Noema Client
The Gateway to Private Intelligence.
The Noema Client is the flagship application, in beta now. Built to run open models directly on your own hardware, in your browser via WebGPU or natively through Ollama, it acts as a local-first firewall for your AI queries.
Local-First by Default
Proving privacy is better than promising it. Interacting with the Noema Client follows this flow:
- Local Intercept: Your prompt is evaluated on-device. The client decides whether it can be answered entirely on your own hardware.
- Local Execution: Whenever your machine can run the model, it does. The prompt generates no outbound request at all, and your browser's network tab proves it.
- Sovereign Routing: Only when a task outgrows your hardware, and only along a path you allow, is it routed to the decentralized GPU network through onion-routed circuits. No relay sees both your identity and your workload.
- Attested Execution: The remote node runs inside a hardware enclave (TEE) that must prove, cryptographically, that it is running the exact open model without tampering or logging before it ever receives your workload.
Architecture Flow
mermaid
sequenceDiagram
participant U as User
participant L as Local Client
participant N as Onion Routing Network
participant G as Sovereign GPU (TEE)
U->>L: Submits prompt
Note over L: Client checks local hardware first
L->>L: Runs locally when it can (nothing leaves)
L->>N: For heavy tasks, routes via Tor-like relays
N->>G: Delivers workload with no link to your identity
Note over G: Attests hardware, then executes in isolated VRAM
G->>N: Returns AI response
N->>L: Delivers response
L->>U: Displays final responseLocal execution is live in the beta today. Sovereign routing to the network ships with the Network phase.
Native Wallet Integration
Connect your wallet (MetaMask, Coinbase Wallet, and other EVM wallets) directly to the client. You pay programmatic micro-fees in $NOEMA on Base only for the decentralized compute you actually use. No subscriptions, no hidden tracking.