Simon Willison's `llm` CLI (a long-standing dev staple for talking to hundreds of LLM

Simon Willison’s llm command-line tool just shipped a major update, and if you’re not a developer, you might have missed it. But it matters because this tool has quietly become the Swiss Army knife for anyone who needs to talk to AI models without wrestling with a dozen different APIs.

The llm CLI lets you query hundreds of LLM providers, OpenAI, Anthropic, Google, local models, from one simple interface. You type a command, you get an answer. No switching between web apps or juggling API keys. Developers love it because it’s fast and consistent. The new version adds reasoning-trace support (so you can see how models like o1 think through a problem), compatibility with OpenAI’s Responses API, and server-side tool calling, which means the model can trigger actions on your behalf without you writing glue code.

What’s interesting is how little noise this release made, according to Willison himself, despite the tool’s wide reach. That’s often a sign that something is genuinely useful rather than hyped. The people who need it are already using it.

Why it matters to you

If your team is experimenting with AI, they’re probably bouncing between ChatGPT, Claude, and whatever else is flavour of the month. That’s fine for exploration, but it doesn’t scale. A unified interface like llm means you can test prompts across models, script repetitive tasks, and build lightweight automations without waiting for a vendor to add a feature.

This is the kind of plumbing we build into the Omni Command Centre, a single layer that talks to multiple models and routes work based on what each one does best. You don’t need to be a developer to benefit, but you do need someone thinking about how your AI tools fit together. Willison’s update is a reminder that the best infrastructure often arrives quietly.