Text Generation

Not yet available. The Agent SDK is tracked to be built — today, use the Client SDKs against the live chat completions API. The docs below describe the intended design.

Text generation is the foundation of every agent turn. callModel() sends a message list to the model and returns the generated text in result.output. This page covers the generation-specific knobs: temperature, token limits, and practical patterns for consistent output.

A minimal generation call#

const result = await client.callModel({
  model: "anthropic/claude-opus-4.8",
  messages: [
    { role: "system", content: "You are a concise technical writer." },
    { role: "user", content: "Explain what a hypertable is in two sentences." },
  ],
});

console.log(result.output);

Temperature#

temperature (0–2) controls randomness. Lower values produce more deterministic output; higher values introduce more variation.

const result = await client.callModel({
  model: "google/gemini-2.5-pro",
  messages: [{ role: "user", content: "List three names for an AI gateway." }],
  temperature: 1.1,
});

Token limits#

Set maxTokens to cap the length of the response. Useful to control cost on bulk generation tasks.

const result = await client.callModel({
  model: "meta-llama/llama-4-maverick",
  messages: [{ role: "user", content: "Summarise this document:" + longDoc }],
  maxTokens: 300,
});

If the model hits the token limit before finishing, result.stopReason will be "max_tokens". The output is still returned as-is — truncated but usable.

Multi-turn text generation#

For multi-turn conversations without tool calls, pass the previous turn's item list as inputItems:

let items = [];

const r1 = await client.callModel({
  model: "anthropic/claude-sonnet-5",
  messages: [{ role: "user", content: "Draft an intro paragraph for a docs page about streaming." }],
});
items = r1.items;

const r2 = await client.callModel({
  model: "anthropic/claude-sonnet-5",
  inputItems: items,
  messages: [{ role: "user", content: "Now make it shorter and more direct." }],
});

Cost awareness#

Each TurnResult includes usage.costMicroDA — the exact DA cost of that turn in micro-DA (bigint). 1 DA = 1,000,000 micro-DA. Log it to build per-task cost tracking.

const daCost = Number(result.usage.costMicroDA) / 1_000_000;
console.log(`This turn cost: ${daCost.toFixed(4)} DA`);

Also see#

• Streaming — stream tokens as they arrive.
• Message Formats — multimodal inputs and content types.
• Dynamic Parameters — vary temperature or model per turn.
• API Reference — full parameter reference.