framework self-hosts on any infrastructure and integrates with databases, Git repositories, and Slack workflows through configuration files rather than hardcoded connectors.

744 stars. Actively maintained with 130 commits in the last 30 days.