llm

llm Plugin

llm is Simon Willison's unix-friendly LLM CLI. Pipe text to any model, save prompt templates, log every conversation to SQLite, and build embedding collections — extended by a large plugin ecosystem.

Commands

Prompting

• llm prompt run — Send a prompt to the default model (stdin is appended to the prompt)
• llm embed run — Embed text/files into a collection
• llm _ _ — Passthrough (llm chat, llm similar, llm schemas, llm install)

Management

• llm self version — Show version
• llm models list — List models (--options shows per-model options)
• llm plugins list — List installed plugins
• llm keys manage — keys list, keys set <name>, keys path
• llm templates list — List stored prompt templates
• llm logs list — Conversation history (-n 10, --json, -m <model>)

Usage Examples

• "Summarize this file with GPT-4o"
• "Set my Anthropic API key in llm"
• "Show my recent llm prompts"
• "Embed these docs for similarity search"

Installation

pipx install llm
# or: pip install llm / brew install llm / uv tool install llm

Examples

# Pipe a file through a system prompt
cat app.py | supercli llm prompt run -s "Find bugs" -m gpt-4o

# Extract structured data
supercli llm prompt run "Who wrote Hamlet?" --schema "name, year:int"

# Save and reuse a template
supercli llm --system "Summarize as bullet points" --save bullets
cat doc.md | supercli llm -t bullets

# Set keys and extend with plugins
supercli llm keys manage set anthropic
supercli llm _ _ install llm-gemini llm-ollama

# Embeddings
supercli llm embed run docs README.md -d docs.sqlite3
supercli llm _ _ similar docs -d docs.sqlite3 -c "how to install"

Key Features

• Model plugins: llm-anthropic, llm-gemini, llm-ollama, llm-mlx, and more
• --schema extracts structured JSON from unstructured text
• -t/--template reusable system+model prompts, llm chat for multi-turn
• Every call is logged to SQLite — query with llm logs or Datasette
• llm embed/llm similar/llm embed-multi build local embedding search
• Fragments (-f) attach files, URLs, and previous logs to a prompt