llm 0.32rc2

CLI utility and Python library for interacting with Large Language Models from organizations like OpenAI, Anthropic and Gemini plus local models installed on your own machine.

LLM

A CLI tool and Python library for interacting with OpenAI, Anthropic’s Claude, Google’s Gemini, Meta’s Llama and dozens of other Large Language Models, both via remote APIs and with models that can be installed and run on your own machine.

Watch Language models on the command-line on YouTube for a demo or read the accompanying detailed notes.

With LLM you can:

• Run prompts from the command-line
• Store prompts and responses in SQLite
• Generate and store embeddings
• Extract structured content from text and images
• Grant models the ability to execute tools
• … and much, much more

Quick start

First, install LLM using pip or Homebrew or pipx or uv:

pip install llm

Or with Homebrew (see warning note):

brew install llm

Or with pipx:

pipx install llm

Or with uv

uv tool install llm

Use LLM to run prompts or start chats against an arbitrary OpenAI-compatible Chat Completions endpoint, such as LM Studio. With uvx, you can do this without installing LLM first:

uvx llm openai endpoint http://localhost:1234/v1 \
  -m google/gemma-4-12b \
  "What is the capital of France?"

uvx llm openai endpoint http://localhost:1234/v1 \
  -m google/gemma-4-12b \
  --chat

Add --key your-api-key if the endpoint requires authentication. See Run against an endpoint without configuring it for more options.

If you have an OpenAI API key key you can run this:

# Paste your OpenAI API key into this
llm keys set openai

# Run a prompt (with the default gpt-5.6-luna model)
llm "Ten fun names for a pet pelican"

# Extract text from an image
llm "extract text" -a scanned-document.jpg

# Use a system prompt against a file
cat myfile.py | llm -s "Explain this code"

Run prompts against Gemini or Anthropic with their respective plugins:

llm install llm-gemini
llm keys set gemini
# Paste Gemini API key here
llm -m gemini-3.5-flash 'Tell me fun facts about Mountain View'

llm install llm-anthropic
llm keys set anthropic
# Paste Anthropic API key here
llm -m claude-5-opus 'Impress me with wild facts about turnips'

You can also install a plugin to access models that can run on your local device. If you use Ollama:

# Install the plugin
llm install llm-ollama

# Download and run a prompt against the Orca Mini 7B model
ollama pull llama3.2:latest
llm -m llama3.2:latest 'What is the capital of France?'

To start an interactive chat with a model, use llm chat:

llm chat -m gpt-4.1

Chatting with gpt-4.1
Type 'exit' or 'quit' to exit
Type '!multi' to enter multiple lines, then '!end' to finish
Type '!edit' to open your default editor and modify the prompt.
Type '!fragment <my_fragment> [<another_fragment> ...]' to insert one or more fragments
> Tell me a joke about a pelican
Why don't pelicans like to tip waiters?

Because they always have a big bill!

More background on this project:

• llm, ttok and strip-tags—CLI tools for working with ChatGPT and other LLMs
• The LLM CLI tool now supports self-hosted language models via plugins
• LLM now provides tools for working with embeddings
• Build an image search engine with llm-clip, chat with models with llm chat
• You can now run prompts against images, audio and video in your terminal using LLM
• Structured data extraction from unstructured content using LLM schemas
• Long context support in LLM 0.24 using fragments and template plugins

See also the llm tag on my blog.

Contents

• Setup

  • Installation
  • Upgrading to the latest version
  • Using uvx
  • A note about Homebrew and PyTorch
  • Installing plugins
  • API key management
    • Saving and using stored keys
    • Passing keys using the –key option
    • Keys in environment variables
  • Configuration
    • Setting a custom default model
    • Setting a custom directory location
    • Turning SQLite logging on and off

• Usage

  • Executing a prompt
    • Model options
    • Attachments
    • System prompts
    • Tools
    • Extracting fenced code blocks
    • JSON output
    • Schemas
    • Fragments
    • Continuing a conversation
    • Tips for using LLM with Bash or Zsh
    • Completion prompts
  • Starting an interactive chat
  • Listing available models
  • Setting default options for models

• OpenAI models

  • Configuration
  • OpenAI language models
  • Model features
  • OpenAI embedding models
  • OpenAI completion models
  • Adding more OpenAI models

• Other models

  • Installing and using a local model
  • OpenAI-compatible models
    • Run against an endpoint without configuring it
    • Configure an OpenAI-compatible model
    • Extra HTTP headers

• Tools

  • How tools work
  • Trying out tools
  • LLM’s implementation of tools
  • Default tools
  • Tips for implementing tools

