ModelOutput

KIT · interface · Type contract

The model-authored part of a reply, before it becomes JSON or SSE frames.

Import

TypeScript

import type { ModelOutput } from '@jangts/agent-core/mock';

JavaScript

/** @typedef {import('@jangts/agent-core/mock').ModelOutput} ModelOutput */

Other public entry points:@jangts/agent-core/mock.

Definition

interface ModelOutput

Extends or implements: OutputOptions.

Field structure

Field	Type	Required	Description
content	string | null | undefined	Optional	Assistant text; null or omitted means no textual content (common for tool-call replies).
tools	readonly ToolCall[] | undefined	Optional	Tool calls emitted by the assistant; their serialized size also counts toward limits.
finishReason	"length" | "tool_calls" | "stop" | "content_filter" | undefined	Optional	Finish reason sent in the last content frame; defaults to tool_calls with tools, else stop.
usage	Usage | undefined	Optional	Usage object included in non-stream replies and in the terminal usage chunk.
delayMs	number | undefined	Optional	Artificial latency before the reply or first frame, in milliseconds; default 0.
chunkSize	number | undefined	Optional	Maximum Unicode code points per content/tool-call delta; default 24; never splits a code point.
intervalMs	number | undefined	Optional	Delay between SSE frames in milliseconds; default 0 (emit as fast as the consumer pulls).
wireChunkSize	number | undefined	Optional	Optional transport-level byte chunks, independent of content/codepoint chunks.
interruptAfter	number | undefined	Optional	Error the stream after this many data frames, without a success terminator.

Examples

• MockGPT: response descriptions and scenario generators: Fixed results serve strict regression, input-dependent generators explore wiring, and explicit response descriptions cover failure and streaming scenarios.

Encode a validated model result

Extension · Node.js

The adapter handles request validation, pre-first-byte delay, and lifecycle; the encoder does not execute rules or modify MockGPT call records.

TypeScript

import { encodeModelResponse } from '@jangts/agent-core/mock';
const controller = new AbortController();
const response = encodeModelResponse(
  { model: 'demo', messages: [{ role: 'user', content: '状态' }], stream: false },
  { content: '服务在线。' }, 'mock-1', 1700000000, controller.signal,
  (error, cancelled) => console.log({ error, cancelled }),
);
console.log(await response.json());

JavaScript

import { encodeModelResponse } from '@jangts/agent-core/mock';
const controller = new AbortController();
const response = encodeModelResponse({ model: 'demo', messages: [{ role: 'user', content: '状态' }], stream: false }, { content: '服务在线。' }, 'mock-1', 1700000000, controller.signal, (error, cancelled) => console.log({ error, cancelled }));
console.log(await response.json());

Generate an inspectable summary from the input

Beginner · Node.js

TypeScript

import { MockGPT, generators } from '@jangts/agent-core/mock';
const mock = new MockGPT({ rules: [{ generator: generators.summary({ maxChars: 120, count: 2 }) }] });
console.log(await mock.generate({ model: 'demo', messages: [{ role: 'user', content: '服务刚完成维护。健康检查通过。用户可以查询状态。' }] }));
console.log(mock.records);

JavaScript

import { MockGPT, generators } from '@jangts/agent-core/mock';
const mock = new MockGPT({ rules: [{ generator: generators.summary({ maxChars: 120, count: 2 }) }] });
console.log(await mock.generate({ model: 'demo', messages: [{ role: 'user', content: '服务刚完成维护。健康检查通过。用户可以查询状态。' }] }));
console.log(mock.records);

Related definitions

• OutputOptions
• ToolCall
• Usage