llm-echo 0.5

What it claims

llm-echo 0.5 is a debug plugin for Simon Willison's LLM CLI, released 18 October 2026. It provides an echo model that never calls an LLM and returns the prompt it received. Version 0.5 changes the simulated token usage counts so that tool calls, system prompts and reasoning tokens are counted. Willison revisited the plugin while testing pull requests ahead of a new LLM release.

How to use it

The plugin runs without a permanent install by using uvx to add it to a one-off LLM environment:

uvx --with llm-echo llm -m echo 'hi there' -o thinking 1 --usage

The -o thinking 1 option makes the echo model emit a canned reasoning line: "First I consider the prompt, then I decide what to say." The --usage flag prints the simulated token counts. The example run returns this:

First I consider the prompt, then I decide what to say.
{
  "prompt": "hi there",
  "system": "",
  "attachments": [],
  "stream": true,
  "previous": []
}
Token usage: 2 input, 24 output

The JSON shows what LLM passes to a model plugin. It holds the prompt text, the system prompt, any attachments, whether streaming is on, and the earlier turns of the conversation (previous). This makes the plugin useful for checking how LLM assembles a request, for example whether a system prompt or conversation history reaches the model, without spending on API calls.

Limits

The page does not explain how the simulated counts are calculated. In the example, 2 input and 24 output tokens are returned for a two-word prompt with thinking on, so the numbers are not real tokenizer output and should not be used to estimate costs. The post gives no changelog beyond the token-count change, no licence and no repository link. It also does not show how tool calls appear in the echoed JSON.