Player not loading? Watch on YouTube
This walkthrough covers OpenHuman's desktop installation and compares its approach with OpenClaw and Hermes. The speaker installs the Mac app from a .dmg, drags it into Applications, and configures permissions for computer control and voice input. He also describes Homebrew installation, Linux packages and a Windows MSI installer.
OpenHuman keeps memory on the user's machine, but the speaker explains that its default Simple mode sends model calls, search and integration authentication through a hosted backend. Custom mode accepts the user's API keys and service settings. A fully self-hosted configuration takes additional work; installing the desktop app alone does not make it offline.
The AI agent's memory consists of editable Markdown files, with summaries by source, topic and day. The walkthrough shows a vault shortcut and an option to open it in Obsidian. The speaker also describes scheduled tasks, trigger filtering and a background loop that checks standing tasks every five minutes, with approval required for actions such as sending messages.
The review covers early beta version 0.5.6. Its meeting feature is described as Google Meet only, and the speaker says he has not personally tested it. He reports Linux AppImage launch failures, partial MCP support, limited OneDrive search and crashes on some local model embedding requests. His preference for OpenHuman centers on easier desktop setup, while Hermes suits his account of an agent that keeps running on a server.
