Resource request tool¶
Describe the research or teaching activity, the resources that would help and the barriers. You do not need a GPU-hour estimate or a budget. Unknown quantities can stay unknown.
Local by default. Preparing a brief does not submit it to the working group. Optional Luna drafting sends a reviewed public/synthetic brief for AI processing only with your consent. Keep student records and confidential research out of this public tool. A separate, optional submission step stores only your reviewed brief and the contact details you provide in the private pilot queue.
Prepare a local brief. Sending it for Luna drafting or working-group review requires a separate action.
Your local needs brief
Barriers:
Next steps to review
Download a copy for yourself, then review and submit below if you want the working group to see it.
Campus CPU/GPU route: Puma permits class groups. The published 150,000 CPU core-hours per PI/month is shared by research and classes; GPU jobs use the same pool. It is not a class reservation. Queues may delay jobs; instructors support students. Allocation, queue and support limits →
Develop a resource request
This local template turns your brief into a reviewable request and adds documented resource routes. It does not use an LLM or decide eligibility. Candidate routes need owner confirmation.
Optional Luna draft
Use GPT-6 Luna to rewrite your brief and the dated resource facts into a short first draft. Review the result yourself. The local template above is always available.
Limited pilot: five attempts per network each day. Your local template remains available.
Usage limits and data handling
Shared allowance: 20 attempts/day, 200/month, up to 1,000 output tokens each. Failed attempts count; there are no automatic retries. The application keeps allowance counters and rotating network hashes, not your brief or generated text. OpenAI processes the requested text under its API terms; store:false is requested, not a zero-retention guarantee.
Campus alternative: copy or download the public/synthetic prompt and paste it into Arizona AI Assistant ↗. Wolfgang Jentner at ORAI, via Greg Chism, reports that AI Verde will transition to this service with $200/month of Bedrock credit available; credit scope, eligibility and renewal terms need confirmation. Transition and model pricing →. U of A GenAI is another documented NetID route. The site does not connect to either campus service.
Submit a completed brief
A small pilot queue for Barney Maccabe and Greg Chism. Submission is optional and does not use AI. Use your own public/synthetic, non-sensitive brief; no student records, credentials, confidential research or attachments.
Storage, limits and what happens next
Your contact details, brief and review notes are stored privately on the working group’s existing Vercel hosting, outside university Microsoft 365. This is a limited pilot, not university sign-in or an approved channel for sensitive data. Contact addresses are self-reported. Only reviewers with the shared passphrase can access the queue. They may export it or delete requests; there is no automatic deletion schedule. Contact a point person with your reference to request removal. Submission does not send your contact details or brief to Luna/OpenAI, notify reviewers by email, submit an ACCESS application or guarantee support.
Limits: three submissions per network/day, 20 shared/day, 100/month and 100 stored requests. A full queue pauses submissions. There are no automatic retries; keep your download. Reviewers can open the private queue ↗.
If technical sizing would help¶
Use the optional compute estimator for CPU/GPU, concurrency and storage assumptions. It is one attachment to the need, not a requirement for participating. Measured demand and provider access still need confirmation.