GPT-agents is a simple blend between AutoGPT and Perplexity AI, with a focus on usability and user experience.
Assistant
|
? Therapist
|
?? Chef
|
? Investor
|
|---|---|---|---|
| The default agent, with a neutral personality | A therapist that can help you with your problems | A chef with great cooking expertise | An agent that has knowledge about finance and investments |
Features:
There are two options: chat mode and task mode. In chat mode, you can converse with the agent, and in task mode, you can let all agents work together to complete a task.
git clone https://github.com/avocardio/GPT-agents.git
pip install -r requirements.txt
credentials.json (without the "(template)")
Edit the config.json file to your liking
Run the program:
python chat.py
or
python task.py
Chat mode
Follow the on-screen instructions to select an agent (1-4), and then select the conversation history to be used, or press enter to use the current history. After this, you will be able to converse with the agent.
To reset and save a conversation, type "RESET" into the input field and press enter. This will save the current conversation to a timestamped history.json file in the agent's folder. After this, you will have a fresh instance of the agent.
Task mode
Agents
Currently there are 4 agents available.
To create more agents, simply create a file in the agents folder, and add a JSON file, or run the agent_creator.py script.
Browsing
Browsing is enabled when the user prompts using one of the following keywords:
["search", "browse", "research", "look up", "find", "look for", "google", "browsing", "googling", "looking for", "looking up", "looking up"]
This will trigger a google search where the first 3 pages are scraped and summarized, for the agent to read out to the user.
Whisper API
config.json. Then, you will be able to speak to the agent when its your turn ("You:") and accept the message with enter.a) Chef
b) Investor
To print the current message payload sent to the model (for debugging) you can enable debug-mode in the config.json file.