• Schemas

  • Schemas tutorial
    • Getting started with dogs
    • Extracting people from a news articles
  • Using JSON schemas
  • Ways to specify a schema
  • Concise LLM schema syntax
  • Saving reusable schemas in templates
  • Browsing logged JSON objects created using schemas

• Templates

  • Getting started with –save
  • Using a template
  • Listing available templates
  • Templates as YAML files
    • System prompts
    • Fragments
    • Options
    • Tools
    • Schemas
    • Additional template variables
    • Specifying default parameters
    • Configuring code extraction
    • Setting a default model for a template
  • Template loaders from plugins

• Fragments

  • Using fragments in a prompt
  • Using fragments in chat
  • Browsing fragments
  • Setting aliases for fragments
  • Viewing fragments in your logs
  • Using fragments from plugins
  • Listing available fragment prefixes

• Model aliases

  • Listing aliases
  • Adding a new alias
  • Removing an alias
  • Viewing the aliases file

• Embeddings

  • Embedding with the CLI
    • llm embed
    • llm embed-multi
    • llm similar
    • llm embed-models
    • llm collections list
    • llm collections delete
  • Using embeddings from Python
    • Working with collections
    • Retrieving similar items
    • SQL schema
  • Writing plugins to add new embedding models
    • EmbeddingModel
    • Embedding binary content
  • Embedding storage format

• Plugins

  • Installing plugins
    • Listing installed plugins
    • Running with a subset of plugins
  • Plugin directory
    • Local models
    • Remote APIs
    • Tools
    • Fragments and template loaders
    • Embedding models
    • Extra commands
    • Just for fun
  • Plugin hooks
    • register_commands(cli)
    • register_models(register, model_aliases)
    • register_embedding_models(register)
    • register_tools(register)
    • register_template_loaders(register)
    • register_fragment_loaders(register)
  • Developing a model plugin
    • The initial structure of the plugin
    • Installing your plugin to try it out
    • Building the Markov chain
    • Executing the Markov chain
    • Adding that to the plugin
    • Understanding execute()
    • Prompts and responses are logged to the database
    • Adding options
    • Distributing your plugin
    • GitHub repositories
    • Publishing plugins to PyPI
    • Adding metadata
    • What to do if it breaks
  • Advanced model plugins
    • Tip: lazily load expensive dependencies
    • Models that accept API keys
    • Async models
    • Supporting schemas
    • Supporting tools
    • Attachments for multi-modal models
    • Structured messages and streaming events
    • Consuming prompt.messages in build_messages
    • Restoring opaque metadata on subsequent requests
    • Tracking token usage
    • Tracking resolved model names
    • LLM_RAISE_ERRORS
  • Utility functions for plugins
    • llm.get_key()
    • llm.user_dir()
    • llm.ModelError
    • Response.fake()

• Python API

  • Basic prompt execution
    • System prompts
    • Attachments
    • Tools
    • Schemas
    • Fragments
    • Model options
    • Passing an API key
    • Models from plugins
    • Accessing the underlying JSON
    • Token usage
    • Streaming responses
    • Structured messages and streaming events
  • Async models
    • AsyncResponse
    • Tool functions can be sync or async
    • Tool use for async models
  • Conversations
    • Conversations using tools
  • Listing models
  • Running code when a response has completed
  • Other functions
    • set_alias(alias, model_id)
    • remove_alias(alias)
    • set_default_model(alias)
    • get_default_model()
    • set_default_embedding_model(alias) and get_default_embedding_model()

• Logging to SQLite

  • Viewing the logs
    • -s/–short mode
    • Logs for a conversation
    • Searching the logs
    • Filtering past a specific ID
    • Filtering by model
    • Filtering by prompts that used specific fragments
    • Filtering by prompts that used specific tools
    • Browsing data collected using schemas
  • Browsing logs using Datasette
  • Backing up your database
  • The message store
    • Threads, turns, messages and parts
    • A worked example
    • Content addressing as a contract
    • Forking and shared history
    • Storage by reference
    • Table by table
    • Querying the message store
    • Logging from Python
  • SQL schema

• Related tools

  • strip-tags
  • ttok
  • Symbex

• CLI reference

  • llm –help
    • llm prompt –help
    • llm chat –help
    • llm keys –help
    • llm logs –help
    • llm models –help
    • llm templates –help
    • llm schemas –help
    • llm tools –help
    • llm aliases –help
    • llm fragments –help
    • llm plugins –help
    • llm install –help
    • llm uninstall –help
    • llm embed –help
    • llm embed-multi –help
    • llm similar –help
    • llm embed-models –help
    • llm collections –help
    • llm openai –help

• Contributing

  • Updating recorded HTTP API interactions and associated snapshots
  • Debugging tricks
  • Documentation
  • Release process

• Changelog