simonw/llm ? reverse-engineered prompt

Build me a Python command line tool, plus a small importable library, for talking to lots of AI models from one place. I want to be able to save API keys, send a prompt from the terminal, pipe file contents in, attach things like images, and start an interactive chat session. It should work with OpenAI out of the box and also support other providers like Claude, Gemini, and local models through installable plugins.

Please make it feel practical for everyday use. I want prompt and response history stored in SQLite, support for embeddings, a way to extract structured data from text or images using schemas, and support for model tool use where available. It should be easy to switch models, set a default model, and continue conversations.

Also include solid docs, sensible tests, and examples for common flows like asking a question, explaining a file, extracting text from an image, and chatting in the terminal. Look up current docs online if you need to.