llm-openrouter 0.7 Adds Reasoning Traces

Independent developer Simon Willison has released version 0.7 of llm-openrouter, a dedicated plugin designed to connect his command-line LLM tool with models hosted on the OpenRouter platform. This latest update introduces compatibility with LLM version 0.32, unlocking several advanced capabilities for developers who interact with large language models through the terminal.

A primary highlight of the 0.7 release is the ability to display reasoning traces for supported models. As modern artificial intelligence models increasingly rely on chain-of-thought processing to solve complex problems, practitioners can now inspect these internal reasoning steps directly within their terminal workflows. Additionally, the plugin has transitioned to using OpenRouter's official implementation of the Responses API, ensuring more robust and standardized data delivery for all API calls.

The update also introduces three new server-side tools that expand the utility of the command-line interface: Shell, WebFetch, and WebSearch. Users can activate these tools by passing specific options during execution, such as using the -T WebSearch flag to enable real-time web queries. These integrations allow developers to build more dynamic, agent-like workflows without leaving their command-line environment, making it easier to fetch live data or run local commands.

For practitioners, these updates bridge the gap between local command-line efficiency and the powerful features offered by hosted API providers. By exposing reasoning traces and enabling direct web search and shell execution, the plugin transforms the LLM CLI from a simple prompt-and-response interface into a highly capable environment for debugging and executing complex, multi-step AI tasks.