Why self-host
Self-host is for people who want full control of where their data lives and what the brain can see.
Reasons to self-host
Section titled “Reasons to self-host”- Private files and services. Point the brain at code, customer data, internal docs — none of it leaves your network unless you let it.
- Data residency. Keep the brain in a specific region, cloud, or environment.
- Air-gapped. With an offline licence, the box makes zero outbound calls to Gray’s hub.
- Custom model providers. Route the brain through Bedrock, Vertex, or a private gateway instead of Anthropic’s API.
- More control over what gets logged, retained, and stored.
What stays the same
Section titled “What stays the same”- The native app is still your front door.
- Voice still streams to OpenAI Realtime (unavoidable — Realtime is on-device-impossible). Switch to text-only if you can’t accept that.
- The product is the same product. You’re just running half the stack.
What changes
Section titled “What changes”- You bring your own OpenAI key (voice) and Claude access (brain).
- You’re responsible for the box, backups, and uptime.
- You pick the model provider (Anthropic API, Bedrock, Vertex, or a proxy).
Requirements
Section titled “Requirements”- Active paid Gray subscription (Plus or higher).
- A server: any Linux box for Docker, or an Ubuntu 22.04/24.04 VPS for the installer path.
- An OpenAI API key.
- Anthropic access (API key or logged-in Claude CLI).