The open-source
context layer for
data agents.

ktx is an open-source context layer for data agents that makes them much more accurate at querying data warehouses. It gives agents the business context, approved metric definitions, joinable columns, warehouse metadata, and query history they do not get from direct warehouse access. It also includes a query engine, so agents can query through ktx instead of writing SQL from scratch.

Yes. ktx is open source under the Apache 2.0 license. There are no usage limits. You run it yourself with your own LLM API keys.

Run npm install -g @kaelio/ktx, then run ktx setup from your project directory. You can also ask Claude Code, Codex, Cursor, or another coding agent to install and configure ktx for you by giving it the prompt from the ktx quickstart. Setup creates a ktx project, connects your warehouse and context sources, builds context, and can install agent integrations. ktx requires Node.js 22 or newer.

A direct warehouse MCP connector usually gives an agent access to tables, columns, and SQL execution. ktx gives the agent a query engine plus reviewable metric definitions, a mapped join graph that resolves chasm and fan traps, a wiki of business context, and the source evidence behind every definition.

ktx gives agents approved context before they write or run SQL. Agents can search the wiki, find semantic-layer entities, compile governed metric queries, and run read-only SQL through configured connections. This reduces repeated schema exploration, incorrect joins, and invented metric logic.

ktx runs locally. Your warehouse data stays in your warehouse, and local state and secrets stay in the git-ignored .ktx directory. Your LLM provider sees whatever context or query content your agent sends to it, the same as any other agent connected to your data.

ktx stores shareable context in plain files: ktx.yaml, semantic-layer YAML, and Markdown wiki pages. Those files can be reviewed in git. Local runtime state, credentials, caches, and secrets live under .ktx and should stay out of git.

ktx includes its own semantic layer, but it is built to be flexible and agent-native: agents can search, read, patch, and validate context through files, CLI, and MCP. Compared with standalone semantic-layer tools, ktx adds business context, source evidence, warehouse metadata, query history, and a reviewable wiki around the semantic model.

No. ktx complements the tools you already use. It can ingest dbt, MetricFlow, LookML, Looker, Metabase, Notion, warehouse metadata, and query history, then expose that context to agents. Your existing transformations, dashboards, and BI workflows stay in place.

No. ktx consumes dbt instead of replacing it. If you use dbt, ktx can ingest your models, metric definitions, and documentation into its context layer so agents can use that governed logic directly. If you don't use dbt, ktx can still build context from your warehouse schemas, BI tools, documentation, and query history.

ktx is the open-source engine. You run it yourself, locally, with your own LLM keys. ktx Cloud is the hosted version. Same engine, plus hosted runtime, multi-user workspace, review and approval workflows, SSO, continuous ingest, observability, and included LLM credits.

ktx works with Claude Code, Claude Desktop, Codex, Cursor, OpenCode, generic .agents clients, and any custom MCP-compatible agent your team builds. ktx exposes an MCP server and a CLI, so any agent that speaks MCP can use it.

Warehouses: PostgreSQL, Snowflake, BigQuery, MySQL, SQL Server, SQLite, and many more. Modeling and BI: dbt, MetricFlow, LookML, Looker, Metabase, and many more. Documentation: Notion and many more.

Yes. ktx is built around reviewable context files. Agents can propose updates to semantic-layer YAML or wiki Markdown, but your team can inspect, validate, and approve those changes through normal code review before they become trusted context.

ktx is open source. Issues, pull requests, and discussions are welcome on the GitHub repository.