I built a package registry for OSI models - sharing what I learned

[pasar6987]

It's my first time discussion, so I'm a little shy.
Grammar might be weird because I'm not familiar with English.

I'm a data engineer doing semantic layer modeling at my company. While dealing with OSI, I realized that there was no central repository to search, preview and share OSI packages, so I made it myself.
https://hub.rawctx.dev

I want to make rawctx an independent, vendor-neutral package registry for semantic models.

It currently supports OSI and dbt MetricFlow formats.

Key features include:

• You can preview the structure of the OSI package (datasets, measures, dimensions, relationships) before downloading
• Native packages can be downloaded from the CLI
• Indexed packages are clearly marked for preview only and guide the source
• There is also a private workspace for internal team sharing

pip install rawctx

rawctx search "shopify revenue"
rawctx info @dbt-hub/jaffle-shop
rawctx snapshot-download @pasar6987/zoho-expense

What I discovered while actually implementing OSI v1.0 (0.1.0) in the registry:

• Package-level metadata is not part of the spec. Things like version, author, license, and scope are essential for distribution but currently live outside OSI. I had to define my own packaging convention on top of the spec.
• It's not easy to validate OSI files. Flexible specs are good for adoption, but from the registry's point of view, we need to set the "publicizable level" criteria ourselves. I was wondering what others think of validation. (Regarding OSI-level validations? #35, Use templating engine instead of plain yaml #62 discussions)

It's still a nascent service and there are a lot of shortcomings. I welcome feedback from the community. I'm especially curious about how spec should be done to better support packaging and deployment workflows.