AgentVerse
AgentVerse is a framework for deploying multiple LLM-based agents in applications, offering task-solving and simulation frameworks. It solves the problem of coordinating multiple agents to collaboratively accomplish tasks or simulate social behaviors. It is for developers and researchers building multi-agent systems with large language models.
✨ Key features
- Task-solving framework for collaborative multi-agent problem solving
- Simulation framework for custom environments and agent interactions
- Supports local LLMs like LLaMA and Vicuna via FastChat
- Supports vLLM for high-throughput inference
- Includes GUI for simulation environments
- Provides CLI tools for benchmarking and task execution
🎯 Use cases
- Software development system with multiple agents
- Consulting system for collaborative problem solving
- Simulation of social behaviors in games or research
- Running benchmarks like HumanEval with multi-agent collaboration
- Tool-using tasks such as solving 24-point game with web search
📦 Installation
🧰 Requirements: Python >= 3.9, OpenAI API key (or Azure OpenAI key), optional local model dependencies for LLaMA/Vicuna.
git clone https://github.com/OpenBMB/AgentVerse.git --depth 1
cd AgentVerse
pip install -e .
If you want to use AgentVerse with local models such as LLaMA, you need to additionally install some other dependencies:
pip install -r requirements_local.txt
Or you can install through pip:
pip install -U agentverse
🚀 Usage
For simulation (CLI):
export OPENAI_API_KEY="your_api_key_here"
agentverse-simulation --task simulation/nlp_classroom_9players
For task-solving (CLI):
agentverse-tasksolving --task tasksolving/brainstorming
⚠️ Good to know
The simulation framework is being refactored; for a stable version exclusively supporting simulation, use the release-0.1 branch. Documentation and more sophisticated memory for conversation history are coming soon.
❓ FAQ
How do I set up my OpenAI API key?
Export your OpenAI API key as an environment variable: export OPENAI_API_KEY="your_api_key_here". For Azure OpenAI, export AZURE_OPENAI_API_KEY and AZURE_OPENAI_API_BASE.
Can I use local models like LLaMA with AgentVerse?
Yes, AgentVerse supports local models via FastChat. Install additional dependencies with pip install -r requirements_local.txt, launch the local server with bash scripts/run_local_model_server.sh, and set llm_type: local in your config file.
What is the difference between task-solving and simulation frameworks?
Task-solving assembles multiple agents to collaboratively accomplish tasks, while simulation allows users to set up custom environments to observe behaviors among or interact with multiple agents.
How do I run a benchmark like HumanEval?
Use the command: agentverse-benchmark --task tasksolving/humaneval/gpt-3.5 --dataset_path data/humaneval/test.jsonl --overwrite.
📊 Repository
🤖 Overview, features, install steps and FAQ were generated from the project's README on Sep 4, 2026. Always check the original source before running commands.