Connections
How edges between nodes work, and how they share context across AI conversations.
Connections
Connections (edges) are the lines drawn between nodes on the canvas. They do more than organise your diagram visually — they actively shape what the AI knows when you're chatting on any given node.
Drawing a connection
- Hover over a node. Small circular handles appear on its edges.
- Click and drag from a handle toward another node.
- Release over the target node. The connection is created.
You can connect any node type to any other. There are no enforced rules about which nodes can be linked — the structure is yours to define.
Removing a connection
Click an edge to select it, then press Backspace or Delete. The edge is removed immediately.
Removing a connection does not delete either node, nor does it affect existing conversations. However, future AI responses on those nodes will no longer have access to each other's context.
How connections affect AI conversations
When you open a node's chat panel and send a message, the AI receives context from across your canvas — not just the current node. Connected nodes are prioritised.
The context system works as follows:
- The current node — full structured data and recent conversation messages
- Directly connected nodes — full structured data for each connected node
- All other nodes in the project — a one-line summary only
This means a Feature node connected to your Tech Stack node gives the AI complete knowledge of your stack when you're discussing that feature. A Feature node with no connections still gets one-line summaries of your other nodes, but without the full detail.
Connect your Feature nodes to the Tech Stack and Audience nodes that are most relevant. The AI will use that full context to ask better follow-up questions and write more precise acceptance criteria.
Context budget
The context injected from other nodes is size-limited per conversation turn. If your canvas is large, Kommit prioritises connected nodes to ensure the most relevant information always makes it in. This keeps responses fast and focused.
Recommended connection patterns
| Use case | Suggested connections |
|---|---|
| Feature refinement | Feature → Tech Stack, Feature → Audience |
| Competitive positioning | Audience → Competitors |
| Architecture decisions | Tech Stack → Features (high-complexity) |
| Design assets | Assets → Feature they belong to |
| Repository context | Repository → Tech Stack or Project |
These are suggestions, not requirements. Connect nodes in whatever way reflects how the concepts relate in your product.