connect to the AI data CLOUD that you already bought

Architected for Snowflake

Sigma is an agentic execution layer, AI Applications platform, and spreadsheet UI on top of your Snowflake AI Data Cloud. Build Sigma Agents and AI Applications without code.

Zero-Copy Architecture

Run AI natively on warehouse compute without any data extraction.

Closed-Loop Execution

Build Sigma Agents and AI Apps to write governed decisions directly to the warehouse.

Inherited Governance

Use existing role hierarchies and permissions without manual mapping.

Build and deploy AI Apps with Snowflake integrating Horizon Catalogues, security, and governance on Sigma. Easily manage inherited permissions, immutable audit, and maximize your Snowflake AI Data Cloud investment.

Snowflake-native architecture: If you can write a SELECT statement in Snowflake, Sigma can natively query it—including interactive tables and Semantic Views so existing security policies apply automatically.

One workspace, every user: Surface Cortex Agents to incorporte Cortex Search and Cortex Analyst in a spreadsheet or AI App.

Integrated enterprise AI: Write rich, real-time human-in-the-loop insights and actions directly to Snowflake with Sigma Agents.

Explore how Sigma Agents transform your governed data into automated action.

Explore the platform

Architecture at a glance

Execute everything inside the warehouse boundary.

Sigma sits directly on Snowflake—generating SQL, orchestrating Cortex AI functions, and surfacing agent-driven insights through workbooks, paginated reports, AI Apps, and embedded experiences.

When a user asks a question, Sigma determines the optimal path via Cortex Agents, whether Cortex Analyst for structured queries or Cortex Search for unstructured insights.

Snowflake-native execution

All AI processing, Cortex Agents orchestration, and SQL execution runs directly on Snowflake.

Inherited security

Sigma automatically respects Snowflake Role-Based Access Control (RBAC), RLS, and OAuth.

Deterministic outputs

Deliver reproducible, governed results in AI Applications with Sigma Agents for automated action.

End-to-end lineage

Maintain complete auditability of every pushdown and transformation from raw data to writeback.

Under the hood

See the parts architects ask about: compilation, execution paths, governance boundaries, and what you can measure.

Transformation

Sigma translates spreadsheet operations into SQL on the fly. Switch statements become CASE logic, moving averages become window functions, and pivots compile to your warehouse's dialect.

Query History shows the generated SQL for every element, with timing breakdowns and request IDs for warehouse tuning.

Inspect Query History

Execution paths

Sigma exposes query behavior including queue time, Sigma runtime, warehouse runtime, and result fetch time, plus admin usage dashboards and audit logs.

Explore warehouse usage templates

Governance & Roles

Sigma can run as the user (OAuth) or as a service account, and optionally map users/teams to warehouse roles. Or you can define access rules in Sigma.

Data access overview

Execution Paths

Not every click should wake up your warehouse. Sigma’s hybrid query engine evaluates the fastest, lowest-cost execution path by starting in the browser, then escalating through query ID caching, and only then to the warehouse.

Reuse results already in the user’s browser session. If Sigma can satisfy the request from what’s already been returned, there’s no network roundtrip, and no new warehouse work.

When the data is in-cache but the shape changed, Sigma recalculates “what changed” directly in the browser—think: new calcs, filters, regrouping—before issuing anything to the warehouse.

Sigma fingerprints the query structure and keeps a mapping to the warehouse query ID. If it’s been run before (and your platform supports result caching), Sigma can fetch the cached result via that ID without storing your query results in Sigma.

If Sigma has to go back to the warehouse, the warehouse can still return results from its own cache when conditions allow (many platforms keep result caching for up to ~24 hours depending on changes and determinism).

Only then do you light up compute. But Sigma pushes optimized SQL and merges work when possible to reduce the number of separate hits needed to render a full page of tables, charts, and pivots.

Your data stays in your warehouse storage. And for repeat-heavy logic, materialize expensive datasets into reusable warehouse tables and refresh on your schedule, so “fresh” doesn’t have to mean “expensive.”

Read the decision tree + caching tiers

Enterprise-Grade AI, Agents, Apps, and Analytics

Sigma is built for business teams that need flexibility without sacrificing governance or performance.

