LLM Factory is a locally controlled platform for the latest open-weight large language models (LLMs), through an intuitive web interface or programmatically through an OpenAI-compatible API. It is the foundation for the team’s language-model research and applications, and users can securely fine-tune their own models on a platform controlled locally.
Current models include GLM-5.3-Flash and DeepSeek-V4.1-Flash, with a context of 1,048,576 tokens, and the team’s own KOS-V5 models. The API also serves embeddings, Whisper speech-to-text, and Kokoro text-to-speech. On October 22, 2026, it served 1.06 billion tokens across 11,760 requests in 24 hours.
Part of the Kentucky Open Science platform: it runs over the Cresco mesh, with versioned storage for datasets, models, and code in GaiaKeep. See the full stack.
LLM Factory uses parameter-efficient fine-tuning methods such as Low-Rank Adaptation (LoRA) to make model customization efficient and scalable. Users upload their own datasets, train adapters on a base model, privately configure and explore models, and expose them through the OpenAI-compatible API.
Adapters change only a small subset of parameters, so customization reaches performance comparable to full fine-tuning with far less memory and computation.
User data, chat history, and API interactions are stored locally. Because the endpoints are OpenAI compatible, libraries, tools, and systems built for that standard work with it unchanged.
Request access through the information request form. An informational video is available on YouTube.
Users of LLM Factory agree to acknowledge the tool in their research, publications, and products.
V. K. C. Bumgardner, M. A. Klusty, W. V. Logan, S. E. Armstrong, C. N. Leach, C. Hickey, J. Talbert. Institutional Platform for Secure Self-Service Large Language Model Exploration. AMIA Joint Summits on Translational Science Proceedings, 2025, pp. 105-114. PubMed