Nous Research has released the Computer Use feature for Hermes Agent on macOS, directly challenging the capabilities of OpenAI's Codex. By integrating the open-source cua-driver, the agent reverse-engineers Apple's low-level APIs to perform background automation without interrupting user activity.
Background and Positioning
Traditional AI agents often require frequent user permissions or consume significant foreground resources when interacting with the operating system. Hermes Agent bypasses these limitations by driving system APIs directly, enabling complex tasks such as file management, application interaction, and browser control—all while the user continues working undisturbed. This design puts it head-to-head with OpenAI Codex, which aims to achieve similar desktop control via natural language.
Core Innovation: Four-Layer Context Compression
The standout feature is four-layer context compression, optimized for the massive token demands of high-resolution screen observations. At a 1568×900 resolution performing 20 actions, standard models typically consume around 600,000 tokens. Hermes Agent reduces this to just 30,000 tokens—a 95% reduction—by sequentially filtering redundant visual data, encoding actions structurally, caching cross-step states, and summarizing semantic snippets. This dramatically lowers API costs and accelerates response times, making long-duration automation far more practical.
Open-Source Ecosystem and Compatibility
Hermes Agent is fully open-source, built on top of Nous Research's Hermes language model series. The cua-driver is also released as an independent open-source module, allowing developers to extend support to other operating systems. Currently optimized for macOS, future versions may cover Windows and Linux. The agent integrates seamlessly with popular frameworks like LangChain and AutoGPT, lowering the barrier for developers.
Market Impact and Outlook
The launch comes at a time when the AI agent landscape is fiercely competitive. While OpenAI Codex excels at code generation, Hermes Agent differentiates itself through ultra-low token consumption and open-source flexibility. Analysts believe that drastically reduced costs will accelerate enterprise adoption of autonomous desktop automation, such as next-generation RPA and advanced personal assistants. Nous Research plans to further refine multimodal input processing and explore cross-device coordination in the coming months.