ExpressAI

ExpressAI is the SDK module for encrypted AI chat completions. It lets a host application send OpenAI-shaped chat conversations to the ExpressAI service and consume the answer as a live token stream, without the application having to deal with authentication, encryption, or transport details.

What the SDK does for you

• End-to-end encryption to an attested enclave. Prompts are encrypted on-device and only decrypted inside ExpressAI's secure enclave; response chunks arrive encrypted and are decrypted by the SDK. Before anything is sent, the SDK verifies the enclave's attestation against a root CA pinned in the SDK configuration. Intermediaries (including the API edge) never see plaintext conversations.
• Authentication. Requests are authenticated with short-lived signed request tokens derived from the user's subscription. The SDK creates and refreshes these automatically; the host only passes a subscriptionId.
• Censorship-resistant transport (native targets). ExpressAI calls ride the SDK's regular API stack, including Geo Unblocking (DSS & NT) — chat completions keep working in blocked regions with no extra handling. The wasm/browser build uses plain browser fetch instead (see Requirements below).
• Streaming ergonomics. startChatCompletion returns a pull-based ExpressAiChatStream: the host calls next() at its own pace, can drain everything in one call, and owns cancellation.

The first completion of a session performs the one-time enclave attestation and session setup, so it is a little slower; the session is cached and reused for subsequent completions.

Requirements

• License — the SDK license must include the ExpressAI category. sdkManager.expressAi() throws if the feature is not licensed.
• Entitlement — every call takes a subscriptionId that must hold the AI entitlement (see Chat Completions).
• Configuration — the express_ai configuration block must be set, see Enabling & Configuring.
• Platform — ExpressAI is available on all targets. On native it rides the censorship-resistant DSS/NT transport; in the wasm/browser build it uses the browser's fetch transport instead — fully functional and end-to-end encrypted, but with no censorship resistance (calls fail in blocked regions rather than falling back to DSS/NT).

Pages

Page	Content
Enabling & Configuring	The express_ai config block, domain override, programmatic setup
Chat Completions	Starting and consuming completions, chat credits, error handling

Built on ExpressAI

Some SDK features use ExpressAI internally and expose a finished object rather than a token stream — they need the same license and the same AI entitlement, but no chat handling on your side:

Feature	What it returns
explain_ranking_with_ai	A plain-language explanation of a speed-test ranking — read against the user's own line when the ranking was scored against one — or, for a single un-tunnelled measurement, of that connection alone