AtScale

Overview

AtScale is a commercial semantic layer with OLAP heritage, and that heritage is the point. A model is a multi-dimensional layer over tables in your warehouse: star and snowflake schemas, hierarchies and levels, time intelligence and many-to-many relationships, queried in place rather than copied.

It matters most for Excel and Power BI users. AtScale emulates a Microsoft Analysis Services endpoint over XMLA, so Excel pivot tables speak MDX to it and Power BI speaks DAX, which is something few semantic layers offer. It also exposes ODBC, JDBC, a Python library and an MCP server for AI clients.

Performance comes from aggregates. The engine creates system-defined aggregates automatically and lets you add your own, rebuilds them incrementally as new fact data arrives, and offers SQL hints to control when they are used or generated, so queries hit summaries rather than scanning raw fact tables.

Models are authored in a visual design centre or written as code. Its modelling language, SML, was open-sourced under Apache 2.0 in 2024, with models stored as YAML in Git, which gives version control and review over metric definitions. dbt semantic models can be converted into SML.

Governance covers role-based access, row-level security through a security dataset, and catalogue integration with Alation and Collibra.

Supported warehouses are Databricks, BigQuery, Snowflake, Redshift, PostgreSQL and InterSystems IRIS. It is deployed as containers on Kubernetes with Helm rather than as a vendor-run service, and pricing is by deployed semantic object rather than by seat or query.

Key features and capabilities

The same headings are used for every semantic layer & metrics entry, so two tools can be read side by side.

How metrics are defined

• Multi-dimensional models over warehouse tables, with star and snowflake schemas
• Hierarchies, levels, time intelligence and many-to-many relationships
• Visual design centre, or models as code in SML, stored as YAML in Git
• dbt semantic models can be converted into SML

Query interfaces

• XMLA endpoint emulating Analysis Services: MDX for Excel, DAX for Power BI
• ODBC and JDBC connections per published catalogue
• AI-Link Python library for programmatic access
• MCP server for AI clients such as Claude and ChatGPT

Performance and caching

• System-defined aggregates created automatically, plus user-defined aggregates
• Incremental rebuilds process only new windows of fact data
• SQL hints control when aggregates are used or generated
• Queries resolve against aggregates instead of scanning raw fact tables

Governance and testing

• Git is the system of record for models, with per-object YAML files and branch controls
• Role-based access, with directory groups mapped to roles
• Row-level security through a security dataset and attribute filter keys
• Catalogue integration with Alation and Collibra

BI and tool integrations

• Warehouses: Databricks, BigQuery, Snowflake, Redshift, PostgreSQL, InterSystems IRIS
• BI: Excel, Power BI, Tableau, Looker and Google Sheets
• dbt metrics translator for converting semantic models
• AI and data science: Snowflake Cortex, Databricks Genie, Dataiku, DataRobot, pandas and Spark frames

How it runs

• Containers on Kubernetes, requiring Helm 3 or later
• Available through Google Cloud Marketplace and Snowflake Snowpark Container Services
• No vendor-run SaaS is published; named distributions and regions are not published

Pricing

Price on request

Quote-only: the pricing page publishes no figures. Pricing is consumption-based on deployed semantic objects, meaning published metrics, dimensions and models, with no per-seat or per-query fees, and volume discounts against a contracted commitment. The only number AtScale publishes is a 2024 blog post saying companies can start from around $2,500 a month. A free trial and a developer edition were announced in 2024 but are no longer advertised.

Vendor pricing page →

Demos and videos

• AI and semantic layers: live demo of Claude querying data with AtScale and MCP Product demo · YouTube
• Build semantic models with AI: AtScale MCP server demo Product demo · YouTube
• AtScale channel Channel or playlist · YouTube

About AtScale

AtScale, Inc. was founded by veterans of Yahoo's data team and says it introduced the first cloud-ready semantic layer in 2013. Its head office is in Boston, with a second office in Sofia, Bulgaria, and Chris Lynch is chief executive. Published funding runs from a $7m Series A in 2015 through a $50m Series D led by Morgan Stanley in 2018, with a further strategic investment led by Snowflake Ventures announced in December 2025, amount undisclosed. The platform is proprietary; only its modelling language, SML, is open source.

Founded 2013 · Boston, Massachusetts · atscale.com