What it does
LLM is a command-line tool and Python library for talking to language models. Its documentation lists OpenAI, Anthropic's Claude, Google's Gemini, Qwen, Gemma, Kimi, DeepSeek and Mistral among dozens of supported models, reached either through remote APIs or as models installed on your own machine. From the terminal you can run a prompt, start an interactive chat, pipe a file in with a system prompt, attach an image to extract text, and store every prompt and response in SQLite. It can also generate and store embeddings, extract structured data with schemas, and let models run tools. Other providers and local runtimes such as Ollama are added by installing plugins, and any OpenAI-compatible endpoint, for example one served by LM Studio, can be used directly. Installation is through pip, Homebrew, pipx or uv, and hosted providers need your own API keys. The documentation also covers prompt templates, fragments, model aliases and a tutorial for writing your own model plugin.
Capabilities
- Run prompts and chats from the command line or from Python
- Hosted models and local models, added through plugins such as the Ollama one
- Logs prompts and responses to SQLite
- Generates, stores and searches embeddings
- Schemas for extracting structured data from text and images
- Tools, templates, fragments and a plugin system
When to use it
- Trying the same prompt across several providers from a terminal
- Scripting model calls inside shell pipelines
- Keeping a searchable log of your prompts and responses
- Writing a plugin to add a model that is not yet supported
