2 min

# Try the guided demo

The demo is self-guided inside the product, and the same sandbox also runs from your terminal. Either route shows the core Cassis loop: an agent hits a context gap, you teach it once, and that knowledge becomes governed context for every future question.

## Start the tour

1.  [Create a free account](https://app.getcassis.com/auth/sign-up). Cassis opens a seeded fictional marketplace project called **Stallora**.
    
2.  Follow the tour in the product. It first asks a question the ontology cannot answer reliably, so you can see the missing context rather than a polished result.
    
3.  Teach Cassis the missing definition in one line. The tour shows how that correction becomes a versioned metric, including the SQL behind it.
    
4.  Ask the question again. This time Cassis uses the new context immediately. Then continue in the sandbox with your own questions.
    

That gap-to-fix loop is the product: context improves through use, while your data team keeps every change reviewable and governed.

## Prefer the terminal?

The same sandbox runs headless. You ask over MCP, write the fix in a checkout, prove it with the CLI, and publish it by merging a pull request. Press **Use this template** on [cassis-demo-stallora](https://github.com/GetCassis/cassis-demo-stallora) for your own repository, pre-wired with the sandbox’s ontology, MCP config, and CI. Its README is the walkthrough, about fifteen minutes.

## Next

-   [Choose how to manage your ontology](/start/choose/): Use git by default, or the web app if the people maintaining it do not work in git.
-   [Understand an answer](/ask/answers/): Read the plan, SQL, and provenance attached to every result.
-   [Connect any MCP client](/ask/mcp/): Use Cassis from any MCP-compatible client, including Claude Code, Claude Desktop, and Cursor.