Spatial Reasoning for
Prompt Engineering
Visualize trade-offs, orchestrate logic, and generate structurally superior directives for LLMs.
Language Models struggle when you give them a flat list of conflicting instructions. They don't know which rule matters most when two rules collide. Constellation is a "Creative Algebra" tool that solves this by allowing you to map your instructions spatially.
What is it?
Constellation replaces the linear text editor with a gravitational canvas. You position "Nodes" (ideas, constraints, roles, goals) relative to a central problem.
The physics of the map generates the logic of the prompt:
Distance = Priority
Nodes closer to the center are treated as "High Amplitude" directives—non-negotiable rules. Nodes at the edge are "Low Amplitude" context—nice to have, but flexible.
Angle = Context
Grouping nodes together creates thematic clusters. The tool reads your map radially to form coherent paragraphs of instruction.
Interference
Visualize how constraints pull against each other. If you want "Creative" and "Concise", placing them on opposite sides tells the system they are competing forces to look out for.
Who is it for?
- Prompt Engineers who need to maintain complex system prompts without getting lost in 2000 words of text.
- Systems Architects designing agent behaviors where trade-offs (e.g., Speed vs. Accuracy) need to be explicitly tuned.
- Creative Strategists who want to "sketch" a strategy before committing to the exact wording.
Operating Instructions
- Define the Phase: Use the top bar to select your mode (Block, Shape, Refine). This sets the preamble for the AI.
- Add Nodes: Click "Add" in the right panel or use the mobile controls. Give your node a name (e.g., "Code Quality") and a role (e.g., "Constraint").
- Position on Stage:
- Drag Important rules to the center (Green Zone).
- Drag Contextual details to the edge (Grey Zone).
- Group related concepts together.
- Generate: Click "Generate Prompt". The tool calculates the weights based on position and compiles a structured Markdown prompt.
- Iterate: Paste the prompt into your LLM. If the AI is too rigid, drag the constraint nodes further away. If it's ignoring a rule, drag it closer.
Use Cases
1. The "Senior Engineer" Persona
Place Security and Performance very close to the center. Place
New Features and Speed at the edge. This generates a persona that
prioritizes stability over innovation.
2. Creative Writing Assistant
Place Vocabulary and Tone nodes in a cluster. Place
Plot Constraints on the opposite side. The generated prompt will balance style
instructions against narrative requirements.
3. Debate Simulator
Create two clusters of nodes on opposite sides of the circle (180 degrees apart). One cluster represents "Pro" arguments, the other "Con". This explicitly structures the prompt to see a polarized view.