Ask the graph instead of the chat
The chat surface has two modes — the default (Chat) feeds the graph; Ask-KG reads from it. Tab between them in the workspace's chat-pane top bar. Same conversation, two different ways to interact with what's been extracted.
Chat vs Ask-KG — when to use which
Use Chat to add to the project (talk through a decision, document a risk, describe a new entity). Use Ask-KG to get something out ("what's our current strategy on X?", "who's responsible for the API rewrite?", "summarise the open risks"). Switching modes mid-conversation is fine; both share the conversation history.
How retrieval works
Each Ask-KG question runs a BFS subgraph walk from semantically-relevant seed nodes (depth 2, capped at 40 nodes). The model receives the subgraph as structured context, then answers. The answer's footer lists every node consulted, so you can audit exactly which entities the model leaned on.
What [uncertain] means in answers
When the answer cites a node that was extracted with confidence below 0.6, the citation is tagged [uncertain]. The model is telling you: "I used this, but the underlying entity isn't well-formed yet — refine it in Chat if it matters." Click an [uncertain] tag to jump to the node in the mindmap pane.
Source-node attribution
Every Ask-KG answer ends with a list of source nodes — the same set the BFS walk surfaced. This is the auditability promise: nothing the model says is unanchored. If a claim doesn't appear in the source nodes, treat it as model knowledge, not project knowledge — usually a sign to send a Chat message that locks the claim into the graph.
Ask-KG works best on projects with at least 20-30 nodes; below that, BFS retrieval often returns nothing useful and the answer falls back to whatever the model already knew. Send a few messages first, then ask.