AgentVerse
https://github.com/openbmb/agentverse- Category
- AI Agents
- Rank
- No. 509Tools index
- Pricing
- Open Source
- Type
- AGENT
- Builder
- openbmb
- GitHub
- 5.1k stars
- Latest release
- v0.1.8.1
- Date
About
OpenBMB's framework for deploying multiple LLM-based agents across applications — simulation, task solving, and multi-agent collaboration.
What it does
AgentVerse lets you build coordinated groups of language-model agents and custom worlds where they act. One mode routes work among specialized roles such as solver, critic, executor, evaluator, and manager. Another advances simulated environments turn by turn, including user interaction. Tasks can run through command-line tools or a graphical interface.
Why it's ranked here
The project offers more than a conversation loop. It supplies distinct task-solving and simulation models, configurable registries, local-model options, benchmark loaders, command-line entry points, and a GUI. That breadth makes it useful for experiments. The verdict is tempered by active simulation refactoring, unfinished documentation, and memory improvements still listed as future work.
What's good
The architecture separates agents, environments, model loading, memory, output parsing, and decision rules through registries. Typed message variants distinguish criticism, execution requests, evaluation results, and role assignment. Users can install one Python package, invoke dedicated commands for benchmarks or workflows, and add optional dependencies for local models.
Tradeoffs
The simulation framework is explicitly undergoing refactoring, while the README directs stability-sensitive simulation users to an older branch. Documentation and richer conversation memory remain unfinished. The dependency set pins several older framework versions and includes heavy packages such as computer vision, language processing, and GUI libraries in the base installation. Some interface rendering code admits task-specific behavior.
How to use it well
Use AgentVerse for Python research prototypes that need explicit agent roles, turn-based environments, benchmark datasets, or visual demonstrations. Start from an existing task, then customize its agents, environment rules, and parsers. Prefer the older simulation branch when stability matters. It does not supply finished documentation or sophisticated long-term conversation memory, so plan to cover those needs separately.
Technical notes+
setup.py defines the agentverse setuptools package for Python 3.9+, installs requirements.txt, exposes a local extra from requirements_local.txt, and registers four console scripts. agentverse/__init__.py exports Simulation and TaskSolving while importing registries and initialization helpers. agentverse/registry.py implements decorator-based class registration and construction. agentverse/message.py models workflow messages with Pydantic. agentverse/gui.py connects Gradio and OpenCV rendering to both backends, but contains scenario-specific branches. pokemon_server.py exposes a FastAPI health check plus chat, decision, and location endpoints backed by Simulation.from_task. dataloader/gsm8k.py and dataloader/mgsm.py register JSON-lines benchmark loaders.
Observed
- License
- Apache Software License
- Primary language
- Python
- Python support
- Python 3.9 or newer
- Packaging
- Setuptools package with base requirements and an optional local-model dependency extra
- CLI
- Console commands for benchmarks, simulation, simulation GUI, and task solving
- Library interface
- Python exports for Simulation and TaskSolving
- Web interfaces
- Gradio GUI support and a FastAPI example server
- Platform classifier
- Operating System Independent
Read from README.md, setup.py, requirements.txt, pokemon_server.py, agentverse/gui.py, agentverse/demo.py, dataloader/mgsm.py, agentverse/utils.py, dataloader/gsm8k.py, agentverse/logging.py, agentverse/message.py, agentverse/__init__.py, dataloader/__init__.py, agentverse/registry.py.
What it can do
Deploy multiple LLM-based agents
Agent configurations and parameters → Running agent instances
Run multi-agent simulations
Simulation parameters and agent definitions → Simulation results and agent interactions
Solve tasks through agent collaboration
Task description and requirements → Task solution or completion status
Coordinate communication between agents
Agent messages and interaction protocols → Structured agent conversations and data exchange
Manage agent workflows
Workflow definitions and agent assignments → Executed workflow results
Monitor agent performance
Agent activity data → Performance metrics and status reports
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.