bmo.cafe

Getting started

Bmo Cafe helps you generate and schedule recurring task scripts using LLMs. Basically, anything that can be thrown in a single Python file can be automated. The two task types that are currently implemented are:

• Generate browser requests based on a text prompt
• Generate and run Python functions and tests based on a text prompt

Clone or fork this repository. Make sure poetry is installed.

If you don't already have your OpenAI key set up, follow the instructions on their site.

Then, run:

• poetry install
• poetry run python main.py

Be sure to run your scripts from the root directory since some files rely on absolute file paths.

Finally, add my_data.yaml with your data. For example, to run the check for SF parking tickets, add the following information:

car_license: 'YOUR_LICENSE_PLATE_NUMBER'
car_state: 'YOUR_LICENSE_PLATE_STATE'

And you're all set up! ☕️

What tasks can this cafe do?

You can check out the example tasks in the tasks folder. You can do things like:

• Do things in the web browser that are based on HTTPS requests and responses
• Write in your style of voice

2 Paths

There are two ways you can create a task:

1. You provide the Python script that is executed.
2. A model automatically generates the Python script. (Sadly this only works at the moment for Python function generation with 1 number input and 1 number output. It does generate test sets though, which is cool.)

A task is represented as a single folder <TASK_NAME>. In this folder, you will find:

• index.py: this is the script that is executed.
• streams/: this is where the output streams of task execution is stored. It should be generated upon running the task for the first time.
• config.yaml: this is where the task configuration is stored.
• tests.csv (Optional): if this exists, it contains the input/output pairs to ensure that index.py does what it is supposed to.

Create a task by running poetry run python main.py and following the prompts. Add your script afterwards, if you are adding your own. Be sure to follow the steps below first.

Future implementations

Some future tasks:

• Generate text based on some writing sample
• Generate text edits based on some writing sample
• Generate natural language checks to ensure your text is faithful to writing sample