Example Agents - supercog-ai/community GitHub Wiki

Building real world agents with Supercog

This page describes some demo and practical agents and how they are constructed.

The Weather agent

This agent demonstrates the "Hello World" of AI agents. We've made the agent a little more interesting by using both the Weather tool and the Image Generator tool.

The agent instructions are simple:

When the user asks for a weather report, get the weather data and then generate
an image which matches the weather report.

To use the agent we can ask it any basic weather question:

What is the weather like today in Los Angeles?

and we get both our weather report and a matching image.

The Meeting Prep Researcher

This agent performs web research to help you prepare for a meeting. The best part is that you can often get good personal and company background information by providing nothing more than the email address of the person you are meeting.

These agent instructions are a little more complicated:

### A meeting researcher agent using web browser and email tools.
### Replace <YOUR EMAIL> with your email address.
### Add your email to the EMAIL_WHITELIST env var.

The user will supply an email address plus some other identifying information. 
Do your best to perform extensive web research:
- About the company the person works for
- About the person themselves

and prepare a detailed "meeting prep" report based on the information you gather. 
You should do up to 5 web search and retrieve up to 5 other pages in order to 
gather information.

Then send the report to <replace with YOUR EMAIL>.

The # lines are just comments. The actual instructions inform the agent that it will get some basic info about the subject of your meeting, and then it should prep a report including personal and company research on the subject.

The last line tells the agent to send the report to your inbox. You should edit the instructions before running the agent.

This agent uses three tools:

  1. The Tavily search tool is purpose built to enable web search from LLM applications.
  2. The Web Browser tool allows the agent to download full pages that may be summarized in the Tavily search results.
  3. The Email tool enables the agent to email you the report.

Note that as a security measure, the Email tool will not send messages to any random address. You must configure the Env Var EMAIL_WHITELIST with one or more email addresses, separated by commas, that you want to enable as recipients. You should add your email to this list so that the agent can email you meeting reports.

The "Roasting" agent

You are a professional insult comic. You have long experience telling hilarious jokes which
insult your target. When you receive a name or email address, do some extensive web
research about the person and then roast them with a few choice jokes. Use markdown 
for the output. Then email the result to <enter email>.

Add the Tavily, Web Browser, and Simple Email tools.

⚠️ **GitHub.com Fallback** ⚠️