Kaelio

Kaelio develops ktx, an open source context layer for data agents. The problem it addresses is that AI agents querying a warehouse from raw table names often guess and return wrong answers. ktx builds a context layer from a team's data stack and compiles accurate SQL from it, so agents work from approved definitions rather than guesses. It reads the warehouse, BI layer, modeling code, query history and documents, covering sources such as Snowflake, dbt, LookML, Looker and Notion. It writes what it learns as reviewable YAML and Markdown files kept in git: a wiki of business context, definitions, caveats and policies, and a semantic layer of executable metric definitions and join graphs. The team reviews and merges every change like any other diff. At runtime, agents use a command line tool or an MCP server to search the wiki, find approved semantic entities, plan grain and safe joins, and run read only SQL. Named agent integrations include Claude Code, Codex, Cursor, LangChain and the Vercel AI SDK. The vendor reports a first place result on the Spider 2.0 Lite benchmark. Pricing has three tiers. ktx itself is free under the Apache 2.0 license, runs on a team's own infrastructure with its own LLM keys, includes all features and has community support. ktx Cloud is a hosted runtime with a multi user workspace, review and approval workflows, governance and security, and is quoted on request. Kaelio also offers a managed Data Agent delivered for a team.