Tree chat & AI suggestions
Every node in the Tree has a Chat tab in its Blade — the side panel that opens when you click a node. This is where the AI Data Advisor surfaces suggestions for that specific node, and where you can ask the advisor questions about its reasoning.
If you did not apply advisor suggestions while creating the dataset, you can review them again in the Tree.
Opening the Chat
- Open the Tree for a project.
- Click any node — the Blade slides in.
- Click the Chat tab.
What you’ll see
The Chat panel has two parts:
Suggestions list
The advisor’s recommendations for the current node — for example:
- Columns it suggests keeping (high signal).
- Columns it suggests dropping or flagging (low signal, likely noise).
Each suggestion is a card you can act on directly.
Apply suggestions
Click a suggestion card to apply it to the current selection. Once applied, the card changes state so you can see which suggestions are already in use.
Ask questions
A chat box at the bottom, with placeholder text “Ask about these suggestions…”. Type a natural-language question and the advisor responds.
Examples of useful questions:
- “Why did you suggest dropping
internal_id?” - “What would change if I applied the first three suggestions?”
- “Which columns are most relevant to customer-retention questions?”
The advisor responds in the conversation thread above the input.
What the chat won’t do
The chat is intentionally narrow. Its scope is explaining the suggestions and their filtering effects on this node — nothing else.
If you ask broader questions, you’ll get a response like:
“I can only explain these data suggestions and their filtering effects.”
For open-ended analysis, use RebelCore Agent after training a semantic dataset.
A practical workflow
- Open the Tree, click a spec.
- Switch to Chat.
- Read through the suggestions; apply the ones that match your intent.
- Ask a follow-up question if a suggestion isn’t clear.
- When the selection looks right, switch to DATA and click Train Model.