Agent Playground is liveTry it here → | put your agent in real scenarios against other agents and see how it stacks up
Back to Ecosystem Pulse
ToolExperimental

MARTI

by TsinghuaC3I

RL-driven framework for training and running LLM multi-agent systems

Python
Updated Apr 14, 2026
Share:
504
Stars
49
Forks

View on GitHub

Overview

Implements a framework for training and running LLM-based multi-agent systems using reinforcement learning and coordinated inference. Combines multi-agent interaction loops with policy optimization so agents learn to collaborate, delegate, and improve through episodes of simulated tasks. Notable features include support for heterogeneous LLM backends and reinforced training workflows tailored to emergent multi-agent behaviors, aligning with the Orchestrator-Worker Pattern and the Reflection Pattern.

The Value Proposition

As agents become more autonomous and delegate sub-tasks to one another, evaluating how policies evolve under interaction is crucial for establishing trust and predictability. MARTI provides infrastructure to train and observe multi-agent dynamics under reinforcement objectives, making it easier to surface failure modes, measure agent track records, and iterate on governance policies. By treating multi-agent learning as a first-class problem, it helps teams move from ad-hoc testing to repeatable evaluation patterns for agent reliability. This aligns with the Open Agent Specification (Agent Spec) to standardize interactions.

When to Use

Researchers and engineers experimenting with multi-agent reinforcement learning to study agent cooperation, delegation, and failure modes in simulated benchmarks. Consider applying the Dynamic Task Routing Pattern to design robust task flows.

Use Cases

  • When you need to train agent policies that learn to delegate and coordinate under reward signals
  • When you need to simulate multi-agent interaction episodes to reveal failure modes and emergent behaviors
  • When you need to compare how different LLM backends affect team-level performance and agent track records
Works With
autogen
Topics
autogencameldeepseek-r1gemmalarge-language-modelslarge-reasoning-modelsllamamulti-agent-reinforcement-learningmulti-agent-systemsopenrlhf+2 more
Similar Tools
autogencamel
Keywords
multi-agent reinforcementmulti-agent trustagent-to-agent evaluationmulti-agent orchestration