Chat completions

const response = await client.inference.chat({
  model: "openai:gpt-4o-mini",
  messages: [
    { role: "system", content: "You are a helpful assistant." },
    { role: "user", content: "Explain x402 payments." },
  ],
  temperature: 0.7,
  max_tokens: 1024,
});

const text = response.choices[0]?.message.content;

Request fields

Field	Description
`model`	Model ID (`provider:model`)
`messages`	Chat messages (`system`, `user`, `assistant`, `tool`)
`temperature`	Optional sampling temperature
`max_tokens`	Optional max completion tokens
`tools`	Optional tool definitions (see Tool calls)
`storeConversation`	Optional — persist the turn (requires auth)
`conversationId`	Optional — continue a managed conversation

Response

OpenAI-style choices, usage, plus optional Galliun fields:

Field	Description
`usage`	Token counts
`cost`	USDC cost breakdown
`payment`	Payment proof echoed on success
`conversationId`	Set when using managed conversations

Client setup: Create a client.

Streaming

Server-side stream: true is not supported as an SSE client flow in the current SDK. Prefer client.inference.chat for complete JSON responses.