Skip to main content
The standard path is provider API + Railway worker + persistent memory. Your ngram’s Linux tools and files live on Railway. Its spatial body runs in your browser or headset.

​
Before you begin

Install Python 3.11+, Git, uv, and Node.js 22+. Have your provider API key and a Railway account ready. No local model or GPU is required.
1

Get ngram

2

Give it a cloud home

Name your Entity, select its personality, and provide an API key. The wizard verifies chat model availability and memory embeddings, then guides you through linking Railway. Review the worker, dedicated Postgres database, and persistent volume before deployment.OpenAI’s memory default is text-embedding-3-small with 768 dimensions. Provider API use and Railway hosting are billed by those services.
3

Enter your world

Run the AR command printed by setup. The Entity already lives on Railway; the browser connects to it through an authenticated bridge.Choose New ngram to walk through connection, memory settings, body, and voice. Saved worker connections appear automatically. A new body connected to an existing worker keeps that Entity’s identity and memories.
For a runtime on your own computer or a home GPU, see setup profiles. For Meta Quest and LAN details, see spatial setup.

How memory works

Persistent experience, semantic recall, and relationships.

Make it yours

Edit identity, voice, tools, and platforms.
If pairing fails, check the worker deployment and pairing token. If embeddings fail, check the provider’s model ID and vector dimensions. Keep existing memories on their original embedding model until you explicitly migrate them. See troubleshooting.