Open Assistant
A self-hosted AI assistant that runs in Docker, connects to email and calendars, and uses local models through Ollama or vLLM or cloud LLM providers.
Open Assistant is a self-hosted personal AI assistant for people who want it to act on their email, calendar, files, and notes. It runs in a single Docker container with a browser interface. A coordinator assigns requests to specialist agents, so a conversation can lead to finding an invoice, scheduling a meeting, or researching a topic.
You can run models locally through Ollama or vLLM, or connect OpenRouter, Anthropic, or Groq. Local model operation can work offline; connected services and external tools need network access. The assistant runs on your infrastructure; cloud model providers and connected online services involve external systems. It's designed for one user, and it encrypts stored credentials.
Its connections cover Gmail and Outlook email, Google Calendar and Outlook Calendar, OneDrive and Nextcloud files, plus Notion and OneNote. Microsoft To Do handles tasks. WhatsApp and Slack provide two-way messaging alongside the web interface, with support for voice notes and images. Whisper transcribes voice messages, while Mistral OCR extracts text from PDFs.
You can create custom agents and control which tools each agent can use through a visual interface. Plugins and remote MCP servers extend the available tools. Scheduled tasks support recurring jobs and one-off requests, while generated pages, dashboards, and documents remain available as saved HTML artifacts.
For web work, it connects to Brave Search and Google News and uses Playwright for browser actions, including logged-in sessions with encrypted cookies.
The project uses the source-available BUSL-1.1 license.
