AI you'll actually use

Omni currently uses AWS Bedrock-hosted Claude models for most tasks and OpenAI models only for advanced AI visualizations. Omni does not have its own LLM. For more, see our docs.

When a user enters a prompt, Omni shares the prompt with AWS Bedrock.

If the prompt requires data summarization, the metadata of the current query and a CSV of the results are shared with AWS Bedrock, which returns a summary.

If the prompt requires a new Omni query, then AWS Bedrock translates the prompt into an Omni semantic query. The semantic query is a collection of metadata –  including field names, filters, sorts, pivots, topic name, and limit – that is translated into SQL. The SQL is run against the data warehouse and the results are returned in the Omni interface.

Omni’s semantic model lets you feed important context and nuance into the AI to promote reliable responses, instead of expecting LLMs to deliver precise answers based on your raw data

In addition, while other tools are solely focused on enabling an “AI chatbot” experience, Omni’s AI lets you do more than just ask questions and get black-box answers. AI in Omni can accelerate every part of your data analysis workflow – including creating queries, filtering, writing Excel formulas, and more.

In addition, AI is embedded into Omni’s workbook experience rather than a separate interface, making it interoperable with the rest of Omni’s analysis methods. As you query with AI, you can drop into the point-and-click UI, SQL, or spreadsheet formulas at any point. Users aren’t locked into AI, even if that’s their starting point.

Customers can also embed Omni’s AI interface directly into their own product, or use Omni’s APIs and MCP server to customize further.

Omni currently doesn’t offer a way to “test” the accuracy of responses. However, we encourage data teams to optimize their data for AI and experiment with adding context to deliver more reliable results. By using this human-in-the-loop approach, teams can continually evolve their data model for AI to deliver more accurate, relevant answers.

Omni’s semantic model ensures AI responses aren’t a black box. For example, you might provide ChatGPT specific prompts like “You’re an expert on the economy. Summarize this article in 50 words” to get the results you want. Omni lets you do the same – you can provide instructions, context, and nuance to your data for more reliable responses.

For example, in your “Sales & Revenue” Topic, you might add context like: “Make sure your responses are easy for a busy sales leader to understand and act on.” You might get even more detailed on your Revenue field and specify: “This is the source of truth for our sales and finance reporting. It is the gross revenue from all of our sales channels. For any questions that ask for total sales, total revenue, total sold, or similar metrics, use this field.” Including these details helps AI better understand your data and deliver more helpful, trustworthy results.

Making your datasets AI-ready doesn’t need to be complicated – we’ve found that adding context in a few key places makes a big difference, and you can continue tuning up the model as you go. For tips, check out this how-to guide from our VP of Product Arielle, and watch our CEO Colin go from zero to AI-ready live.

Omni never shares your data without your permission. When AI features are toggled on, only necessary data is shared with external LLMs. For example, when asking AI to generate queries, or write formulas, only query metadata (column names, sorts, limits, etc.) is shared. If asking AI to summarize a query, then the results set is shared. Learn more in our docs.