Atlas UI 3
Open-source chat interface from Sandia National Laboratories for self-hosted AI assistants with MCP, RAG and fine-grained access rules.
What it does
Atlas UI 3 is an open-source chat interface for language models that you run on your own servers. Through the LiteLLM library you connect models from OpenAI, Anthropic, Google, OpenRouter or your own endpoints. Assistants use MCP to reach tools and data sources, and a RAG connection pulls knowledge from document collections. Admins control access through group permissions, OIDC sign-in, approval of tool calls and data classifications that define which models and sources may handle which content. It also offers personas, chat history with retention rules and S3-compatible file storage.
Who it suits
IT teams in organisations with strict requirements that want to build an internal AI chat with clear access rules. Setting it up is hardly feasible without experience with Docker, reverse proxies and Python.
Data protection for businesses
You host Atlas UI 3 yourself, so you decide where servers and data are located, including in the EU. The vendor does not offer a data processing agreement because it runs no service of its own; you sign agreements with the respective model provider. Whether inputs are used for training also depends on that provider and its plan. There are no business plans; the software is MIT-licensed.
Limits
Not a ready-made cloud service and no guaranteed support. Documentation is available in English only. The documentation does not explicitly mention local model servers such as Ollama.