What problem does it solve? General-purpose AI agents struggle with data tasks because they re-explore the warehouse on every question and invent metric logic, producing inaccurate queries. ktx gives agents an executable context layer combining approved metric definitions, joinable columns, and business knowledge from your data stack. ## Core Features & Use Cases - Semantic Layer Construction: Ingests warehouse metadata, samples tables, detects joinable columns, and builds a join graph that resolves chasm and fan traps across PostgreSQL, Snowflake, BigQuery, ClickHouse, MySQL, SQL Server, and SQLite. - Knowledge Ingestion: Pulls context from dbt, MetricFlow, LookML, Looker, Metabase, and Notion, organizing wiki content, removing duplicates, and flagging contradictions. - MCP Server for Agents: Exposes tools like ktx_search_semantic_layer, ktx_search_wiki, and ktx_query_warehouse so Claude Code, Codex, and Cursor can search metrics and run read-only SQL. - Use Case: An analytics team runs ktx setup and ktx ingest against their Snowflake warehouse, then asks their agent "query revenue by month" and receives SQL grounded in the approved MRR metric definition instead of guessed logic. ## Quick Start Ask your agent to run ktx setup in the project, verify readiness with ktx status, then start the MCP server with ktx mcp start so the agent can search metrics and query the warehouse.
