#Structured output

Free-text answers are fine for chat, but when your agent feeds another program you want a typed, validated result, not prose to parse. Set outputSchema — a Zod object schema — and the run's final answer is forced through it: the SDK registers a final_answer tool with your schema as its parameters, instructs the model to call it exactly once when done, validates the arguments, and surfaces them as result.output.

#Usage

import { createLiteAgent } from "@lite-agent/sdk";
import { anthropic } from "@lite-agent/provider";
import { z } from "zod";

const agent = createLiteAgent({
  model: anthropic(),
  modelName: "claude-sonnet-4-6",
  workdir: process.cwd(),
  outputSchema: z.object({
    name: z.string(),
    deps: z.number(),
  }),
});

const result = await agent.send("Summarize package.json");
result.output; // { name: "…", deps: 42 } — validated against the schema

query() accepts the same option. Its generator resolves to the LiteAgentResult, so drive it manually if you need result.output:

import { query } from "@lite-agent/sdk";

const run = query({
  prompt: "Summarize package.json",
  model: anthropic(),
  cwd: process.cwd(),
  outputSchema: z.object({ name: z.string(), deps: z.number() }),
});

let result;
while (!(result = await run.next()).done) {
  // stream events as usual: result.value is an AgentEvent
}
result.value.output; // LiteAgentResult.output

#How it works

1. A final_answer tool is registered with your schema as its parameter schema, so the model can only produce structurally valid arguments.
2. A ## Final answer section is appended to the system prompt: the model must call final_answer exactly once when the task is complete, and only that call is read as the answer.
3. The tool handler records the validated arguments for the current session; when the run resolves, they are attached as result.output.

LiteAgentResult is RunResult & { output?: unknown } — output is only present when outputSchema is set and the model produced the final answer.

#Details

#See also

• System prompt — how outputSchema extends the prompt with a ## Final answer section.
• Subagents — why children don't inherit the schema.
• Getting started — install and run your first agent.