RESTai

RESTai is a self-hosted AI platform for developers and teams building document assistants, agents and AI services for other applications. Its web interface manages projects with separate models, prompts and tools, plus a chat playground for testing them.

• Document questions: Query uploaded files, rerank matching passages, sync content from web pages or S3, and ask plain-language questions of MySQL or PostgreSQL databases.
• Knowledge graphs: An optional graph for document projects links people, organizations and locations to source documents, with a visual explorer and questions scoped to matching entities.
• Agents and workflows: Agents can use MCP tools and a browser to fill forms or download files, while a Blockly editor connects projects into visual workflows.
• APIs and media: OpenAI-compatible endpoints provide chat, image generation and audio transcription; vision-capable models such as LLaVA and Gemini can accept images alongside text.
• Evaluation and team controls: Compare answer quality across saved prompt versions, track token use, cost and latency by project, and manage team access permissions.

RESTai is open source under Apache 2.0 and runs on your own server with Docker or Kubernetes. Local LLM backends include Ollama and vLLM. Connections to OpenAI, Anthropic, Gemini and Azure send model requests to the selected cloud provider. Image generation supports local and remote backends, including Stable Diffusion, Flux and DALL-E.

An optional MCP server exposes projects to Claude Desktop, Cursor and other MCP clients. RESTai detects NVIDIA GPUs and displays hardware details and utilization in its admin settings.