It's becoming easier and easier to build **AI agents**. Frameworks and platforms to build AI agents are becoming more mature with Microsoft (**Autogen**), Google (**Vertex**) and smaller players like **CrewAI** and **MetaGPT** joining in and creating platforms to make agent building easy and fun.
From an agent building perspective, an agent has two main components:
- Intelligence (reasoning and communication through language model capabilities)
- Tool integration (ability to connect to, usually third party, tools)
With these two components you can basically do anything you want an AI agent to do:
- Set goals and planning
- Communicate with humans
- Communicate with other agents
- Accessing third party data
- Actioning third party tools
Sounds simple and it is indeed very much a feasible task.
Some agent builders still require some coding but most platforms already have a no-code version available. In the no-code versions you can avoid coding for the most part, with an exception for installation (sometimes have to use a terminal) or for creating tools or skills (but you can easily let **ChatGPT** create pieces of code for tools/skills).
Most of the work required for the no-code versions is prompting the agents and configuring the agents and workflows to collaborate in the right way. Screenshot from AutoGen Studio
Although a lot of the platforms are still in their infancy (most are v0.5), you can use these platforms already in beta and join the communities of the platforms to learn how to build and meet other agent builders. Image from AgentOps presentation
Here's a list of the best Agent builders, their characteristics and how to access them.
- CrewAI
- AutoGen
- Google Vertex AI
- Superagent
- MetaGPT
1. CrewAI
CrewAI is a robust framework for building multi-agent systems where agents can assume roles, share goals, and operate cohesively. It supports sequential and hierarchical processes for task management and allows integration with various **LLMs** and third-party tools.
Ease of Use: Requires some coding, especially for creating custom tools and managing complex workflows.
Learning Resources: Comprehensive courses are available on their learning platform, and you can find detailed guides and examples in their documentation.
No-Code Version: While it doesn't have a fully no-code version, it offers user-friendly interfaces for managing agents, and some tasks can be configured with minimal coding.
Cost: CrewAI offers open-source tools, with enterprise-level support available via CrewAI+.
API Documentation: Detailed API docs are available here.
2. AutoGen
AutoGen is a framework by Microsoft designed to create custom GPT-based AI agents. It allows the development of LLM applications using multiple agents that can converse and collaborate to solve tasks.
Ease of Use: Requires coding, primarily for integrating custom tools and setting up the framework.
Learning Resources: There are detailed tutorials and examples available on the AutoGen Studio page and YouTube courses like the AutoGen Studio 2.0 Full Course.
No-Code Version: Yes, AutoGen Studio offers a no-code interface for defining and modifying agent workflows using a point-and-click, drag-and-drop interface.
Cost: AutoGen is available as a free Python package.
API Documentation: Comprehensive documentation can be found here.
3. Superagent
Superagent simplifies the creation of AI agents that can perform complex tasks with minimal configuration. It focuses on usability and integration with various third-party services.
Ease of Use: Typically no-code, with an emphasis on drag-and-drop configurations and prompt-based setups.
Learning Resources: Tutorials and examples are available on their documentation site, along with a YouTube channel for video guides.
No-Code Version: Yes, heavily focused on no-code solutions.
Cost: Pricing details are usually provided upon request or through enterprise agreements.
Learning Resources: Often provided through dedicated customer support and onboarding sessions.
API Documentation: Available through direct contact with the service provider.
4. MetaGPT
MetaGPT provides tools for building AI agents using the GPT framework. It emphasizes ease of use and integration with various tools and platforms to enhance agent capabilities.
Ease of Use: Some coding required, particularly for integrating advanced features.
No-Code Version: Limited no-code capabilities, more suited for users with some coding knowledge.
Cost: Pricing varies, usually based on enterprise needs and usage.
Learning Resources: Learning materials are available through technical documentation and community forums.
API Documentation: Usually provided to users with access; details are shared through direct engagements.
For detailed learning resources, DeepLearning.ai and YouTube have numerous walkthroughs and tutorials on these platforms, which can be immensely helpful for beginners and advanced users alike.
5. Google Vertex AI
Google Vertex AI offers a low-code environment for building AI models and agents. It integrates well with other Google Cloud services and allows for easy deployment and management of AI agents.
Ease of Use: Mostly no-code, with some coding required for advanced features and custom integrations.
Learning Resources: Extensive resources available on Google Cloud's training and certification site.
No-Code Version: Yes, it provides a no-code interface for many tasks.
Cost: Pay-as-you-go model, based on usage of Google Cloud services.
Learning Resources: Extensive resources available on Google Cloud's training and certification site.
API Documentation: Comprehensive documentation can be found here.
Pro tip: on DeepLearning.ai and Youtube you can find a lot of useful walkthroughs for the mentioned agent frameworks.
Enjoy building.