RecursiveMAS
by RecursiveMAS
Framework for recursive task-decomposing multi-agent systems
How It Works
Implements Recursive Multi-Agent Systems to coordinate agents that spawn and reason about sub-agents recursively. Uses a recursive task-decomposition pattern where agents create specialist sub-agents, aggregate their results, and iterate until a termination condition. Notable for separating reasoning layers and supporting latent collaboration strategies for scale and composability. See the A2A Protocol Pattern for related coordination patterns.
The Value Proposition
Target Use Cases
Researchers and engineers building or prototyping agent pipelines that use recursive delegation and need to study interaction/ failure patterns. This approach aligns with established patterns such as Tree of Thoughts Pattern and supports robust design through Defense in Depth Pattern.
Applications
- Decomposing complex tasks into recursive sub-agents to evaluate end-to-end reliability
- Experimenting with recursion depth and aggregation strategies to find failure modes
- Tracing how sub-agent outputs influence parent decisions for reputation and trust signals