MetaLayer is a plugin that helps your agents query data. It makes their answers more consistent and less prone to error by capturing metrics, joins, and other lore about your data warehouse.
Install yourself
$ git clone git@github.com:ryanjanssen/metalayer.git
$ cd metalayer
$ ./setup.shHave your agent install
Send this link to your agent and ask it to set it up.
https://github.com/ryanjanssen/metalayerBuilt for LLMs, not dashboards
Traditional semantic layers are built for BI platforms. MetaLayer is built natively for AI agents:
- Your data model is stored in plain markdown files,
- turned into a knowledge graph with
[[wikilinks]], - maintained by your AI agent as you pull data, and
- very un-opinionated.
Import your existing data model
Already have a data model in dbt, LookML, or Zenlytic YAML? Just tell the agent where it is:
Import the data model from /path/to/my/models
The agent reads the files, translates them into MetaLayer markdown, verifies against the warehouse, and presents the result for your approval. Starting from scratch? Just ask a data question — the agent will query your warehouse and build the data model from what it learns.
Requirements
- Claude Code or OpenClaw
- Python 3.11+
- Node.js (for QMD)
- A warehouse connection accessible by the agent (Snowflake CLI, MCP server, etc.)