On-premises large language model access for UCSB staff and faculty; incidental student use is supported. All inference runs on CIT-owned hardware, and your prompts stay on campus unless you enable tools such as web search that call external services.
The L&S AI services are built by CIT staff using open-weight models and open-source software. They are useful in their current configurations, but because LLM technology is developing so quickly, they should be considered beta services. Development is ongoing. We aim to keep services and model offerings stable within each academic quarter, but security or operational issues may cause short service interruptions at any time.
Responsible use: These open-weight models include built-in safety behavior, but no safeguard is complete. Output may be inaccurate, biased, or harmful. You are responsible for verifying output and for how you use it. See Model Safety and Responsible Use.
| Chat interface | chat.ai.college.ucsb.edu |
| API endpoint | api.ai.college.ucsb.edu/v1 |
| Sign in | UCSB NetID and Password |
| API keys | By request — cloud.college.ucsb.edu/contact (Other Requests) |
| Status | Beta |
New here? Start with the Getting Started Guide, the FAQ, and Known Limitations.
The following are the primary flagship models supported by the L&S AI Inference platform.
| Model | Best for | Context |
|---|---|---|
| gemma-4-26b-a4b-it | General chat, writing, brainstorming, and image input | 256k |
| qwen3.8-27b | STEM, research, long documents, and code | 256k |
| gpt-oss-120b | Complex reasoning, synthesis, and deep research | 128k |
Other models may be available through the API but are not covered in the flagship documentation. See API Access for how to list them.
Open WebUI User Guide: Model switching, document upload and RAG, web search, and system prompts.
Choosing the Right Model: Guidance on choosing between the flagship models.
Context Windows and Effective Prompting: How context limits work, strategies for long conversations, and tips for getting better results from open-weight models.
API Access: What the API is for and how to request a key.
Python Quickstart and cURL & REST Quickstart: Minimal working examples.
Pi Coding Agent: A lightweight, minimal coding agent for power users, focused on simplicity and an extension-driven architecture.
Configurations for aider, opencode, and Continue (VS Code) are also available.
Questions or issues? Contact help@cit.ucsb.edu.