Ktx provides an executable context layer enabling AI agents (e.g., Claude Code, Codex) to query SQL data warehouses accurately. It solves AI data task limitations by automatically building a semantic layer, ingesting company knowledge, and mapping the data stack. This ensures agents use approved metrics and canonical SQL, offering reliable data insights without manual semantic layer upkeep.

How It Works

Ktx synthesizes diverse knowledge: raw table metadata, approved metrics, and unstructured company knowledge (wikis, Notion). It automatically detects joinable columns, resolves join traps, and builds a unified semantic layer. This layer, combined with full-text search, is exposed via CLI and MCP tools for contextual AI agent querying at execution. This automates context discovery and knowledge integration, surpassing manual maintenance or agents re-exploring data independently.

Quick Start & Requirements

• Install: npm install -g @kaelio/ktx
• Prerequisites: SQL warehouse (PostgreSQL, Snowflake, BigQuery, etc.), LLM API keys (Anthropic, Google Vertex AI, AI Gateway) or Claude Pro/Max subscription, Node.js. Integrations with dbt, Looker, Notion supported.
• Setup: Run ktx setup for project configuration and context building. Agent integration via npx skills add Kaelio/ktx --skill ktx.
• Links: Quickstart, CLI Reference, Agent Setup.

Highlighted Details

• Automated semantic layer and data context construction from warehouse introspection and external knowledge.
• Ingests and synthesizes business knowledge from wikis/Notion, flagging cross-source contradictions.
• Automatically detects joinable columns and resolves chasm/fan traps in warehouse schemas.
• Provides CLI and MCP interfaces for AI agents, enabling execution with combined semantic and full-text search.

Maintenance & Community

• Community support via Slack.
• Bug reporting and feature requests via GitHub Issues.
• Open-source development model.

Licensing & Compatibility

• Apache License, Version 2.0.
• Permissive license suitable for commercial use and closed-source integration.

Limitations & Caveats

• Requires an existing SQL data warehouse; not a standalone platform.
• Dependent on configured LLM providers and associated API keys/subscriptions.
• Collects anonymous usage telemetry by default (opt-out available).