Which model should I use?
See the Model Selection Guide for a full breakdown. Short version: gemma-4-26b-a4b-it for general chat and writing, qwen3.8-27b for STEM, long documents, and code, and gpt-oss-120b for complex reasoning.
Are these the same as ChatGPT / Claude / Gemini?
No. These are open-weight models running on UCSB hardware. They are capable and useful, but smaller than frontier commercial models. They work best with explicit, specific prompts. See Getting Better Results from Open-Weight Models.
Will the available models change?
Yes. This is an early access service and the model lineup will evolve — models will be added, updated, and retired as the platform matures. We aim to keep the lineup stable within each academic quarter and to communicate changes in advance where possible.
Can I request a specific model be added?
Yes — email help@cit.ucsb.edu with your request and use case. We can't guarantee availability, but we do take requests into account.
Why did the model refuse my request?
Models will occasionally decline requests they interpret as harmful, inappropriate, or outside their scope. If you believe a refusal was incorrect, try rephrasing your prompt more clearly. If the issue persists, contact help@cit.ucsb.edu. Note that built-in safety behavior is imperfect in both directions; see Model Safety and Responsible Use.
Does my data leave UCSB?
All inference runs on UCSB hardware. Your prompts and responses stay on campus — with one exception: when web search is enabled in Open WebUI, your query is sent to an external search engine. Disable web search for sensitive topics.
Can I use this for research involving human subjects data?
No. Do not submit FERPA-protected, HIPAA-covered, CUI, or P3/P4 data. If you're unsure whether your data is appropriate, contact help@cit.ucsb.edu before proceeding.
Can I use this for my thesis or dissertation?
Yes, for drafting, brainstorming, summarizing, and similar tasks — subject to your department's academic integrity policies. Check with your advisor or department if you're unsure what's permitted.
Is my data used to train models?
No. Your conversations are not used to train or fine-tune any models.
Can Letters and Science IT staff see my conversations?
Conversations may be accessible to platform administrators for troubleshooting and service operation purposes, consistent with UCSB's standard IT policies.
What file types can I upload?
PDF, DOCX, TXT, Markdown, and CSV are supported.
How large can uploaded files be?
Per-file size limits apply. If an upload fails, split the document or upload only the relevant sections.
Why isn't the model using my uploaded document?
By default, retrieval only injects the parts of a document that match your question. Try rephrasing using terminology from the document, or switch the file to full-context mode. See Document Upload and RAG in the Open WebUI User Guide.
Does document content leave UCSB?
Document content is processed and stored on UCSB infrastructure. It is not sent externally unless web search is enabled alongside a document query.
Is there a usage limit?
There are no hard per-user usage limits, but the service is shared infrastructure and controls may be applied to prevent any one user from monopolizing it. Please check with help@cit.ucsb.edu before running automated high-volume workloads.
Is there an SLA or uptime guarantee?
No. This is a beta service, not a production platform with guaranteed availability. Plan accordingly for any workflows that depend on it.
How do I report a bug or give feedback?
Email help@cit.ucsb.edu. Bug reports, feature requests, and general feedback are all welcome.
Where do I find out about outages or model changes?
Announcements appear as banners in the chat interface, and current issues are listed on the Known Limitations page.