FractalMind
FractalMind
FractalMind is a recursive multi-directional thinking engine. It decomposes complex objectives into a tree of sub-agents, each exploring a different dimension, then synthesizes results from the bottom up — producing deeper analysis than a single LLM call can achieve.
FractalMind is patent pending.
How It Works
- You provide an objective — a question or task that benefits from multi-perspective analysis
- FractalMind creates a root agent connected to your character
- The root spawns sub-agents in three spatial directions, each exploring a different dimension
- Sub-agents can spawn further sub-agents, creating a fractal tree
- Results are synthesized bottom-up — leaf agents feed into their parents, which feed into the root
- The final synthesis is returned as the result
Each agent in the tree makes its own LLM call through your character's configuration, maintaining the character's identity and personality throughout.
Three Spatial Directions
Every branch in the fractal tree explores one of three directions:
| Direction | What It Explores | Example |
|---|---|---|
| Meta | Broader context — systemic patterns, assumptions, domain framing | "What larger forces shape this problem?" |
| Horizontal | Alternatives — different approaches, paradigm shifts | "What other approaches could we take?" |
| Focus | Depth — concrete steps, edge cases, technical details | "What specific implementation challenges exist?" |
This three-dimensional exploration ensures the analysis covers breadth, depth, and context simultaneously.
Three Temporal Modes
Control how long FractalMind runs:
| Mode | Behavior | Best For |
|---|---|---|
| Fast | Rush execution. Max depth 2, 60-second limit | Quick insights when time is limited |
| Capped | Run until time or token budget is exhausted, then synthesize what's available | Balanced analysis within a budget |
| Thorough | Wait for natural completion at all depths (default) | Maximum depth and quality |
Using FractalMind
From Strings
If your character has the FractalMind node enabled, it can trigger a session via:
[TRIGGER:fractal_mind:analyze the ethical implications of autonomous AI decision-making]
The session runs in the background. Results appear in the character's next response or in the FractalMind tab of the QUI Core dashboard.
From ThinkThing
Use the FractalMind node in a ThinkThing graph. Connect it to an Anima node and configure the objective in the node settings.
From the Dashboard
The FractalMind tab in the QUI Core dashboard lets you launch sessions directly, monitor progress, and visualize the branching thought tree.
[Screenshot: FractalMind visualization showing the branching tree structure]
Configuration
| Setting | Default | Description |
|---|---|---|
| Max Depth | 3 | How many levels deep the tree can grow |
| Max Agents Per Level | 3 | Maximum sub-agents spawned per parent |
| Max Total Agents | 20 | Total agent cap across the entire tree |
| Token Budget | 100,000 | Total tokens across all agents |
| Time Limit | 300 seconds | Maximum execution time |
| Write to Memory | true | Save results to the character's memory |
When to Use FractalMind
Good for:
- Complex strategic questions that benefit from multiple perspectives
- Research topics requiring breadth and depth simultaneously
- Decision analysis where you want to explore alternatives and consequences
- Creative problem-solving where unexpected connections matter
Not needed for:
- Simple factual questions
- Tasks with clear, single-path solutions
- Time-sensitive queries (even Fast mode takes multiple LLM calls)
Cost awareness: Each agent in the tree makes its own LLM call. A tree with 20 agents = 20 LLM calls. Use token budgets and agent limits to control costs.
