OpenHuman is an open-source, agentic AI assistant designed to seamlessly integrate into users’ daily workflows, offering a private, simple, and powerful experience. Its primary goal is to serve as a personal super-intelligence, providing context-aware assistance by connecting to a wide range of third-party services and maintaining persistent memory of user data. OpenHuman distinguishes itself with a clean, UI-first desktop experience, minimizing setup friction and enabling users to get started in just a few clicks without requiring terminal or configuration expertise. The agent features a desktop mascot that interacts with users, participates in meetings, remembers past interactions, and continues processing information in the background.
A standout feature of OpenHuman is its extensive integration ecosystem, supporting over 118 third-party services such as Gmail, Notion, GitHub, Slack, Stripe, Calendar, Drive, Linear, and Jira. Integrations are handled via one-click OAuth, and the agent auto-fetches data every twenty minutes, ensuring up-to-date context without manual intervention. All connected data is processed and stored locally, forming a hierarchical memory tree in SQLite and an Obsidian-compatible Markdown vault. This approach enables users to browse, edit, and manage their knowledge base directly, inspired by Karpathy’s Obsidian wiki workflow. The memory tree compresses and summarizes information, allowing the agent to rapidly learn about the user’s workflow and provide relevant assistance within minutes of setup.
OpenHuman includes a comprehensive set of built-in tools: web search, a web scraper, coding utilities (filesystem, git, lint, test, grep), and native voice capabilities (speech-to-text input, ElevenLabs text-to-speech output, mascot lip-sync, and live Google Meet participation). Model routing is integrated, automatically selecting the appropriate large language model (LLM) for each task, whether it requires reasoning, speed, or vision, all under a unified subscription. Optional local AI support via Ollama allows users to run workloads on their own devices, enhancing privacy and control.
To optimize performance and reduce costs, OpenHuman employs smart token compression (TokenJuice), which processes tool outputs, emails, and scraped content to minimize token usage before sending data to LLMs. This includes converting HTML to Markdown, shortening URLs, deduping verbose outputs, and preserving multi-byte text like CJK and emoji. The result is up to 80% reduction in cost and latency, while maintaining information fidelity.
Privacy and security are central to OpenHuman’s design. All workflow data remains on the user’s device, encrypted locally, and treated as personal property. Messaging channels are supported for inbound and outbound communication, allowing users to interact with the agent across their preferred platforms. The project is actively developed and currently in early beta, with a focus on minimizing vendor sprawl, keeping workflow knowledge on-device, and providing persistent memory beyond simple chat interactions.
OpenHuman is open-source under the GNU license, with contributions encouraged via a straightforward workflow. The repository provides detailed documentation, architecture guides, and setup instructions for new contributors. It also offers optional integration with agentmemory for users who self-host durable memory backends. Overall, OpenHuman aims to deliver a uniquely human-centric, context-rich AI assistant that learns quickly, respects privacy, and empowers users through deep integration and persistent knowledge.
