Player not loading? Watch on YouTube
OpenHuman is presented as a desktop AI agent that builds persistent context from connected accounts and conversations. The speaker describes an open source Rust app built with Tauri under the GPL-3 license, with installers for Mac, Windows and Linux. The comparison with Hermes Agent focuses on their different goals: Hermes learns task skills, while OpenHuman collects context about the user and their work.
The memory pipeline converts incoming material to Markdown, splits it into chunks of up to 3,000 tokens, and builds source, topic and daily summary trees. AutoFetch refreshes active connections every 20 minutes. A local database stores chunks, while an editable Markdown vault opens in Obsidian so users can inspect and correct the stored memory.
The setup walkthrough selects local operation and a custom model configuration. The dashboard includes manual sync, ingestion statistics and an entity graph. Examples include email digests, matching commits to tickets and preparing meeting context.
The speaker recommends Ollama for local AI processing and says the app includes a small model for background work. Claims that nothing leaves the computer are the speaker's assurances, not an independent privacy audit. Connected services still involve sensitive account access, and cloud model prompts can contain user data. The review also notes upgrade pop-ups in the beta and recommends trying a few accounts on a spare machine before expanding access.
