Today’s release of Janas brings Brebus, its app (Linux for now, Windows next): the models of this machine, a chat with them and a server for other programs, in a window of its own. No cloud, no account, open source (GPL-3.0).
Models from Hugging Face, two ways. The 18 JNS models of this organization - Qwen3, Qwen3.5, Qwen3.6, Qwen3.8, Qwen3-Coder, Gemma 4, Nemotron 3.5 Bolt - download from Brebus’s catalog, each checked against its SHA-256, with what it asks of your memory shown first. Any other model is searched on the Hub by name: Brebus lists the repositories with GGUF files, each file with its size, and converts the one you pick on your machine.
An engine that adapts to the machine. Janas-LLM measures the memory, the CPU, the GPU and the NPU when it opens a model and chooses how to run it: the whole model in memory when it fits; when a mixture of experts is larger than the RAM, the experts each token needs read from the NVMe SSD and the frequent ones kept in memory; a GPU and Intel’s NPU used where they are faster. On a laptop with 32 GB (Core Ultra 9 185H), Qwen3-Next-80B, 48 GB, writes at 26.4 tok/s, where llama.cpp manages 3.0, under the same conditions wherever possible.
In the chat:
In the picture: Qwen3-30B-A3B in Brebus made a mortgage calculator as a React artifact; 250,000 at 3.5% over 25 years gives 1,251.56 a month, as it should.*
https://github.com/prabanta-dev/janas · release: https://github.com/prabanta-dev/janas/releases/tag/v2026.11.16