AB
AB
AiBoss

All Categories

project

ROMA - Sentient AGI open-source multi-agent framework

ROMA (Recursive Open Meta-Agent) is an open-source multi-agent system framework from the Sentient AGI team. It uses a recursive hierarchical structure to break down complex tasks into parallelizable subtasks, which are then assigned by parent nodes to child nodes for execution...

What is ROMA?

ROMA (Recursive Open Meta-Agent) is an open-source multi-agent system framework from the Sentient AGI team. It uses a recursive hierarchical structure to break down complex tasks into parallelizable subtasks, which are then assigned by parent nodes to child nodes for execution, and the results are aggregated. ROMA supports multimodal input and output and includes a built-in general task solver, deep research agent, and financial analysis agent, making it suitable for various scenarios ranging from research and analysis to financial decision-making. ROMA's transparent execution process facilitates debugging and optimization, and it performs exceptionally well in multiple benchmark tests, making it a powerful open-source tool for DeepResearch.

ROMA's main functions

  • Recursive task breakdownIt supports automatically breaking down complex tasks into parallel subtasks, solving them step by step, and then integrating the results.
  • Multimodal supportThe framework can handle various data types such as text, images, and code, adapting to different scenario requirements.
  • Tool IntegrationSupports MCP protocol and API integration, and can call external tools and models.
  • Transparent debuggingEach step of the execution process is clearly visible, making it easy to debug and optimize.
  • Built-in professional agentExamples include general task solvers, deep research agents, and financial analysis agents, catering to diverse needs.

ROMA Technical Principles

  • Recursive HierarchyThe system uses a tree structure, where the parent node breaks down the task into sub-tasks, and the sub-nodes execute the tasks and then send the results back to the parent node.
  • Core components:
    • Atomizer: Determine if the task is an atomic task; if not, trigger disassembly.
    • PlannerBreak down complex tasks into subtasks and recursively assign them.
    • Executor: Execute atomic tasks, call LLM, API or other agents.
    • AggregatorIntegrate the results of subtasks and send them back to the parent node.
  • Context flow managementDecompose tasks from top to bottom and aggregate results from bottom to top to ensure clear information delivery.
  • Modular designIt supports inserting any Agent, tool, or model at the node level, and is highly scalable.

ROMA project address

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

Application scenarios of ROMA

  • Research AnalysisThe Deep Research Agent automatically breaks down complex academic or market research tasks and integrates multi-source information to generate reports.
  • Financial DecisionsThe financial analysis agent monitors the cryptocurrency market in real time and integrates multiple data sources to generate investment analysis reports.
  • Project PlanningThe general task solver breaks down project tasks, assigns them, and tracks progress, facilitating efficient project management.
  • Enterprise AutomationBuild multi-agent workflows to automate internal enterprise processes and improve operational efficiency.
  • Educational toolsStudents use natural language to create research agents that automatically collect and integrate information to generate research reports.
Back