Show HN: Strata – an expressive semantic layer that can say no to your LLM

Strata is a governed semantic layer and BI platform that prioritizes correctness and discoverability: naming is the model, dimensions with the same name are conformed and blendable, measures are typed (five measure types) and grain-safe blending and federated routing prevent invalid joins or aggregations. It supports AI analytics agents that translate natural-language questions into model-respecting queries and will refuse or explain when a requested cross-domain blend would be incorrect. The stack is “open OLAP” - you can plug in any OLAP engine without vendor lock-in - and Strata provides beautiful default dashboards, report building, and cohort/segment visuals atop the semantic layer.

The product runs locally or on your server via a Dockerized Strata Server and a Ruby-based CLI; a project is a git folder of YAML model files and an AGENTS.md that instructs coding agents (Claude Code, Codex, etc.) how to draft models. An agent inspects the warehouse, proposes fact/dimension design and joins (one join per table pair; role-playing tables for multiple keys), and writes YAML only after approval. Workflows center on strata audit and strata deploy. Connectors cover Snowflake, Databricks, Postgres, Redshift, MySQL, Trino, DuckDB and others; credentials live in a gitignored .strata file and AI features are opt-in. The site includes troubleshooting, a TPC-DS sample model, dbt/LookML transpilation, a free tier, and setup help.