llm-echo 0.5
Simon Willison’s Weblog Simon Willison
llm-echo 0.5 now tracks tool calls, system prompts, and reasoning tokens. It’s a fake model for debugging, but the counts are getting more real.
Based on reporting by Simon Willison’s Weblog, Simon Willison — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Simon Willison has updated llm-echo 0.5, the debug plugin that stands in for a model by simply echoing back whatever prompt it gets. There’s no real LLM behind it. That is the point.
The new version changes how simulated token usage is counted. Tool calls, system prompts, and reasoning tokens are now part of the tally, instead of being left out of the pretend math.
Willison said he revisited the plugin while working on a new LLM release and testing pull requests. That’s a pretty good reminder that the boring little tools often get sharpened when the bigger machinery is under active development.
An example shows the plugin being run with uvx and llm, asking for output from the echo model on the prompt “hi there” with thinking turned on. The result includes the echoed prompt data and a usage line reading 2 input and 24 output tokens.
So yes, this is a debug plugin that imitates a model by not being one. But it now imitates the accounting more faithfully, which is exactly the sort of detail that matters when you’re trying to understand what an LLM stack is actually doing.
My take — AI-written commentary, not fact-checked reporting
The useful part of AI tooling is increasingly the plumbing, not the poetry. A fake model that keeps better books is more honest than half the glossy demos floating around. That’s the real trend: fewer magic tricks, more receipts.
Read more about this at: Simon Willison’s Weblog
