Cassis reads your data environment, builds a structured layer of business meaning on top of your data (the ontology), and lets anyone or any agent ask business questions in natural language. Every answer is grounded in the ontology, comes with the SQL that produced it, and carries full provenance.
The ontology sharpens with every use. Every correction and clarification from chat feeds back, and recurring confusion or failing evals surface as prioritized refinements you can review.
You can use Cassis through three interfaces: the web app, where you chat with your data and curate the ontology; the MCP server, which exposes the same engine to the analytics agents you run in Claude Code, Claude Desktop, Cursor, and other MCP clients; and the Slack app, where your team asks the bot questions in the channels they already work in.
Access is limited to selected design partners today. Leave your email at getcassis.com to be notified when it opens up.
What is an ontology?
A structured representation of how your business thinks about its data. It sits between your raw warehouse tables and whoever is asking questions, so an agent reasons about business meaning first, then the data path.
What it captures
The domains your data is organized into, the tables and columns that hold it, how those tables join, the metrics that matter, and the business context and rules behind each one.
Richer than a semantic layer
A semantic layer is just dimensions and metrics. An ontology also carries your business concepts, the rules behind each metric, and how every piece maps to the data.
Richer than a data catalog
A catalog documents what exists. An ontology also encodes what it means and how to use it correctly.
Carries provenance
Every piece tracks where it came from, who changed it, and when.
How it works
Reads your environment
Connects to your dbt project, warehouse schema, and documentation, or parses a DDL file when there's no live connection. Combines all of them to bootstrap the ontology.
Builds a structured ontology
Domains, enriched tables and columns, joins, and metrics, with business context written into each domain. Each piece carries the source it came from, so you can trace it back to a dbt model, a SQL column, or a piece of docs.
Versions every change
You publish ontology versions like commits, each with a label and a diff of what changed. Connect a GitHub repository and publishing opens a pull request for review, so each version goes live once it merges. You can also import the ontology back from the repo.
Improves through use
Every question asked, every ambiguity clarified, every correction made feeds back. Recurring gaps and failing evals cluster into prioritized refinements your data team reviews. Nothing becomes truth without their sign-off, so the ontology gets sharper as a side effect of normal work, and stays trustworthy.
Serves humans and agents
The same engine powers the web app, the MCP server, and the Slack app. People and analytics agents query the same ontology, with the same grounding, and get the same auditable answers.
Ways to use Cassis
Cassis runs one engine behind three surfaces: the web app, the MCP server, and the Slack app. The docs are organized by job, so jump to the part that fits you. Most teams touch all three.