OpenHuman: desktop setup and memory walkthrough

OpenHuman is presented as a desktop AI agent centered on memory across connected tools. The speaker contrasts its unified interface with terminal-first agents such as Hermes Agent and OpenClaw, while still preferring Hermes for personal use.

The walkthrough describes a local-first memory tree with structured Markdown and SQLite storage. Users can read and edit memory through an Obsidian-style wiki, ingest files or folders, and build a summary tree. The speaker reports background syncing every 20 minutes and describes a Google Meet agent that can transcribe meetings into memory.

Setup covers native installers for macOS, Linux and Windows, a choice of local or cloud runtime, and custom model and voice providers. Settings > Advanced > AI configuration is the route shown for configuring a local LLM. A local runtime alone does not establish that connected model providers process data locally.

For the demonstration, the speaker requests a comparison report, a locally created PDF and delivery through Gmail. They switch to their own provider after encountering friction with the free tier and identify GPT 5.5 as the model used for the completed report. This is a hosted-model example, rather than evidence of offline execution. The speaker also criticizes monetization popups and advises caution when syncing personal data. They describe OpenHuman as partially open source under GPL3.