Fully Private
Conversations and model weights stay on your hardware. Zero cloud LLM APIs for inference.
Network boundary
Traffic flows from your device through the Tailscale mesh to your cluster. The cloud node is deliberately disconnected — no edge exists because no data flows there.
What stays on your network
Conversations
Every chat message stays local, stored in Electron SQLite. No sync to any cloud.
Model Weights
Downloaded once, stored on your SSD, never uploaded. You own the bits.
Metrics
CPU, GPU, RAM, power metrics stay on your cluster for local ops — not sold or streamed to ad networks.
Configuration
Cluster config, model preferences, deployment history. All local files.
Voice Data
Speech-to-text runs locally on device. Audio never leaves your network.
Search History
No query logging to any external service. Your questions are yours alone.
How auth works
Cluster access is mesh membership. The marketing site is separate.
On your private cluster, Tailscale/WireGuard identity is the auth boundary — devices on the tailnet reach asmi and local services without cloud LLM APIs. This public website (download / beta) uses Supabase Auth (Apple, Google, or email) only for account and release access — not as the cluster login.
Any device on your Tailscale network gets full access. Add a new Mac to the cluster, install Tailscale, and it is immediately part of the mesh — no config files, no firewall rules, no port forwarding.
r1o vs Cloud AI
| Feature | r1o | Cloud AI |
|---|---|---|
| Data location | Your hardware | Their servers |
| API keys needed | None | Required |
| Inference telemetry | Stays on your cluster | Vendor-side logs |
| Vendor lock-in | None | Full |
| Conversation history | Local SQLite | Their database |
| Model weights | Your SSD | Their GPU |
| Cluster access | Mesh membership | Email + password |
| Cost | Hardware only | Per-token billing |
Zero
Cloud LLM APIs
On-prem
Inference data
WireGuard
Encryption
Mesh-based
Access model
Local SQLite
Storage