Canner
Support for Metrics Layer and Semantics Layer Integration (e.g., dbt Semantic Layers and Cube)
Is your feature request related to a problem? Please describe. I'm always frustrated when data pipelines lack a standardized way to define metrics and semantics. This can lead to inconsistencies, difficulty in debugging, and unreliable data outputs. A solution like a metrics layer (e.g., dbt Semantic Layer or Cube) could help ensure consistent data definitions and outputs across the stack.
Describe the solution you'd like I would like WrenAI to support integration with a metrics layer and semantics layer. This could be done by adopting or integrating with tools like dbt Semantic Layers or Cube, which provide a centralized way to define metrics and semantics. By doing so, WrenAI could:
- Standardize metrics definitions across the stack.
- Enable self-serve analytics with reliable, consistent data.
- Simplify maintenance by centralizing business logic and metrics computation.
- Enhance compatibility with modern data tools that leverage metrics layers.
Describe alternatives you've considered Alternatives include manually managing metrics definitions across different parts of the data stack, but this is error-prone and difficult to scale. Another alternative could be to use ad-hoc methods to define semantics and metrics in WrenAI, but this risks divergence and inconsistency over time.
Additional context Supporting a metrics and semantics layer would align WrenAI with modern data practices, as seen in tools like dbt and Cube. It could significantly improve data reliability, simplify integration with other tools, and provide a better experience for data consumers.
Links for reference:
@yu-iskw thanks for this feature request with so much details!
I'll put it into backlog of our roadmap first.
We did think about integrating existing semantic layer by transforming the semantics of dbt semantic layer or cube into our MDL format (https://docs.getwren.ai/oss/engine/guide/modeling/model). For example, Cube also has data modeling files with their own syntax (https://cube.dev/docs/product/data-modeling/syntax). Should be feasible to read these config files and turn into our MDL.
