Website • Hugging Face • LinkedIn • Discord
300k+ downloads of our models and their community builds • 100+ community quants and fine-tunes • 950+ likes on Hugging Face
UkisAI is an applied AI lab from Belgrade, Serbia, operating across Europe and the US. We post-train open models, publish the data and evals behind them, and build open-source tools for AI developers.
- Models. We post-train open models to reason faster and cost less to run. Our models on Hugging Face and the 100+ community quants and fine-tunes built on them have passed 300k downloads.
- Datasets. We release training data, like Qwen3.8-27B-multi-turn-agent-sft: about 15,200 multi-turn agent traces for terminal, coding, and software-engineering tasks.
- Developer tools. Open-source tooling for building AI products and doing model research from the terminal.
- Production AI. Products and custom systems for startups, enterprises, and public institutions.
vibe-ukis is our CLI for vibe coding AI products. It sets up Claude Code skills, Cursor/Windsurf/Antigravity rules, and MCP docs servers for your coding agent in one step.
pip install vibe-ukis
vibe-ukis start- Swift-Qwen3.8-27B-evals: every response, score, and config behind our Swift model's benchmarks
- ukis-research-helper: multi-agent research report writer on LlamaIndex
More is on the way. Follow the org to catch new releases first, or join our Discord and ask for access to UkisAI-EarlyAccess, where our community tries new tools before they go public.
Custom models, licensing, and production AI: ukisai.com/contact.
