Show HN: Ktx – An Open-Source Executable Context Layer for Data Agents

A self-learning context layer that helps data agents accurately query information from data warehouses. It focuses on solving the metric definition errors, repetitive data exploration, and incorrect numbers returned that typical AI agents run into during data work.

Key features include:

• Enterprise Knowledge Learning: Collects and organizes Wiki content, removes duplicates, and identifies contradictory information to pass on to the agent.
• Data Stack Mapping: Helps agents write optimal queries through table sampling, capturing metadata and usage patterns, and detecting joinable columns.
• Semantic Layer Construction: Combines raw tables with high-level metrics through a Join Graph, automatically resolving complex SQL problems so agents can fetch metrics declaratively.

Users can run Ktx at no additional cost using their own LLM API key or a Claude Pro/Max subscription.