When you're building LLM-powered tooling, the last thing you want is to burn API credits debugging your plumbing. Simon Willison's llm-echo plugin solves this cleanly: it's a "model" that doesn't call any LLM. It just echoes your prompt back. That's it.
Debug Your LLM Pipelines Without Burning Tokens
Simon Willison just dropped llm-echo 0.5, a dead-simple debug plugin that echoes prompts back instead of hitting a real model — now with accurate token accounting for tool calls, system prompts, and reasoning tokens.
This is a Pro article
Subscribe to read the full article and support independent AI journalism.