Firebird can now host and run open-source large language models directly on the cluster using Ollama, a tool that makes it straightforward to pull down a model (and Ollama has LOTS of models!) and start working with it. Anyone with a Firebird account can spin up a capable model and run it against the cluster’s GPUs. Models run entirely locally; nothing you send to the model ever leaves Firebird.

Options for interacting with such models are essentially endless and are similar to ways in which you might use existing AI tools (Claude, Gemini, ChatGPT, etc.), such as:

Working with sensitive or restricted text. Because everything stays on the cluster, you can work with material you would never paste into an outside service, such as unpublished manuscripts, licensed corpora, or data covered by a use agreement. Summarizing, classifying, or coding that kind of text is now something you can do without the data leaving a controlled environment.

Large-scale text processing. It is possible to use models as part of batch jobs with Slurm. Instead of processing documents one at a time or in small batches, you can queue up thousands and extract structured fields, label, translate, or summarize in bulk.

Teaching and AI literacy. A professor can easily give an entire class access to models. And because the models are open, students can look under the hood: compare different models, adjust how they run, and study how they actually behave rather than treating them as a black box.