Spatial Chat
HiveMind OS offers two chat modes: linear (classic top-to-bottom conversation) and spatial (an infinite 2D canvas). Spatial mode turns conversations into visual maps where the arrangement of ideas carries meaning.
The Canvas
The spatial canvas is an infinite 2D workspace. Every message — yours and the agent's — becomes a card positioned in space. Cards connect via edges that show relationships. Drag, group, and arrange cards freely to organize your thinking.
Card Types
| Card Type | Description |
|---|---|
| Prompt card | Your question or instruction |
| Response card | The agent's reply, linked to its prompt |
| Artifact card | Code, images, documents — first-class objects, not inline blobs |
| Reference card | Pinned external context (files, URLs, data) |
| Cluster card | A named group for related cards — like a folder on the canvas |
Edges & Relationships
Cards connect via typed edges:
- Reply-to — conversational flow (prompt → response)
- References — "this card draws on context from that card"
- Contradicts — flags tension between two ideas
- Evolves — a later card supersedes or refines an earlier one
The agent auto-creates edges as the conversation flows, but you can also draw edges manually to express relationships the agent might not infer.
Contextual Prompting
Instead of a single global input box, you can drop a prompt card near relevant context. The agent automatically scopes its response to nearby cards. Ask a question near your database schema cards, and it knows you're asking about data modeling — no re-explaining needed.
You can also explicitly select cards to include in the next prompt's context: select them, then type your question. Only the selected cards are sent — giving you precise control over what the agent "sees."
Three ways to prompt
- Global prompt — type at the bottom bar; card appears at canvas center
- Contextual prompt — click empty canvas space; inline text field appears at that position
- Card-attached prompt — click the reply button on any card for a follow-up
Fork & Explore
Drag a card to an empty region to start a tangent. The agent treats it as a sub-conversation while retaining awareness of the parent thread. This lets you explore two approaches side-by-side without cluttering a single thread.
This works naturally with session forking — fork the conversation at any card to create an independent branch.
Gather & Synthesize
Select multiple cards, right-click → Synthesize. The agent reads all selected cards and produces a summary card that distills the key points. This is perfect for converging after divergent exploration — research three options, then synthesize the trade-offs into one card.
Agent Behaviors on the Canvas
The agent is a co-organizer, not just a responder:
- Auto-clustering — notices thematic overlap and suggests grouping
- Gap detection — spots two clusters with no connection and asks "should these relate?"
- Conflict surfacing — draws a red edge between contradictory cards
- Spatial memory — remembers where things are; you can say "go back to what we discussed in the top-left"
When to Use Spatial vs. Linear
| Scenario | Mode |
|---|---|
| Focused single-thread Q&A | Linear |
| Quick task or command | Linear |
| Brainstorming with multiple threads | Spatial |
| Comparing approaches side-by-side | Spatial |
| Research with many sub-topics | Spatial |
| Architecture discussions | Spatial |
| Visual mapping of agent reasoning chains | Spatial |
Switch between modes at any time from the session toolbar. Your conversation history is preserved — switching to spatial lays out existing messages as cards, and switching back to linear presents them chronologically.