My Notes on the LLM CLI Tool

The LLM CLI tool by Simon Willison provides a simple yet powerful way to interact with various language models directly from your command line. This guide covers installation, configuration, and practical usage patterns to help you get started with this versatile tool.

Links

• GitHub: https://github.com/simonw/llm
• Documentation: https://llm.datasette.io/en/stable/

Installation

Using uv:

uv tool install llm

Plugin Installation

Install plugins for specific providers (using Azure as example):

llm install llm-azure

• llm-azure: https://github.com/fabge/llm-azure

API Key Configuration

Set your API key:

llm keys set azure

Azure Setup

Add a configuration file to the required azure directory of llm installation:

dirname $(llm logs path)

The configuration file should look like:

- model_id: gpt-4-32k
  model_name: gpt-4-32k
  api_base: https://your_deployment.openai.azure.com/
  api_version: '2023-05-15'

- model_id: text-embedding-3-small
  embedding_model: true
  model_name: text-embedding-3-small
  api_base: https://your_deployment.openai.azure.com/
  api_version: '2023-05-14'

• model_id: Used by the LLM CLI
• model_name: Passed to the Azure API

Set Default Model

llm models default <model_id>

Using LLM Prompts

Pipe Input

cat my_python_script.py | llm "explain this code"

Attachments

llm "describe this image" -a https://static.simonwillison.net/static/2024/pelicans.jpg

Supports both URLs and file paths.

System Prompts

echo "import fastapi" | llm "explain this code" -s "You are a python expert and explain complex concepts concisely with simple examples"

Templates

List Templates

llm templates

Edit Templates

llm templates edit summarize

Provide default values for parameters:

prompt: 'Translate to $language: $input'
defaults:
  language: 中文

Use Templates

llm -t translate -p language '中文' "I love you"