USE CASES / SEMANTIC MODELING
Semantic modeling: build the model in days, and keep it current.
AtScale CoModel drafts your semantic model from the documentation you’ve already got, and a person on your team approves every definition.
A semantic model says what your business numbers mean and how they’re computed. Building one by hand has taken teams months, and keeping it current takes more. AtScale combines AI-assisted modeling with a visual workspace and an open modeling language, so your team spends its time on the decisions only people can make.
The challenge
Building a semantic model by hand takes months.
Definitions drift as tables, columns and business rules change.
Context is scattered across catalogs, dbt projects, docs and people’s heads.
What you get
A first model drafted from your existing context, ready for review.
Definitions that stay current, with every change tested and approved.
THE CHALLENGE
Automated definitions need a person’s review.
Anthropic’s data team found that metric definitions generated automatically from raw tables scored worse than a smaller set a person curated (source). AI can do the drafting, but a person has to own each definition. Doing all of it by hand is the bottleneck, and it’s why many teams still don’t have a semantic layer.
HOW IT WORKS
From your existing documentation to a model in production.
- Gather the context. Point CoModel at your warehouse schemas, data catalogs, dbt projects, documentation and modeling standards. It saves what it finds as modeling memory.
- Draft the model. CoModel writes the model in SML, AtScale’s open-source modeling language, adds the context agents need, and flags the calls only a person can make.
- Review and approve. Your data engineer reviews each definition, in the coding agent they’re already using or in Design Center, AtScale’s visual workspace.
- Test and ship. Canary tests the draft against your questions, and Governance promotes it through review from dev to QA to production.
- Keep it current. When Canary finds a miss or a live question exposes a gap, CoModel drafts the fix and the loop runs again.
CoModel and Canary are coming this fall. Design Center and SML are available today. See how AtScale works →
WHAT’S IN A MODEL
Metrics, dimensions and the rules that connect them.
An AtScale model holds connections, datasets, metrics, calculations, dimensions with their hierarchies, relationships and row security. Each object is a YAML file in Git with a plain-language description, so people, agents and BI tools all ask for Revenue instead of querying a raw table.
THE PROOF
All 22 BIRD-Interact models, generated in a couple of hours.
For the BIRD-Interact benchmark, a preview of CoModel generated all 22 semantic models in a couple of hours from the benchmark’s schemas and knowledge base, and they weren’t tuned by hand. With those models, an LLM was twice as accurate. AtScale’s results are preliminary.
FAQ
Questions about semantic modeling.
Can we build a model before CoModel ships?
Yes. Design Center, AtScale’s visual workspace, and SML are available today. CoModel and Canary are coming this fall.
Do we need a data catalog first?
No. CoModel, coming this fall, will read a catalog if you’ve got one. It’ll also work from schemas, dbt projects and documentation.
Which LLM does CoModel use?
The one in your coding agent, such as Claude Code or Codex. AtScale will publish approved configurations.
Can we import models we've already built?
Yes, as context. CoModel will read the models you’ve already built and use them when it drafts.
Does CoModel change the model on its own?
No. Every change it drafts will be an SML diff that’s tested and approved by a person before it’s promoted.
Is SML open source?
Yes. SML is AtScale’s open-source modeling language. Each object is a YAML file in your Git repository with a plain-language description.
Book a workshop.
The workshop takes thirty days from kickoff to results.
- Bring us your most controversial metric and the business questions your team and your agents ask about it every week.
- We benchmark your AI on them, with AtScale and without it.
- You keep the semantic model we built to run the test.