🐫 CAMEL is an open-source group devoted to discovering the scaling legal guidelines of brokers. We imagine that learning these brokers on a big scale gives useful insights into their behaviors, capabilities, and potential dangers. To facilitate analysis on this discipline, we implement and assist numerous forms of brokers, duties, prompts, fashions, and simulated environments.
CAMEL Framework Design Rules
🧬 Evolvability
The framework permits multi-agent methods to constantly evolve by producing information and interacting with environments. This evolution could be pushed by reinforcement studying with verifiable rewards or supervised studying.
📈 Scalability
The framework is designed to assist methods with tens of millions of brokers, guaranteeing environment friendly coordination, communication, and useful resource administration at scale.
💾 Statefulness
Brokers preserve stateful reminiscence, enabling them to carry out multi-step interactions with environments and effectively sort out refined duties.
📖 Code-as-Immediate
Each line of code and remark serves as a immediate for brokers. Code needs to be written clearly and readably, guaranteeing each people and brokers can interpret it successfully.
Why Use CAMEL for Your Analysis?
We’re a community-driven analysis collective comprising over 100 researchers devoted to advancing frontier analysis in Multi-Agent Techniques. Researchers worldwide select CAMEL for his or her research based mostly on the next causes.
| ✅ | Giant-Scale Agent System | Simulate as much as 1M brokers to check emergent behaviors and scaling legal guidelines in complicated, multi-agent environments. |
| ✅ | Dynamic Communication | Allow real-time interactions amongst brokers, fostering seamless collaboration for tackling intricate duties. |
| ✅ | Stateful Reminiscence | Equip brokers with the power to retain and leverage historic context, enhancing decision-making over prolonged interactions. |
| ✅ | Assist for A number of Benchmarks | Make the most of standardized benchmarks to carefully consider agent efficiency, guaranteeing reproducibility and dependable comparisons. |
| ✅ | Assist for Totally different Agent Sorts | Work with a wide range of agent roles, duties, fashions, and environments, supporting interdisciplinary experiments and various analysis functions. |
| ✅ | Information Technology and Device Integration | Automate the creation of large-scale, structured datasets whereas seamlessly integrating with a number of instruments, streamlining artificial information era and analysis workflows. |
What Can You Construct With CAMEL?
1. Information Technology
2. Job Automation
3. World Simulation
Fast Begin
Putting in CAMEL is a breeze due to its availability on PyPI. Merely open your terminal and run:
pip set up camel-ai
Beginning with ChatAgent
This instance demonstrates the right way to create a ChatAgent utilizing the CAMEL framework and carry out a search question utilizing DuckDuckGo.
- Set up the instruments package deal:
bash pip set up 'camel-ai[web_tools]'
- Arrange your OpenAI API key:
bash export OPENAI_API_KEY='your_openai_api_key'
- Run the next Python code:
“`python from camel.fashions import ModelFactory from camel.varieties import ModelPlatformType, ModelType from camel.brokers import ChatAgent from camel.toolkits import SearchToolkit
mannequin = ModelFactory.create( model_platform=ModelPlatformType.OPENAI, model_type=ModelType.GPT_4O, model_config_dict={“temperature”: 0.0}, )
search_tool = SearchToolkit().search_duckduckgo
agent = ChatAgent(mannequin=mannequin, instruments=[search_tool])
response_1 = agent.step(“What’s CAMEL-AI?”) print(response_1.msgs[0].content material) # CAMEL-AI is the primary LLM (Giant Language Mannequin) multi-agent framework # and an open-source group targeted on discovering the scaling legal guidelines of brokers. # …
response_2 = agent.step(“What’s the Github hyperlink to CAMEL framework?”) print(response_2.msgs[0].content material) # The GitHub hyperlink to the CAMEL framework is # https://github.com/camel-ai/camel. “`
For extra detailed directions and extra configuration choices, take a look at the set up part.
After operating, you possibly can discover our CAMEL Tech Stack and Cookbooks at docs.camel-ai.org to construct highly effective multi-agent methods.
