Nous Research Launches Hermes Desktop App for Self-Improving AI Agent
The desktop app removes the terminal barrier for Hermes Agent, making a self-learning AI that saves reusable skills accessible to non-developers on mainstream operating systems.
Key Facts
- Nous Research released a public preview of Hermes Desktop on macOS 12 or later and Windows 10/11, with Linux still requiring a terminal.
- Hermes Agent saves discovered methods as reusable skills without re-explanation.
- The app includes persistent memory, scheduling, web browsing, image generation, sub-agents, and support for local, Docker, and SSH backends.
- Hermes Agent is accessible via Telegram, Discord, Slack, WhatsApp, Signal, email, and CLI.
- Pricing includes a free tier and three paid plans (Plus, Super, Ultra) with monthly credits for over 300 models and tools.
Reporting from 1 source: GameBusiness.jp.
Nous Research has released a public preview of Hermes Desktop, a GUI-based desktop app for its open-source autonomous AI agent Hermes Agent. Previously requiring terminal commands, the app allows setup and operation on macOS and Windows via native installers, with features including persistent memory, scheduling, web browsing, image generation, and multi-platform access.
Nous Research has released a public preview of Hermes Desktop, a native GUI application for its open-source autonomous AI agent Hermes Agent. The app provides installers for macOS 12 or later and Windows 10/11, with Linux installation still requiring a terminal. Users can set up and operate the agent entirely through the graphical interface, eliminating the previous need for command-line interaction.
Hermes Agent is designed to handle multi-step tasks such as web searches and file operations, and it learns from its own actions: when it discovers a new method, it saves the procedure as a skill for reuse without re-explanation. The desktop app includes persistent memory for recording projects and solutions, scheduling for repetitive tasks in natural language, web browsing, browser operations, image generation, sub-agents for parallel processing, and support for multiple execution backends including local, Docker, and SSH. The same agent can also be accessed from Telegram, Discord, Slack, WhatsApp, Signal, email, and CLI.
The software is licensed under MIT, with source code available on GitHub. Pricing includes a free tier and three paid plans-Plus, Super, and Ultra-each offering monthly credits usable with over 300 models and tools such as web search and image generation.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.