openhuman: Local AI Orchestrator Syncing Every 20 Minutes

While most AI agents take weeks to set up initially or rely on external APIs, OpenHuman builds personalized context in minutes simply by connecting accounts. It automatically collects data from over 100 OAuth services (including Gmail, Notion, and GitHub) and over 5,000 MCP servers every 20 minutes, storing it in a local SQLite database. This data is compressed into a scored Markdown tree structure rather than a vector DB, and synced with an Obsidian vault that users can directly view and edit.

It controls agent swarms using graph-based orchestration rather than a single loop. Workflows running on the tinyagents framework can be paused and resumed via checkpoints, and blocked agents return root cause reports. Inter-agent communication uses end-to-end encryption from the Signal protocol, ensuring servers cannot see plaintext. It also supports an agent economy via x402 USDC rewards and 3-level sub-agent spawning.

It offers a visual workflow canvas similar to n8n or Zapier, but agents directly propose automation processes. Once the user approves, durable automations that respond to schedules, webhooks, and channel events are saved. It supports 17 messaging channels including Telegram, Slack, and WhatsApp, as well as native email (IMAP/SMTP), allowing agents to respond where users already work. When Privacy Mode is enabled, all inference is processed only within the local Rust core, ensuring no data leaves the device.

A single subscription includes Exa web search, Seedream image generation, and Veo video generation, with the option to route to local models like Ollama when needed. The TokenJuice feature compresses tool outputs before model input, reducing token costs by up to 80%. Installation and execution are possible entirely through the UI without terminal setup, and it can join Meet, Zoom, and Teams meetings to perform real-time transcription and speaking.