Skip to content

AI Chat

The AI Chat panel gives you a conversational interface to your product graph. The assistant understands your full model — domains, capabilities, features, relationships, and pending suggestions — and can both answer questions and apply changes.

Opening the panel

Click the wand icon (✦) in the top-right corner of the app to toggle the AI Chat panel open or closed. It slides in from the right without moving the main content.

What the assistant knows

The assistant has read access to the entire product graph for the active branch, including:

  • All nodes and their fields (titles, descriptions, statuses, etc.)
  • All relationships between nodes
  • Pending AI suggestions in the inbox
  • Commit history for the active branch

When you ask a question, it reads the relevant part of the graph to give you an accurate, up-to-date answer.

Asking questions

You can ask anything about your product model in plain language:

“What features are linked to the Checkout domain?”

“Which capabilities have no features yet?”

“Does the Express Checkout flow have all steps defined?”

“What needs are not addressed by any feature?”

“Show me all features with status ‘planned’ in the Payments domain.”

The assistant responds in natural language and references specific nodes from your graph.

Making changes

The assistant can also create and update nodes for you. Just describe what you want:

“Add an Acceptance Criteria to One-click Purchase: ‘User completes checkout in ≤ 3 taps’”

“Create a new Feature called ‘Guest Checkout’ under the Payment Processing capability with status planned”

“Add a Metric called ‘Checkout completion rate’ with a target of 80% (direction: up) to One-click Purchase”

“Rename the ‘Payments’ domain to ‘Checkout’”

Changes made through the chat go through the same pipeline as manual edits — they’re recorded as events, appear in your working state, and can be committed and rolled back.

Using AI Fix

From the AI Suggestions inbox, clicking AI Fix on a suggestion opens the chat with the suggestion context pre-filled. The assistant already knows what the issue is and what node it affects — you can immediately ask it to fix the problem.

Tips

  • Be specific about the node. “Add a metric to the login feature” works better than “add a metric” when you have many features.
  • Ask for summaries. “Summarize the state of the Checkout domain” gives you a quick health check without clicking through every node.
  • Iterate. If the first response isn’t quite right, follow up — the assistant remembers the conversation context.
  • Combine with the Node Generator. Use chat to ask questions and get context, then use the Node Generator to bulk-create a whole tree from a description.