Ask your data
Ask in plain language. Every data step is laid out for you and runs with the asker's own permissions.
A sentence you have heard
Can I actually trust the number the AI gave me?
Why it happens
A general-purpose AI can write a polished conclusion without saying where the number came from or what it means, and it may read data it should not. It looks like an answer and works like a risk.
How Shuzi does it
Ask your data. Answers come with provenance.
Ask in plain language. The assistant reaches data only through the semantic layer: list models, read metadata, run a query. Every step is laid out for you, and runs with the asker's own permissions.
Which region earns more?
- cube_list_models1 model
- cube_get_metadatarevenue_model
- cube_queryrevenue × region
Only data you are allowed to see
Sample answer: the North region earns more. Choose a region and quarter to see the sample result for that scope.
cube_list_models → cube_get_metadata → cube_query
- 01
Visible steps
Which tool was used and which members were queried, all in plain view and open to review.
- 02
One permission boundary
Models and rows you cannot see are invisible to the assistant too.
- 03
Your choice of model service
Admins configure an OpenAI-compatible provider, a model and a key, and the key is stored encrypted.
In detail
Semantic layer only
The assistant and external MCP clients share three tools, and neither reads physical tables.
Stateless conversations
The browser carries the history, and the server does not store conversations.
A step limit
Each conversation caps the rounds of tool calls, and errors are handed back to the model to handle.
Generate data apps
The assistant can also generate an HTML data app and preview it in a sandbox.
Take your datafrom tangled to clear.
Try it on sample data first, then bring your own.