Developer Simon Willison has released version 0.5 of llm-echo, a specialized debug plugin designed for his llm command-line interface. The tool serves as a mock model that does not actually communicate with a large language model. Instead, it repeats the user’s input back to them, allowing developers to test the surrounding software infrastructure without incurring API costs or waiting for network latency. The update was prompted by testing requirements for an upcoming release of the core llm software.
Functional Updates and Significance
The primary advancement in this release is the improvement of simulated token usage reporting. While previous versions provided basic feedback, llm-echo 0.5 now generates usage metadata that accounts for system prompts, tool calls, and reasoning tokens. This functionality is critical for developers who need to ensure their applications correctly handle the complex metadata returned by modern AI models. By simulating these specific categories, the plugin allows for the verification of user interface elements and logging systems that track how different types of tokens are consumed during a session.
Unknowns and Future Development
While the release notes confirm the new simulation capabilities, several technical details remain unaddressed. The announcement did not specify the exact logic used to calculate these simulated counts or whether they are intended to mimic any specific provider’s tokenization method. Furthermore, the announcement did not specify a release date for the upcoming version of the main llm tool that necessitated these changes. It remains unclear if future versions of the echo plugin will support more complex simulated behaviors, such as mock error states or multi-turn conversation histories.
The llm-echo 0.5 update is currently available for users of the llm utility. It provides a cost-effective way to refine command-line AI workflows during the development phase.
Lead image: Rafael Minguet Delgado / Pexels
