LLM by Datasette

LLM is Simon Willison's open-source command-line tool and Python library for talking to large language models from a terminal. Out of the box it speaks to OpenAI, Anthropic, Google Gemini, Meta Llama and dozens of other providers through a plugin system, and it stores every prompt and response in a local SQLite database that you can query with Datasette. It supports multi-modal inputs (text, images, audio, video), schema-based structured extraction, tool use, embeddings, fragments and templates, and interactive chat sessions.

It is aimed at developers, researchers and power users who want a single, scriptable interface to many models without committing to one vendor's SDK. Pricing is free under Apache 2.0 — you only pay the underlying model providers, and you can run entirely offline against local models via the Ollama or llama.cpp plugins. The CLI composes well with Unix pipes, which makes it a favourite for shell-based pipelines, batch evaluations and ad-hoc data extraction work.

The plugin ecosystem is the real differentiator: community plugins add new model backends, embedding stores, tool integrations and output formatters, and the SQLite log doubles as a queryable history you can mine later. The trade-off is that it is a developer tool — there is no GUI, no hosted service, and no team features.