Run uv run ngram setup to choose an inference route and create an Entity. The wizard distinguishes where the runtime lives from where inference happens.
Three setup profiles
Run only the command for the profile you want. Hosted is the shortest path when you already have a provider account. Local is useful when you want inference on your own hardware. Hybrid is an operational deployment choice.
- Your Entity name and configuration.
- The inference provider and model selection.
- Local environment variables and provider credentials.
- An embedding route for semantic memory.
- Advanced gateway and tunnel setup when you explicitly choose hybrid.
Follow the exact next-step command printed at the end. Keep .env local and restart an already running process after changing its environment.
Add a spatial body
For an existing local Entity, use manual spatial setup. For an integrated hosted-inference and Meta Quest lab workflow, use:
The lab launcher performs provider checks and supervises the Entity and WebXR processes. It exposes the HTTPS spatial surface to your trusted LAN while keeping the Entity bridge on loopback. See Quest lab for requirements and pairing.
Verify before adding another layer
- Confirm the Entity can reply in the terminal.
- Confirm embedding requests succeed.
- Inspect state paths with
uv run ngram status rook.
- Connect one additional surface.
- Verify stop behavior and persistence before leaving a daemon running.
Cloudflare is only needed when your chosen topology exposes a protected home gateway. It is not part of ordinary hosted or local first-run setup.
