</>MCP Agents Market
Agent

ROMA

by sentient-agi5.2kPythonUpdated 2026-02-16

Recursive-Open-Meta-Agent v0.1 (Beta). A meta-agent framework to build high-performance multi-agent systems.

ROMA (Recursive Open Meta-Agent) is a meta-agent framework that enables developers to build high-performance multi-agent systems using hierarchical task decomposition. It breaks down complex problems into parallelizable subtasks through a recursive plan-execute loop, with modules for atomization, planning, execution, aggregation, and verification. Built on DSPy, it provides transparent development with clear debugging paths and supports any LLM provider. The framework includes nine built-in toolkits for crypto, search, code execution, and file operations, with both minimal and production-ready Docker deployment options.

Key Features

Recursive hierarchical planning that decomposes complex tasks into atomic subtasks with dependency graphs
Five core modules: Atomizer (task classification), Planner (decomposition), Executor (execution), Aggregator (synthesis), Verifier (validation)
Built on DSPy with support for multiple prediction strategies (CoT, ReAct, CodeAct, BestOfN, Refine)
Nine built-in toolkits including crypto (CoinGecko, Binance, DeFiLlama), E2B code execution, and MCP server integration
Production-ready deployment with Docker including PostgreSQL persistence, MLflow observability, and REST API
Multi-LLM support with per-module model configuration (OpenRouter, OpenAI, Anthropic, Fireworks)
Execution-scoped storage with automatic Parquet serialization for large responses
Async execution support with graceful fallback to synchronous operations

Use Cases

  • 01Building autonomous research agents that decompose literature reviews into retrieval, analysis, and synthesis tasks
  • 02Creating crypto analysis agents that combine on-chain data, market metrics, and DeFi protocol information
  • 03Developing code generation systems that plan, implement, test, and verify complex software components
  • 04Constructing multi-step search agents that verify factuality across conflicting sources (benchmarked on SEAL-0, FRAMES, SimpleQA)
  • 05Implementing travel planning agents that coordinate research, booking, and itinerary creation
  • 06Building data migration agents that inventory, plan, execute, and validate database transitions

Related Agents

View more

ROMA — FAQ

What is ROMA and what does it do?+

ROMA is a meta-agent framework that uses recursive hierarchical planning to solve complex tasks by breaking them into parallelizable subtasks. It provides five core modules (Atomizer, Planner, Executor, Aggregator, Verifier) that work together in a plan-execute loop, with each subtask recursively processed until atomic operations are reached.

How do I install ROMA?+

For quick evaluation, install with `uv pip install roma-dspy` or `pip install roma-dspy`, then set your `OPENROUTER_API_KEY` environment variable. For production deployment with persistence and observability, use the Docker setup via `just setup` which configures PostgreSQL, MLflow, and the REST API server.

Which LLM providers does ROMA support?+

ROMA supports any LLM provider compatible with DSPy, including OpenRouter, OpenAI, Anthropic, Fireworks, and others. You can configure different models for different modules, and the default minimal installation uses OpenRouter with Claude Sonnet 4.5 for execution and Gemini 2.5 Flash for planning.

What API keys or prerequisites are required?+

At minimum, you need an OpenRouter API key (or OpenAI/Anthropic key). Optional features require additional keys: E2B API key for code execution, CoinGecko API key for crypto toolkit, and Serper API key for web search. The minimal installation works with just an LLM provider key.

Is ROMA free and open source?+

Yes, ROMA is open source under the Apache 2.0 license. However, you pay for the LLM API calls to your chosen provider (OpenRouter, OpenAI, etc.). Optional services like E2B code execution also have their own pricing.

What toolkits are included and how do I configure them?+

ROMA includes nine built-in toolkits: FileToolkit, CalculatorToolkit, E2BToolkit (code execution), CoinGeckoToolkit, BinanceToolkit, DefiLlamaToolkit, ArkhamToolkit, SerperToolkit (web search), and MCPToolkit (universal MCP server connector). Configure them in YAML files or via the executor's toolkit parameter with class names and API keys.

How do I install ROMA?+

Open the source repository on GitHub and follow its README. ROMA is a agent — MCP Agents Market links you directly to the official repo.

Is ROMA free?+

ROMA is an open-source project hosted on GitHub. Check the repository for its license and any usage requirements.

Related searches