- Published on
Llm-echo 0.5
- Authors
- Name
- DeepHorus
The llm-echo debug plugin has been updated to version 0.5, introducing more accurate simulated token counts. Announced on October 18, 2026, the update now accounts for tool calls, system prompts, and reasoning tokens. This matters because developers use the plugin to test the LLM framework without invoking a real language model.
The llm-echo plugin provides a "model" that does not involve an LLM. Instead, it simply repeats back the prompts fed to it. This makes it a useful tool for debugging and testing. The update comes as the developer works toward a new LLM release, with testing pull requests prompting a revisit of the plugin.
Key changes include the simulated token usage calculations. Previously, these counts may have overlooked certain inputs. Now, tool calls, system prompts, and reasoning tokens are included. An example command illustrates the feature: uvx --with llm-echo llm -m echo 'hi there' -o thinking 1 --usage.
The example output shows the plugin's response. It first prints "First I consider the prompt, then I decide what to say." Then it outputs a JSON object with fields such as prompt, system, attachments, stream, and previous. It reports token usage: 2 input tokens and 24 output tokens.
The implications for developers are significant. More accurate token counts allow for better cost estimation and performance tuning. This is useful when testing applications that rely on LLM interactions. The update ensures that even without a real model, the simulated usage reflects realistic scenarios.
The update is part of ongoing development. The developer noted that testing PRs led to revisiting the plugin. Recent articles on the blog cover other topics, including Python version changes and a new blog feature. These indicate a broader context of active development.
Further refinements to llm-echo are likely as the new LLM release approaches. The plugin will continue to serve as a lightweight debugging tool. Its enhanced token simulation will remain valuable for developers seeking to optimize their applications.
