| name | getting-started |
| description | Get set up to build on the NVIDIA xr-ai stack (XR AI / DeviceIOHub). Use when a user asks to install, clone, or do initial setup of xr-ai; not for routine work in an already-set-up clone. |
Getting started with xr-ai
Pointers only; the linked docs hold the procedures. Paths are relative to the
repo root, so clone before following them.
First: two setup questions
Ask the user, then tailor the pointers below to the answers.
- Release or main? Build against the latest release tag (default) or
main. This should match the ref this skill was fetched from; re-fetch the
skill from the chosen ref if it does not. Releases that predate the skill
bank are not covered: with the user's OK, use main for both this skill
and the checkout. The answer selects both the git ref to check out and the
version of the docs site to read. 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
latest published release.
- 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: the LLM and VLM can run on hosted endpoints, and no
local GPU is needed for the agent or hub. STT and TTS stay local. Read
docs/source/components/ai-services.md (section "Hosting models on
NVIDIA NIM") and docs/source/getting_started/credentials.md.
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, matching what the docs site
serves).
- Working contract: read
AGENTS.md first. Every change must satisfy it,
and its canonical-references table routes deeper topics.
- Docs:
docs/source/; published versioned site at
https://nvidia.github.io/xr-ai/.
- Reference sample:
agent-samples/simple-vlm-example/, the one to study
and copy.
- Scaffolding a new sample:
docs/source/guides/adding-a-sample.md, full
boilerplate templates.
- Quickstart order: follow
https://nvidia.github.io/xr-ai/<ref>/getting_started/quickstart.html, with
<ref> replaced as above. Start agent-samples/model-servers and wait for
it to report readiness before starting agent-samples/simple-vlm-example
or another agent sample.
