A scientist debugging a simulation can have two jobs on the same afternoon. One is to get the computation to run. The other is to figure out why it failed last time. The Texas Advanced Computing Center at the University of Texas at Austin is building Sagebrush for work that includes the second job. In its November 18 announcement, the center said the National Science Foundation-funded project would run as a persistent set of AI services alongside the NSF’s Leadership-Class Computing Facility.
The distinction is in how a researcher asks for work. In a batch system, the researcher submits a job for the machine to schedule. With Sagebrush, a client could send a query to a hosted model. The system description says the service uses an OpenAI-compatible interface and describes plans for a browser chatbot and a Jupyter environment. The interface describes how software connects. TACC says the system is designed to serve foundation models using SambaNova hardware, and SambaNova’s announcement confirms that the center plans to expand its existing deployment.
One planned service would use computing-facility documentation and code examples to help researchers diagnose failures. TACC also describes a way for teams to supply their own document collections for retrieval-augmented generation, which gives the model relevant passages to work with when it answers. For a lab considering that workflow, a small trial could start with a dozen past errors and the manual passages that resolved them. Run the same questions with and without retrieval. Count how many answers point to the correct passage and propose a change that actually fixes the problem. A fluent diagnosis alone earns no credit in that test. TACC’s description of the planned retrieval services
For labs budgeting their time, access is the distinction that matters first. The press release says TACC will seek early users with specific projects while it builds a system for allocating access. The system page gives fall 2027 as a target for new allocations and names the Leadership-Class Computing Facility, ACCESS and the National AI Research Resource as the organizations TACC will work with on that process. A lab with a deadline this year should plan around resources it can already use and treat early participation as a separate opportunity.