Ask, Build, and Act with Sigma AI.

Most standalone AI tools force your team to manually validate every output. Sigma Assistant and Sigma Agents reason over the business logic already defined in your workbooks and dbt semantic models while providing complete transparent lineage so you can view the exact formulas, filters, and tables the AI used to build its answer.

Sigma Assistant is how you learn and build. Ask a question, explore your data, generate a chart, and assemble a full application with natural language. Sigma Agents are how you deploy intelligence. They are configured by your team, scoped to specific data and instructions, and embedded directly in workbooks and apps to run on your behalf. They answer questions, monitor metrics, trigger API calls, write data back to your warehouse, and push alerts to external systems like Salesforce or Slack.
Sigma Assistant creates the product. Sigma Agents power it after it ships.

Yes. Sigma Agents call external APIs and can act. Common examples include creating Salesforce opportunities, posting Slack or Teams alerts, triggering Zendesk tickets, and writing enriched rows back to your warehouse. Actions can be user-initiated within the conversation or scheduled to run automatically. Human-in-the-loop approval steps can be added wherever sign-off is required.

Sigma Agents are scoped to specific data sources and instructions your admins define so that responses are grounded in approved context, not open-ended inference. Every response surfaces the data behind it, so users drill down instead of taking answers on faith. For higher-stakes workflows, configure inline response feedback and pair it with a Q&A audit log written back to your warehouse to ensure a continuous signal for improving the models underneath.

No. Sigma Assistant lives directly inside our familiar spreadsheet interface, empowering non-technical users to build full applications conversationally. Using simple text prompts, users can instantly generate complex spreadsheet calculations, structure dashboards, and assemble workflows. If your team knows how to describe their business rules, Sigma Assistant can write the syntax.

A dashboard shows data. An AI App acts on it. The difference is functional: AI Apps combine data visualization with write and actions functionalities. Users can input, update, or approve data and trigger conditional actions to external workflows or API calls. Integrate a Sigma Agent into your AI App for maximum value and automation potential.

Yes. Sigma’s AI inherits your existing warehouse row-level security (RLS) and column-level masking. The AI can only query data the specific user is authorized to see. There is no shadow AI or secondary permission model to manage, even in embedded deployments. This applies across all surfaces whether it's Sigma Assistant or Sigma Agents so no AI pathway bypasses your existing warehouse access controls.

Yes, using AI Query. You can call LLM functions directly within your workbook columns to classify, extract, and summarize unstructured data at warehouse scale. Because this processing runs natively on your cloud data warehouse compute, you can enrich your data without extracting it, standing up external infrastructure, or writing custom Python scripts. Furthermore, enrichment runs as a warehouse function so your standard column-level permissions and audit logs apply. Unstructured AI analysis stays within your existing data governance framework.

Sigma Assistant gives you a live, interactive data product. When you ask a standalone chatbot a question, it returns static markdown text or a disconnected chart. If you want to drill down into those numbers, you have to start over. Sigma Assistant instantly generates a fully functional, explorable workbook connected directly to your warehouse.

Start by asking a question in natural language with Sigma Assistant. One click turns the chat into a production-ready, live application. From there, you can deploy agents to monitor the app's metrics and automate actions—moving from question to enterprise workflow on one governed platform.

You control this. Sigma is completely model-agnostic. If you choose a warehouse-native model like Snowflake Cortex, your data never leaves that environment. If you opt for external providers like Azure OpenAI, Sigma only sends the minimum required inputs under your secure enterprise agreement.

When using warehouse-native models, your data stays entirely within your cloud environment and is never used to train public models. When using external providers like Azure OpenAI or Anthropic, your enterprise agreement with that provider governs data usage. Standard enterprise contracts often explicitly prohibit using customer prompts or data for model training. Sigma itself never stores, retains, or uses customer data beyond generating your requested AI outputs.

We don't force a specific LLM. Administrators can configure Sigma to use your existing enterprise AI contracts—whether that’s Anthropic, Azure OpenAI, or warehouse-native models—ensuring total alignment with your already-approved corporate AI strategy. Sigma integrates with the model your security team has already approved so you're not forced to adopt a new AI vendor or negotiate a new contract.

Claude and Sigma serve two complementary halves of your AI strategy. Claude helps with conversational intelligence, while Sigma provides a secure enterprise execution layer and AI Applications platform. By connecting them, your teams can use Claude's reasoning capabilities and still rely on Sigma to enforce row-level security, guarantee accurate calculations, and turn text answers into production-ready, automated workflows built for enterprise scale.

Sigma operates as a bidirectional Model Context Protocol (MCP) hub. As a client, Sigma Assistant and Sigma Agents can pull context from external systems like your CRM or document stores. As a server, Sigma securely exposes IT-approved semantic models and workflows to trusted external AI tools. MCP support is currently in private beta. Contact your account team for more detail.