Set up an XR AI checkout

Prepare a checkout and its prerequisites for development. The linked docs hold the procedures. Paths are relative to the repo root, so clone before following them.

First: two setup questions

Use choices the user already supplied, ask for any missing choice, then tailor the pointers below to the answers.

1. Release or main? Build against the latest release tag (default) or main. This selects <ref> for the skill, checkout, and documentation. Before giving setup steps, require a copy of this skill from <ref>. If this copy's source ref differs or is unknown, retrieve it from https://raw.githubusercontent.com/NVIDIA/xr-ai/<ref>/skills/setup/SKILL.md and continue with that copy. If the file is absent at a release tag, explain that setup is unavailable at that release and ask before using main for the skill, checkout, and docs. For other retrieval failures, stop and ask whether to retry. Use https://nvidia.github.io/xr-ai/main/ for main, or replace main with the selected release tag. https://nvidia.github.io/xr-ai/ opens the newest complete documentation set and can fall back to main when the release lacks current entry points.
2. Local or remote GPU? Models on the local GPU, or a hosted/remote endpoint?
   • Local: read the matching version of the requirements page at https://nvidia.github.io/xr-ai/<ref>/getting_started/requirements.html, including the VRAM table. Replace <ref> with main or the selected release tag.
   • Remote/hosted: read the same versioned requirements page: the hub still needs an NVIDIA GPU with NVENC and NVDEC, even when inference is remote. Follow docs/source/guides/deploying-with-nim.md to configure and validate the application's required endpoints, including remote speech when selected. Read docs/source/getting_started/credentials.md for credential sources.

Where everything is

• Repo: git clone https://github.com/NVIDIA/xr-ai.git, then check out the ref chosen above (latest release tag, run in the clone: git tag --list 'v*' | python3 .github/scripts/select_latest_docs_release.py; it prefers stable releases over prereleases). Use the selected tag in documentation URLs; the site landing page can fall back to main. If the selector prints no tag, remain on main and tell the user that no release tag was found.
• Working contract: read AGENTS.md for shared engineering guidance, workflow routing, and canonical references.
• Docs: docs/source/; published versioned site at https://nvidia.github.io/xr-ai/.
• Reference sample: agent-samples/conversation-example/, the one to study and copy for new voice-and-vision applications. The separate run-demo workflow still uses the deprecated Simple VLM browser demo.
• Building your application: docs/source/guides/building-your-app.md, the consumer path for choosing SDK layers, copying the reference, and verifying without hardware.
• Preparing an upstream sample: docs/source/guides/adding-a-sample.md, full repository boilerplate templates.
• Prerequisites: check the selected environment against docs/source/getting_started/requirements.md and report missing requirements. Ask before privileged host changes, system package installation, daemon restarts, or reboots. Configure credentials using docs/source/getting_started/credentials.md. Use existing credential sources without printing secrets.
• Running a sample: the quickstart document is at https://nvidia.github.io/xr-ai/<ref>/getting_started/quickstart.html, with <ref> replaced as above.

Setup complete

Return the checkout path and ref, selected model environment, and any remaining prerequisite or credential blockers. Do not launch model servers or the sample.