ROMA - A Multi-Agent Framework Open-Sourced by Sentient AGI

What Is ROMA

ROMA (Recursive Open Meta-Agent) is a multi-agent system framework open-sourced by the Sentient AGI team. Through a recursive hierarchical structure, it decomposes complex tasks into parallelizable subtasks, which are assigned by parent nodes to child nodes for execution, and then the results are aggregated. ROMA supports multimodal input and output, and comes with a general task solver, a deep research agent, and a financial analysis agent, among others, making it suitable for a variety of scenarios from research analysis to financial decision-making. ROMA's transparent execution process facilitates debugging and optimization, and it performs excellently in multiple benchmarks, making itDeepResearchan open-source powerful tool.

ROMA's Main Features

• Recursive Task Decomposition: Supports automatically decomposing complex tasks into parallelizable subtasks, solving them step by step, and then integrating the results.
• Multimodal Support: The framework can handle multiple data types such as text, images, and code, adapting to the needs of different scenarios.
• Tool Integration: Supports the MCP protocol and API integration, and can call external tools and models.
• Transparent Debugging: Every step of the execution process is clearly visible, facilitating debugging and optimization.
• Built-in Professional Agents: Such as the general task solver, deep research agent, and financial analysis agent, meeting diverse needs.

ROMA's Technical Principles

• Recursive Hierarchical Structure: It uses a tree structure, where parent nodes decompose tasks into subtasks, and child nodes return results to parent nodes after execution.
• Core Components：
  • Atomizer: Determines whether a task is an atomic task, and if it is not atomic, triggers decomposition.
  • Planner: Decomposes complex tasks into subtasks and assigns them recursively.
  • Executor: Executes atomic tasks and calls LLMs, APIs, or other agents.
  • Aggregator: Integrates subtask results and returns them to the parent node.
• Context Flow Management: Decomposes tasks top-down and aggregates results bottom-up, ensuring clear information transfer.
• Modular Design: Supports inserting any agent, tool, or model at the node level, with high scalability.

ROMA's Project Address

• Project Website：https://blog.sentient.xyz/posts/recursive-open-meta-agent
• GitHub Repository：https://github.com/sentient-agi/ROMA

ROMA's Application Scenarios

• Research Analysis: The deep research agent automatically decomposes complex academic or market research tasks and integrates multi-source information to generate reports.
• Financial Decision-Making: The financial analysis agent monitors the cryptocurrency market in real time, integrates multiple data sources, and generates investment analysis reports.
• Project Planning: The general task solver decomposes project tasks, assigns and tracks progress, and supports efficient project management.
• Enterprise Automation: Builds multi-agent workflows to automate internal enterprise processes and improve operational efficiency.
• Educational Tools: Students use natural language to create research agents that automatically collect and integrate information to generate research reports.