We offer a 
demo showcasing a dialog between two ChatGPT brokers enjoying roles as a python programmer and a inventory dealer collaborating on creating a buying and selling bot for inventory market.
Discover various kinds of brokers, their roles, and their functions.
In search of Assist
Please attain out to us on CAMEL discord for those who encounter any subject arrange CAMEL.
Tech Stack
Key Modules
Core parts and utilities to construct, function, and improve CAMEL-AI brokers and societies.
| Module | Description |
|---|---|
| Brokers | Core agent architectures and behaviors for autonomous operation. |
| Agent Societies | Parts for constructing and managing multi-agent methods and collaboration. |
| Information Technology | Instruments and strategies for artificial information creation and augmentation. |
| Fashions | Mannequin architectures and customization choices for agent intelligence. |
| Instruments | Instruments integration for specialised agent duties. |
| Reminiscence | Reminiscence storage and retrieval mechanisms for agent state administration. |
| Storage | Persistent storage options for agent information and states. |
| Benchmarks | Efficiency analysis and testing frameworks. |
| Interpreters | Code and command interpretation capabilities. |
| Information Loaders | Information ingestion and preprocessing instruments. |
| Retrievers | Information retrieval and RAG parts. |
| Runtime | Execution surroundings and course of administration. |
| Human-in-the-Loop | Interactive parts for human oversight and intervention. |
| — |
Analysis
We imagine that learning these brokers on a big scale gives useful insights into their behaviors, capabilities, and potential dangers.
Discover our analysis tasks:
Analysis with US
We warmly invite you to make use of CAMEL in your impactful analysis.
Rigorous analysis takes time and sources. We’re a community-driven analysis collective with 100+ researchers exploring the frontier analysis of Multi-agent Techniques. Be a part of our ongoing tasks or check new concepts with us, attain out by way of e-mail for extra data.
Artificial Datasets
1. Make the most of Numerous LLMs as Backends
For extra particulars, please see our Fashions Documentation.
Information (Hosted on Hugging Face)
2. Visualizations of Directions and Duties
Cookbooks (Usecases)
Sensible guides and tutorials for implementing particular functionalities in CAMEL-AI brokers and societies.
1. Primary Ideas
2. Superior Options
3. Mannequin Coaching & Information Technology
4. Multi-Agent Techniques & Functions
5. Information Processing
Contributing to CAMEL
For individuals who’d prefer to contribute code, we respect your curiosity in contributing to our open-source initiative. Please take a second to assessment our contributing pointers to get began on a easy collaboration journey.🚀
We additionally welcome you to assist CAMEL develop by sharing it on social media, at occasions, or throughout conferences. Your assist makes a giant distinction!
Group & Contact
For extra data please contact [email protected]
-
GitHub Points: Report bugs, request options, and monitor improvement. Submit a difficulty
-
Discord: Get real-time assist, chat with the group, and keep up to date. Be a part of us
-
X (Twitter): Comply with for updates, AI insights, and key bulletins. Comply with us
-
Ambassador Challenge: Advocate for CAMEL-AI, host occasions, and contribute content material. Study extra
Quotation
@inproceedings{li2023camel,
title={CAMEL: Communicative Brokers for "Thoughts" Exploration of Giant Language Mannequin Society},
creator={Li, Guohao and Hammoud, Hasan Abed Al Kader and Itani, Hani and Khizbullin, Dmitrii and Ghanem, Bernard},
booktitle={Thirty-seventh Convention on Neural Data Processing Techniques},
12 months={2023}
}
Acknowledgment
Particular due to Nomic AI for giving us prolonged entry to their information set exploration software (Atlas).
We’d additionally prefer to thank Haya Hammoud for designing the preliminary brand of our mission.
We applied wonderful analysis concepts from different works so that you can construct, evaluate and customise your brokers. In case you use any of those modules, please kindly cite the unique works: – TaskCreationAgent, TaskPrioritizationAgent and BabyAGI from Nakajima et al.: Job-Pushed Autonomous Agent. [Example]
License
The supply code is licensed underneath Apache 2.0.
