Use OpenRouter with Simon Willison’s LLM CLI: One setup, many models

Simon Willison’s LLM command‑line tool now works with OpenRouter, giving developers a single CLI to try dozens of frontier models without rewriting code.

Reference: Willison’s announcement and setup notes: llm + OpenRouter.

Why this matters

• One interface, many models: Swap Anthropic, OpenAI, Google, Meta, and open‑weights via a single flag.
• Faster iteration: Compare outputs and costs model‑to‑model without new SDKs.
• Portability: Keep your prompt workflow stable while the model landscape shifts.

Quick start (5 minutes)

• Install LLM (recommended via pipx): pipx install llm
• Add the OpenRouter plugin: llm install llm-openrouter (or pip install llm-openrouter)
• Set your API key (either works): llm keys set openrouter or set OPENROUTER_API_KEY
• See available models: llm models (look for entries prefixed with openrouter/)
• Run a prompt: llm -m openrouter/<provider>/<model-slug> "Summarize the key risks and mitigations."

Run your first comparisons

• Anthropic example: llm -m openrouter/anthropic/claude-3-5-sonnet "Draft a 120-word product update in a confident tone." -o temperature 0.4
• OpenAI example: llm -m openrouter/openai/gpt-4o-mini "Draft the same product update." -o temperature 0.4
• Google example: llm -m openrouter/google/gemini-1.5-flash "Create 5 alt-text options for this product image."
• Stream tokens as they arrive: add -s
• Provide a system style: --system "You are a precise, terse technical writer."
• Prompt from a file: llm -m openrouter/<provider>/<model> -f brief.txt

Tips for teams

• Version discipline: Pin exact model slugs and note date tested—frontier models change frequently.
• Cost control: Use smaller “mini/flash” variants for drafts; save premium models for the final pass.
• Reproducibility: Log prompts, model slug, and options (e.g., temperature) alongside outputs.
• Safety posture: Treat anything you send as leaving your boundary; avoid secrets in prompts.

Caveats

• Data flows through OpenRouter; review docs and model‑specific terms.
• Availability and latency vary by model/provider. Build timeouts and retries.
• Pricing differs per model—check OpenRouter’s pricing.

Learn more: the LLM project on GitHub (simonw/llm) and the OpenRouter plugin (simonw/llm-openrouter).

Takeaway: If you already use LLM, adding OpenRouter gives you fast access to a broad model marketplace with minimal changes to your workflow.

Get weekly, no‑fluff AI tips in your inbox—subscribe to The AI Nuggets newsletter.