Skip to main content
Source: examples/agent-fake-model

​
Overview

This template demonstrates how to use fake() from aixyz/model to build and test an agent with zero API calls and no API key. The fake model’s transform function deterministically maps each user message to a response, making every test fast, repeatable, and CI-safe. The example agent checks whether the user’s input is a palindrome, or reverses it if not — using both the lastMessage and the full prompt history from the transform signature.

​
Project Structure

agent-fake-model/
├── aixyz.config.ts         # Agent metadata
├── app/
│   ├── agent.ts            # Agent using fake() model
│   └── agent.test.ts       # Fully deterministic test suite
├── package.json
└── vercel.json

​
The Fake Model

Import fake from aixyz/model and pass a transform function: 
import { fake } from "aixyz/model";

export const model = fake((lastMessage, prompt) => {
  const reversed = [...lastMessage].reverse().join("");
  const isPalindrome = lastMessage.toLowerCase() === reversed.toLowerCase();
  const turn = prompt.filter((m) => m.role === "user").length;

  if (isPalindrome) {
    return `"${lastMessage}" is a palindrome! (turn ${turn})`;
  }
  return `"${lastMessage}" reversed is "${reversed}" (turn ${turn})`;
});

export default new ToolLoopAgent({
  model,
  instructions: "You analyze text for palindromes and reverse it when asked.",
});
The transform receives:
  • lastMessage — the text content of the last user turn
  • prompt — the full conversation history, useful for tracking turn count or extracting prior context

​
Test Pattern

Because the fake model is deterministic, every test runs without an API key: 
import { describe, expect, test } from "bun:test";
import type { Prompt } from "aixyz/model";
import { model } from "./agent";

describe("palindrome checker (fake model)", () => {
  test("detects a palindrome", async () => {
    const prompt: Prompt = [{ role: "user", content: [{ type: "text", text: "racecar" }] }];
    const result = await model.doGenerate({ prompt });
    expect(result.content[0]).toEqual(
      expect.objectContaining({ type: "text", text: expect.stringContaining("is a palindrome") }),
    );
  });

  test("tracks turn number from prompt context", async () => {
    const prompt: Prompt = [
      { role: "user", content: [{ type: "text", text: "hi" }] },
      { role: "assistant", content: [{ type: "text", text: "..." }] },
      { role: "user", content: [{ type: "text", text: "level" }] },
    ];
    const result = await model.doGenerate({ prompt });
    expect(result.content[0]).toEqual(
      expect.objectContaining({ type: "text", text: expect.stringContaining("turn 2") }),
    );
  });
});

​
Key Features

  • No API key — fake() never makes network requests
  • Fully deterministic — same input always produces the same output
  • CI-safe — all tests run in any environment
  • Full prompt context — the transform’s second argument exposes the entire conversation history

​
Running Tests

cd examples/agent-fake-model
bun install
bun test

​
Running the Agent

cd examples/agent-fake-model
bun run dev
Endpoints available at http://localhost:3000:
EndpointProtocolDescription
/.well-known/agent-card.jsonA2AAgent discovery card
/agentA2AJSON-RPC task handler