llm-echo 0.5 has been released. The plugin acts as a debug "model" that does not run a real large language model (LLM) but simply echoes back prompts it receives. While preparing a new LLM release and testing pull requests, the maintainer revisited the llm-echo plugin and updated its simulated token-usage accounting.
What the plugin does
llm-echo provides a model-like interface for testing and debugging: instead of invoking an actual LLM, it repeats the input prompts. This makes it useful for integration tests and development workflows where you want predictable, fast responses without downstream model costs.
Changes in version 0.5
The main change in the 0.5 release is that the simulated token counter now includes:
- tool calls,
- system prompts,
- and internal "reasoning" tokens.
Including these elements gives a more accurate simulation of token usage in scenarios where tool interactions, system-level instructions, or extra reasoning steps contribute to token counts.
Example usage
A representative command and its output shown in the release notes:
uvx --with llm-echo llm -m echo 'hi there' -o thinking 1 --usage
The plugin prints a short explanatory line, the echoed prompt structure, and a token-usage line. The example output is:
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
In this example the tool reports 2 input tokens and 24 output tokens.
Why it matters
Although llm-echo does not replace real models, the improved token accounting helps developers and testers estimate costs and quota consumption more realistically without making live model calls. That is useful for integration testing, automated checks, and rapid prototyping where token usage matters.
Tags
llm
