Agentflow is a powerful yet user-friendly tool to run workflows powered by LLMs. You can:

• Write workflows in plain English in human-readable JSON files.
• Use variables for dynamic outputs that change based on user input.
• Build and execute custom functions to go beyond text generation.

Agentflow fills the gap between chat and autonomous interfaces:

• Chat (e.g. ChatGPT) can't run workflows because they're conversational.
• Autonomous (e.g. Auto-GPT) run them unreliably because they have too much freedom.

Agentflow offers a balanced solution: Workflows that LLMs follow step-by-step.

Agentflow is currently in development. To try it:

1. Sign up for the OpenAI API and get an API key
2. Clone or download this repository.
3. Create a .env file from example.env and add your OpenAI API key.
4. Run pip install -r requirements.txt to install dependencies.

Now you can run flows from the command line, like this:

python -m run --flow=example

python -m run --flow=example_with_variables --variables 'market=college students' 'price_point=$50'

python -m run --flow=example -v

Copy example.json or example_with_variables.json or create a flow from scratch in this format:

{
    "system_message": "An optional message that guides the model's behavior.",
    "tasks": [
        {
            "action": "Instruct the LLM here!"
        },
        {
            "action": "Actions can have settings, including function calls and temperature, like so:",
            "settings": {
                "function_call": "save_file",
                "temperature": 0.5
            }
        },
        {
            "action": "..."
        }
    ]
}

Copy save_file.py and modify it, or follow these instructions (replace "function_name" with your function name):

1. Create function_name.py in the functions folder.
2. Create a class within called FunctionName that inherits from BaseFunction.
3. Add get_definition() and execute() in the class. See descriptions of these in BaseFunction.

That's it! You can now use your function in function_call as shown above. However, you should probably:

4. Add tests in tests! Then you'll know if workflows are failing because of your function.

Agentflow is licensed under the MIT License.