Your data stays yours
Models, chats and documents live on your own hardware. Your prompts stay on your servers and are never used to train any model.
- Works fully offline
- Encrypted chat caches
- Role-based access
- Never used for training
UpupaBitUpupaBitA self-hosted inference engine and a private AI chat workspace that run large language models on your own hardware.
Get startedRead the docs
Everything you expect from a hosted AI service, running on machines you control.
Models, chats and documents live on your own hardware. Your prompts stay on your servers and are never used to train any model.
Quantized kernels make good use of a plain CPU, with an NVIDIA GPU when you have one.
Use hardware you already own. Costs stay flat as usage grows.
Roles, shared channels and a knowledge base for the whole team.
Start on one machine, then add nodes with load balancing and failover.
Written from scratch in Python, so every step is easy to follow, check and change.
GGUF models from Hugging Face, Ollama or Matricxon engines, and MCP tool servers.