LLM-OpenRouter 0.7 adds reasoning traces and three server-side tools

👁 116 reads

LLM-OpenRouter Plugin Updated with Reasoning Traces and Server-Side Tools

On 21 August 2026, Simon Willison released llm-openrouter 0.7, bringing significant updates to the LLM plugin for OpenRouter-hosted models. This release introduces compatibility with LLM 0.32 and adds several new capabilities that enhance how users interact with language models through the OpenRouter platform.

OpenRouter is a service that provides unified access to multiple language models through a single interface, allowing developers and users to work with different models without managing separate API integrations. The llm-openrouter plugin connects the LLM command-line tool to these OpenRouter-hosted models, making it easier to query and interact with various language models from the terminal or within applications.

Understanding Reasoning Traces in Language Models

The most visible change in this release is reasoning-trace support. Because the plugin now targets LLM 0.32, it can surface the reasoning traces produced by models available through OpenRouter. Reasoning traces are step-by-step explanations that show how a language model arrives at its answer, revealing the intermediate thinking process rather than just the final output. This transparency helps users understand why a model gave a particular response and can be valuable for debugging, education, and building trust in model outputs.

The update also transitions models to OpenRouter’s own implementation of the Responses API, replacing the previous approach. This change represents a shift in how the plugin communicates with OpenRouter’s infrastructure, potentially improving reliability and feature support.

New Server-Side Tools for Enhanced Functionality

Three server-side tools arrive in this release: Shell, WebFetch, and WebSearch. Users enable them through option flags, with the announcement giving -T WebSearch as the example syntax. The tools run server-side, meaning the heavy lifting happens within OpenRouter’s infrastructure rather than on the client.

This server-side execution model offers several practical advantages. Users don’t need to install additional dependencies on their local machines, the tools can leverage OpenRouter’s more powerful infrastructure, and there are no concerns about local security when executing shell commands or fetching web content. For someone using the WebSearch tool, for example, the search queries and result processing happen entirely on OpenRouter’s servers, with only the final results returned to the user’s machine.

FAQ

What is llm-openrouter and who released version 0.7?

llm-openrouter is a plugin for the LLM tool that provides access to OpenRouter-hosted models. Simon Willison released version 0.7 on 21 August 2026, updating it with compatibility for LLM 0.32 and new features including reasoning-trace support and server-side tools.

What are the three server-side tools added in this release?

The three server-side tools added are Shell, WebFetch, and WebSearch. Users can enable them through option flags, such as -T WebSearch, and they execute on OpenRouter’s infrastructure rather than on the user’s local machine.

What does reasoning-trace support mean for users?

Reasoning-trace support allows the plugin to surface the reasoning traces produced by models available through OpenRouter. This means users can see the step-by-step thinking process that models use to arrive at their answers, providing greater transparency into how responses are generated.

// read moreResearchers extracted hidden reasoning traces from GPT and ClaudeNewsLLM 0.33 Adds Repeatable Templates, OpenAI Library 3.x SupportAI ToolsTaiwan Indicts Nine Over Scheme to Smuggle 74 Nvidia AI ServersNews

Kinetyra