The College of Letters & Science IT (CIT) offers access to LLMs through an OpenAI-compatible API as well as the supported chat interface. API access is especially useful for more complex workflows and for using your own agentic tools.
API access is ideal for:
- Setting up your own tools
- Running workflows from your own devices
- Agentic workflows and coding agents
- Building your own RAG pipelines or other applications on top of the models
API access does not provide:
- The ability to add or run your own models
- Direct access to GPU/CUDA resources
- Saved conversation history. The API does not keep a record of your conversations for you; use is ephemeral unless you configure your own tools to record it.
CIT doesn't support the external tools you connect, but we do support the most complex and costly part: API-level access to the models. Reasons to use CIT's API include:
- Open-weight models. The models are publicly released, so your results are more reproducible and your workflows are less dependent on a single vendor's pricing or decisions. (Each model is governed by its own license.)
- Data and ideas stay at the institution. Data sovereignty is an ongoing concern for researchers using AI. An agreement not to train on your data is not the same as an agreement not to use it at all, and intellectual property is difficult to retract once shared. With CIT's API, inference runs on UCSB hardware.
- Cost. When you're starting out with AI, mistakes can be expensive. A bad prompt or a runaway loop can generate millions of tokens in minutes, which could mean a large bill from a commercial vendor. We track token usage to keep access equitable, but we don't charge for usage or set per-user spend allowances. Controls may be applied to prevent any one user from monopolizing the shared service.
- Flexibility. With API access you control the prompts, settings, data, and tools. We provide the model; you control everything else.
Output from these models can be inaccurate, biased, or inappropriate, and you are responsible for how you use it. See Model Safety and Responsible Use.
Documentation will always lag behind updates. To get the current list of available models, log in to https://api.ai.college.ucsb.edu/ with your API key, or run:
curl https://api.ai.college.ucsb.edu/v1/models \
-H "Authorization: Bearer sk-YOURAPIKEY"
Replace sk-YOURAPIKEY with your actual API key.
See the Python Quickstart and cURL & REST Quickstart for working examples.