I've been trying to create a tool using AI coding LLMs for months to parse rss feeds, pull the best articles, and summarize into a series of tweets. I've tried multiple LLMs, chat interfaces, AI coding platforms, etc. They all came up short and left me frustrated and annoyed and a few times made me doubt what I've been predicting where I think AI will take us. I ran into errors with installing the binaries 30 days ago for Agent Zero, got back to work and didn't work to try to install it. Today I did the manual install, told it my goal and what I wanted to accomplish and 15 minutes later, it is giving me everything I wanted. I've often tweeted over the last 2 years that all apps will die and that we will end up in an agentic world where everything is goal/task oriented and in this platform I am seeing the beginning stages of that world. This is really really powerful. I am excited to keep learning how to make this work even better, but first run and I am blown away, because it is delivering exactly what I wanted. 🙏 @Jan Tomášek for creating something that actually works and can get us to where I think we need to be.

This is a quick example of how to add a new tool to agent zero, there may be other ways to accomplish this, but since i couldn't find a good example i will share my code here. This uses a free joke api just for an example proof of concept , most llms already contain many jokes by default, so its not very practical, but a current weather api implementation would be coded similar. To add a new tool to Agent Zero we need to create two new files and modify a third file. The steps to add a basic tool that requests a joke from an api and returns it to agent zero are: (note that this tool does not take any arguments, look at the existing 'knowledge_tool.py' on how to add arguments, such as 'question') 1. Add the new tool code by adding file: query_joke_api_tool.py 2. Create a markdown file that agent zero will include in the prompt so that the LLM knows what the tools capabilities are: agent.system.tool.joke.md 3. Modify the agent.system.tools.md file so that our new tool is included in the LLM prompt, practically registering our new tool. Agent zero automatically knows where to look to find and execute this tool. Explicitly ask the LLM to use this tool and it should use it. Step 1 Create file: # /python/tools/query_joke_api_tool.py import requests from urllib.parse import urlparse from python.helpers.tool import Tool, Response from python.helpers.errors import handle_error class QueryJokeApiTool(Tool): async def execute(self, **kwargs): url = "https://icanhazdadjoke.com/" headers = { "Accept": "text/plain"} print("*** Agent Zero has called the QueryJokeApiTool") try: # Attempt to make the API request response = requests.get(url, headers=headers, timeout=5) # Check if the request was successful (status code 200) if response.status_code == 200: myjoke = response.text.strip() return Response(message=f"Joke content:\n\n{myjoke}", break_loop=False)

I think it may be a good idea to add the possibility of using TAVILY API (tavily.com) for search, as they have a free tier.

I want to create a tool that can SSH into a server, and another tool that can connect to a database. So on the first call, these tools would have to establish connection, and on subsequent calls they would have to use the open connection. So the tool can't be completely 'stateless'. I suppose it could theoretically reconnect every time for every request, but that doesn't seem desirable. My question is, can tools have 'state', like an open connection handle?

Good evening A0 community, I hope you are well, healthy, and happy! It's been a few weeks since I've done a community check-in and figured it would be a good opportunity to see what kind of common or rare issues are being faced - potentially pushing users away from utilizing the Agent Zero framework! 1) What is something you have been trying to overcome but just can't seem to figure out? 2) What are some common error messages or system warnings you receive, often enough to disrupt your workflow? 3) What are some existing functionalities that you are experiences issues with, or perhaps a disappointment in result(s) with the use of these functions? (Example: tools, prompts, memory, etc) 4) What else do you want to share in the spirit of constructive criticism regarding your experiences thus far? 5) What can the community do to provide more personal assistance and ensure you don't feel as though you are being left out from conversation(s) such as this? Please leave your responses in the comments below and touch upon any or all of the subjects above. Do not hold back and do not be afraid to stand tall among the large number of community members, as that is the true way of growing, learning, expanding, connecting, and succeeding in life! I look forward to hearing your feedback and wish you all nothing but the absolute best, ALWAYS!