Fractals is an experimental open-source framework designed to orchestrate complex tasks using swarms of AI agents organized in a recursive structure. The system takes a high-level goal and decomposes it into a hierarchy of smaller subtasks, forming a self-similar tree that resembles a fractal structure. Each leaf node of this tree represents a specific executable task that can be processed independently by an AI agent. The framework runs these subtasks in isolated Git worktrees so agents can work on separate components of a project without interfering with one another. By coordinating these agents through a recursive task decomposition strategy, the system allows large problems to be solved collaboratively by multiple AI processes. The architecture is designed to experiment with large-scale agent coordination and distributed problem solving.
Features
- Recursive task orchestration that decomposes goals into fractal-like subtask trees
- Agent swarm architecture that distributes work across multiple AI agents
- Execution of subtasks within isolated Git worktrees for safe parallel development
- Automatic generation of executable subtasks from high-level objectives
- Framework for experimenting with distributed AI collaboration systems
- Infrastructure for managing large multi-step workflows across multiple agents