ktx

Context layer CLI for data and analytics agents

TLDR

Initialize or resume a local ktx project

$ ktx setup

Check project readiness

$ ktx status

Build context from configured sources

$ ktx ingest

Search semantic-layer metrics

$ ktx sl "revenue"

Search local wiki pages

$ ktx wiki "refund policy"

Start the MCP server for agent clients

$ ktx mcp start

ktx is the command-line interface for the ktx context layer, a local tool that teaches AI agents how to query a data warehouse accurately. It ingests warehouse metadata, dbt or MetricFlow definitions, BI tool exports, and wiki content, then exposes approved metrics, join paths, and business knowledge through CLI and MCP tools.ktx setup creates or resumes a project directory containing ktx.yaml, a semantic-layer/ tree, wiki/ pages, and a git-ignored .ktx/ state directory. ktx ingest rebuilds context from every configured connection. ktx sl and ktx wiki search the semantic layer and wiki respectively. ktx mcp start launches the MCP daemon that agent clients (Claude Code, Codex, Cursor, OpenCode) call at query time.Connections are read-only; ktx never writes to the warehouse. Supported databases include PostgreSQL, Snowflake, BigQuery, ClickHouse, MySQL, SQL Server, and SQLite. LLM backends include Anthropic API, Google Vertex AI, AI Gateway, Claude Code session, and local Codex authentication.Install globally with npm install -g @kaelio/ktx. Project resolution defaults to KTX_PROJECT_DIR, then the nearest ktx.yaml, then the current directory.

PARAMETERS

--project-dir path

Override project directory for scripting.

ktx sl query

Full-text and semantic search over semantic-layer entities.

ktx wiki query

Search local wiki pages.

ktx mcp start

Start the local MCP server; use the path printed by ktx status if needed.

CONFIGURATION

Project layout after setup:

$ my-project/
├── ktx.yaml
├── semantic-layer/<connection-id>/
├── wiki/global/
├── wiki/user/<user-id>/
├── raw-sources/<connection-id>/
└── .ktx/ # local state and secrets (git-ignored)

Commit ktx.yaml, semantic-layer/, and wiki/; keep .ktx/ local. See the official docs for LLM provider and connector configuration.

CAVEATS

Requires a SQL warehouse; ktx sits on top of one rather than replacing it. Remote LLM calls send prompts built locally, not raw warehouse rows. The MCP daemon must be running before opening an agent client when ktx status indicates it is required.

HISTORY

ktx is built and maintained by Kaelio (Y Combinator P25). It combines automatic warehouse introspection, semantic-layer ingestion, and wiki reconciliation into a single agent-facing context layer.

RESOURCES

Source code · Homepage · Documentation

ktx