@wsl8297: Discovered a recursive multi-agent framework on GitHub: ROMA (Recursive Open Meta-Agent). It solves complex problems through a hierarchical recursive structure, decomposes tasks into parallelizable components, and enables agents to handle more intricate reasoning tasks. GitH…

Discovered a recursive multi-agent framework on GitHub: ROMA (Recursive Open Meta-Agent). It solves complex problems through a hierarchical recursive structure, breaking down tasks into components that can be processed in parallel, enabling agents to handle more complex reasoning tasks. GitHub: https://github.com/sentient-agi/ROMA… Core Workflow: - Atomizer: Determines whether a task can be executed directly or requires further planning. - Planner: If planning is needed, decomposes the task into smaller subtasks. - Executor: Executes atomic tasks, supporting calls to LLMs, APIs, or other agents. - Aggregator: Collects and integrates results from subtasks to generate the answer for the parent task. - Verifier (Optional): Validates whether the aggregated results meet the original objectives. The information flow is tri-directional: tasks are recursively decomposed top-down, results are aggregated bottom-up, and dependencies are executed sequentially from left to right. This structure allows the system to flexibly handle complex problems while maintaining clear transparency. The framework is built on DSPy and supports various prediction strategies (CoT, ReAct, CodeAct, etc.). It offers two installation methods: a minimal installation for a 30-second setup via pip to use core features; and a full installation supporting Docker deployment, including production-grade features such as PostgreSQL persistence, MLflow observability, and REST API servers. The default configuration uses OpenRouter (Claude Sonnet 4.5 + Gemini 2.5 Flash), but also supports other LLM providers like OpenAI and Anthropic. It includes built-in toolkits for calculators, file operations, etc., and can integrate extended functions such as the E2B code execution sandbox and S3 storage. Suitable for developers who need to build hierarchical multi-agent systems and handle complex reasoning tasks, or those interested in understanding the design of recursive agent frameworks.