> ## Documentation Index
> Fetch the complete documentation index at: https://docs.zprotocol.org/llms.txt
> Use this file to discover all available pages before exploring further.

# z.infer() API

The Z Inference Gateway exposes its primary interface as an OpenAI-compatible REST endpoint at `/v1/chat/completions`. Tier 4 also uses `/v1/handshake` for ephemeral key exchange before the encrypted request is transmitted.

The `z.infer()` function is the equivalent interface available to smart contracts on the host EVM chain. Z Trade and Z Lend use this path to compose inference with on-chain actions. The REST interface is the primary path for application developers; the contract-callable path is the primary path for on-chain finance activities.

## Call Format for REST API

A standard request mirrors OpenAI's Chat Completions:

```http theme={null}
POST /v1/chat/completions
Authorization: Bearer <api_key_or_wallet_signed_credential>
Content-Type: application/json

{
  "model": "deepseek-v3:tee",
  "messages": [
    {"role": "user", "content": "Summarize this contract clause..."}
  ],
  "stream": true,
  "max_tokens": 1024
}
```

The response follows the OpenAI streaming SSE format. On Tiers 1 through 3 the response is plaintext and ordinary OpenAI client code consumes it without modification.

On Tier 4 the response stream is ciphertext; an additional `z_metadata` event is emitted after the final encrypted chunk, carrying a silicon-signed completion summary that the gateway reads to finalize billing. The Z TypeScript SDK handles both the decryption and the metadata verification transparently.

## Tier selection

Tier is selected by suffixing the model identifier (`deepseek-v3:tee`) or by setting the `X-Z-Tier` header. The two are equivalent and either is sufficient. See [Trust tiers](/docs/inference/trust-tiers) for the operational meaning of each tier.

## Privacy mode

For Tiers 1 through 3, the gateway is the trust boundary: prompts arrive at the gateway in plaintext, are routed to the selected upstream, and the response is relayed back. Z does not retain content on any tier as a matter of policy.

For Tier 4, the trust boundary is the TEE node itself. Before issuing the request, the client performs a handshake against `/v1/handshake`, receives a long-term TEE identity key and an ephemeral session public key, validates the attestation chain on the long-term key, and encrypts the prompt to the ephemeral session key under AES-256-GCM with a fresh random nonce. The gateway relays the ciphertext without inspection and holds no decryption key.

## On-chain invocation

`z.infer()` is callable from Solidity contracts. The call format from a contract is structurally similar to the REST call: a model identifier (optionally tier-suffixed), a payload, and a callback for receiving the response. The contract pays from its own balance and acts on the response when it returns.