Visit Trust Center

Zero-copy query model

Sigma doesn't require you to duplicate warehouse tables into a separate store to get interactivity. Results can be reused via cache paths instead of persisting a separate copy.

Private connectivity (AWS/Azure/GCP)

Support for PrivateLink / Private Service Connect patterns when your security team wants to keep traffic off the public internet.

Auth or service account

Run per-user (OAuth) or via a service account. Choose what fits your governance model and auditing requirements.

Role-aware access control

Dynamically map users/teams to warehouse roles so row/column policies are enforced at query time.

Audit logs for admin events

Track key admin activities (logins, permission changes, connection changes, and more) for operational visibility.

Compliance artifacts in the Trust Center

SOC 2 Type II, ISO/IEC 27001, GDPR/CCPA posture, and other reports live in one place for review.

Fits into the rest of your stack

Sigma connects to your warehouse, and it also plays well with the systems around it whether its catalog, transformation, monitoring, or reverse ETL.

Reuse standardized metrics and keep business logic centralized.

Let users discover governed tables and definitions where they already look.

Keep an eye on pipeline and data quality issues that impact downstream analysis.

Operationalize curated outputs from the warehouse into downstream tools.

See our partner integrations

Sigma + Snowflake FAQ

The questions that usually come up once someone starts mapping Sigma into their Snowflake AI Data Cloud and governance model.

Does every click run a Snowflake virtual warehouse query?

No. Sigma’s query engine is optimized to integrate with Snowflake's native Result Cache which doesn't require active compute. We minimize redundant queries during exploration so that fresh SQL is pushed down to your virtual warehouse only when necessary.

What does warehouse-native architecture mean for Sigma and Snowflake?

It means Sigma translates interactions into optimized Snowflake SQL and pushes execution entirely down to Snowflake's compute engine. Sigma also natively integrates with Snowflake Cortex REST API to handle agentic orchestration. No data is extracted; Snowflake does the heavy lifting.

How do you keep Snowflake compute costs from getting out of hand?

Sigma limits unnecessary query generation. For highly complex or reused logic, teams materialize the output back into a Snowflake table. Downstream apps hit this precomputed table, drastically reducing compute costs.

Can we see the actual SQL Sigma is generating?

Yes. Query History shows the exact SQL, timing, and Snowflake Query ID. Data engineering teams can trace any Sigma action directly back to Snowflake’s Query History for auditing and tuning.

What does “inside the warehouse boundary” mean?

Snowflake remains your absolute system of record. Every user query, AI action, and data writeback executes entirely within Snowflake's security perimeter. There is no shadow database to secure.

What does “zero-copy” mean?

Sigma never extracts or duplicates your Snowflake tables into a separate BI engine. You are always querying live Snowflake data.

How do roles / RLS / CLS work with Snowflake?

Sigma fully supports Snowflake's role-based access control (RBAC). Through OAuth passthrough or Sigma's dynamic role-switching, the user's identity or role is directly passed to Snowflake. Your existing Snowflake Row-Level and Column-Level Security policies are automatically inherited and enforced at query time. No duplicate permission models are required.

What does "private connectivity" actually mean here?

If your instance is locked down, Sigma supports private networking (AWS PrivateLink, Azure Private Link). Traffic routes privately from Sigma to your Snowflake deployment without ever traversing the public internet.

What about data freshness?

Because there are no extracts, you are looking at live Snowflake data. You control cache duration (TTL) and can force a live query whenever up-to-the-second results are required.

Where does AI processing run in Sigma?

Sigma leverages Snowflake Cortex-hosted models like Mistral, Claude from Anthropic, Meta Llama, OpenAI, and more. This means all AI runs directly on Snowflake compute and data never leaves your perimeter for inferencing.

Architected for Snowflake

More powerful together.

Architecture at a glance

Under the hood

Workbooks generate warehouse-optimized SQL

You can attribute spend and tune from real usage

Drive access from the warehouse, Sigma, or both

You pay for queries. Not curiosity.

Enterprise-Grade AI, Agents, Apps, and Analytics

Fits into the rest of your stack

Sigma + Snowflake FAQ

AI Apps. Analytics.