Simon Willison just shipped a new release of llm, his terminal-first interface for large language models. If you build with AI, this is one of the cleanest ways to make prompts reproducible, scriptable, and auditable—right from your shell. Read the release post here: New release of llm.
What is llm?
llm is a command‑line tool that routes prompts to models from major providers (e.g., OpenAI, Anthropic) and local runtimes via plugins, then stores prompts and outputs in a local database for search, reuse, and auditing. It’s open source and built with developer ergonomics in mind. Docs: llm.datasette.io, GitHub: simonw/llm.
Why this release matters now
- Reproducibility: prompts and outputs are logged locally, so you can diff, review, and roll back.
- Automation: pipe files and commands into models to build CI jobs, Makefile targets, and cron tasks.
- Privacy choice: use cloud APIs or run local models via plugins when data must stay on-device.
- Team workflows: share saved prompts and conventions so everyone ships consistent results.
Quickstart (3 minutes)
- Install: pipx install llm (pipx keeps tools isolated and easy to upgrade).
- Set API keys (examples): export OPENAI_API_KEY=… and/or export ANTHROPIC_API_KEY=….
- List available models and aliases: llm models.
- Send your first prompt: for example, echo “Summarize this README in one paragraph” | llm -m gpt-4o (pick any configured model).
- Explore help and options: llm –help. For the full command surface, see the official docs.
High‑impact workflows you can ship today
- Documentation pipelines: pipe markdown or code comments into llm to generate summaries, release notes, or API docs as part of CI.
- Structured outputs: request JSON responses and feed them to linters or validators before merging PRs.
- Human‑in‑the‑loop edits: generate first drafts (specs, SQL, unit tests), review diffs locally, and commit with traceable provenance.
Plugins and local models
llm supports a growing plugin ecosystem—including options to connect to local runtimes—so you can prototype offline or keep sensitive data on your machine. Browse official plugins and setup notes here: llm.datasette.io.
Release notes and deeper reading
For what’s new, performance tweaks, and UX improvements, see the announcement: New release of llm. For day‑to‑day usage patterns, consult the official documentation and GitHub repository.
Takeaway
If you’re serious about operationalizing AI, put it in the terminal. llm turns ad‑hoc prompting into versioned, testable, automatable workflows—so your team ships faster with less guesswork.
Get more practical AI tips in your inbox. Subscribe to our free newsletter: theainuggets.com/newsletter.
