The universal semantic layer for ai
If it’s not AtScale, it’s just a guess.
AtScale is a universal semantic layer with a built-in AI Computation Engine (ACE). Everyone in your company can chat with any LLM and get the exact same number for “revenue.” You can automate with agents and trust the numbers they act on, and your compute costs stay contained. And yes, your Power BI dashboards still work too.
THE PROBLEM
Prompt an LLM for a business metric and it guesses at the SQL, even with a skill or an ontology to help. Prompt it again tomorrow and you can get a different number. Every guess scans your warehouse again, so you’re paying for answers that don’t match. In preliminary BIRD-Interact results, AI models with no semantic layer got about one enterprise question in five right.
Prompt the AtScale semantic layer and there’s no guessing. AtScale maps each prompt to the right business metric and calculates the answer for you. Every time.
“Revenue” can mean bookings, recognized revenue or contract value, and the model picks one without asking.
The database doesn’t say which column holds a metric, and the same value is spelled three different ways.
Real metrics take several passes over the right calendar, and a model that’s working alone takes shortcuts.
Teams rarely test AI answers, so they can’t tell which ones are wrong.
Agents write their own SQL. Then they act on whatever comes back, right or wrong.
one platform
Every request for a metric comes to AtScale first. ACE computes it in your data platform and sends back the answer.
Snowflake
Databricks
Google BigQuery
Microsoft Fabric
Data Warehouse
Ontologies Catalog
Spark
Ontologies
Data Catalogs
Metadata
Claude
Cursor
Chat GPT
Copilot
Codex
Power BI
Tableau
Excel
01
Turns metrics into SQL. Computes in place
02
Build and maintain semantic models using AI.
New
03
Navigate prompts to one metric. Ask when unsure.
New
04
Inherited access control. Enforced once.
New
05
Scores answers against a gold standard.
AI
BI
Semantic model
Prompt
Clarify
Answer
Serve
what’s in the platform
Ava works out which metric a question means before anything runs, whether a person or an AI agent asked it.
Interprets each question against your governed metric definitions
Confirms the meaning when a question could mean more than one thing
Gives people and AI agents the same answer to the same question
what’s in the platform
CoModel uses AI to build and maintain your semantic model.
Chooses which columns to use and how to compute each metric
Defines what each business term means and cleans up messy values
Sends every change to your data engineers for approval before it goes live
Coming this fall
what’s in the platform
Canary tests your semantic layer against your own business questions before your users do.
Runs the questions your business actually asks
Checks each answer against known-correct results
Sends every failure back to CoModel to fix
Coming this fall
what’s in the platform
The AI Computation Engine computes complex metrics in your data platform the same way every time.
Returns consistent results for complex metrics
Runs multi-pass and time-series calculations directly in your data platform
Keeps the aggregates it builds so repeat questions cost less
what’s in the platform
Governance applies your existing warehouse permissions in one place, so you don’t rebuild access rules.
Uses the warehouse permissions you already have
Versions and reviews metric definitions like code
Rolls back any change when you need to
Two ways to run it
You get the full platform on the warehouse you already use: Snowflake, Databricks, BigQuery, Redshift or Azure.
AtScale for Snowflake runs natively inside Snowflake, and it’s the engine inside Semantic Views.
integrations
Keep the data platform, BI tools and AI agents you have. If you switch one later, you won't redefine a single metric.
AtScale is a universal semantic layer. It defines your metrics once and computes them inside your own data platform. Your AI agents, analysts and apps all get the same answer, and repeat questions cost less to compute.
A semantic layer is where your company defines its metrics and computes them. When an AI agent or an analyst asks for revenue, it gets your definition of revenue, calculated the same way every time. AtScale is a universal semantic layer, so that holds for every model, agent, BI tool and data platform you use.
AtScale is one platform. ACE computes your metrics inside your data platform. Ava works out which metric a question means before anything runs. Governance carries your access rules through and versions every definition like code. CoModel, for building and maintaining the model with AI, and Canary, for testing answers against known-correct results, are coming.
No. AtScale computes each metric in your data platform and returns the same result to the BI tools and AI agents you already use.
Ava works out which metric a question means before anything runs, and ACE computes that metric in your data platform the same way every time. AtScale also gets more accurate as it runs. On the BIRD-Interact benchmark, accuracy rose from the first run to the third as it built up a memory of queries it had confirmed correct. Canary, coming soon, will test answers against known-correct results so you can see that improvement on your own questions.
Yes. ACE keeps the aggregates it builds, so the next question on the same data costs less to answer.
Yes. Blue Yonder consolidated a thousand dashboards and 800 tables into ten semantic models, cutting a four-day analysis to about ninety seconds. Snowflake chose AtScale as the engine inside Semantic Views and invested in the company. Carrefour France migrated 3,000 metrics and dimensions across 40 countries to AtScale.
Yes. AtScale queries data where it sits and doesn’t copy it out of your warehouse. To speed up queries, it builds aggregate tables inside your own warehouse, where your existing security controls still apply.
Permissions are defined once in the semantic model and enforced for every person and AI agent, whatever tool they use.
Most customers have their first governed semantic model live in 6-8 weeks. With AtScale’s new AI CoModel, we are shortening the time to value. A typical rollout connects AtScale to the warehouse, builds or imports the first model, then connects BI tools and AI agents.
AtScale is priced on consumption. Check out our pricing page for more information and contact sales for a quote.
