Title: Request for Enhanced Multi-Agent Customization in Agent-Zero

[alexey2baranov]

Issue Description:

First of all, thank you for the incredible work on Agent-Zero! The framework is truly impressive.

I have been working on extending the code to fit my specific needs, but I encountered some challenges that I’d like to share with you. As you correctly pointed out in your YouTube video, Agent-Zero is extremely easy to customize because all prompts and tools are visible and accessible. This is particularly convenient for single-agent systems or systems with many similar agents.

However, when it comes to creating a state-of-the-art specialized agent (e.g., a programmer agent), I’ve found it challenging to do so without impacting other agents in the system. For example, if I update prompts, Docker configurations, and tools for a new programmer agent (Agent-1), these changes also affect the original Agent-0. As a result, Agent-0 becomes an expert programmer but loses its versatility in other areas.

Given that Agent-Zero is a multi-agent framework, it would be great to have a built-in mechanism to create agents with different characteristics without them interfering with one another.

Proposed Solution:

What I envision for Agent-Zero is a structure where:

Library with Default Configurations: Agent-Zero could provide a core library containing default prompts, tools, and Docker configurations.

Custom Agent Configuration: Users could then install this core library and create custom agents in their projects. Each agent could have its own folder structure, such as prompts/agent-1, prompts/agent-2, and prompts/agent-3, allowing them to use either the default prompts from the library or custom prompts.

Separate Docker Environments: Similarly, Docker environments and work_dir folders should be isolated for each agent to prevent conflicts. This would be especially important for specialized agents that require different Docker setups. Also files in work_dir's between different agents shuldn't be mix / conflict / affect each other.

I’m really interested to hear your thoughts on the possibility of setting up different agents in Agent-Zero, as well as the future direction of the framework.

Thanks again for the amazing work!