Use Claude API from command line with llm tool

I've found Anthropic's Claude LLM to be very useful for writing code. The $20/month subscription is well worth it in terms of the efficiency I gain – but since I use it almost entirely for code, I'm also paying for lots of features I don't use (image upload, Google Docs integration, etc.)

Using the Claude API, which is charged by usage rather than flat rate, is cheaper for my needs. So I'm trying out Simon Willison's LLM CLI tool to interact with Claude from the command line.

First I installed LLM with pip:

pip install llm

Then following instructions here, I used llm to install the Anthropic plugin and set API key (generated from Claude console):

llm install llm-anthropic
llm keys set anthropic
# Then paste key when prompted

Now I can run prompts from command line, by first choosing the model and then providing the prompt as a string. For example:

This is great not only because I can interact with Claude more cheaply (and without opening a browser), but also because I can easily switch between LLMs.

I'm also particularly interested in downloading and running open-source LLMs on my local machine!