Skip to main content
A drop-in LanguageModelV3 implementation that maps user messages through a transform function — no API key, no network calls, fully deterministic. 
import { fake, type Prompt } from "aixyz/model";

​
fake(transform)

Creates a fake language model conforming to the Vercel AI SDK LanguageModelV3 specification. The returned model can be passed directly to ToolLoopAgent or any AI SDK function that accepts a LanguageModel. 
function fake(transform: (lastMessage: string, prompt: Prompt) => string): LanguageModelV3;

​
Parameters

ParameterTypeDescription
transform(lastMessage: string, prompt: Prompt) => stringFunction that receives the last user message text and the full prompt, and returns the model output string
The transform function receives two arguments:
  • lastMessage — the text content of the most recent user message (empty string if none)
  • prompt — the full LanguageModelV3Prompt conversation history, useful for tracking turn count or prior context

​
Return value

A LanguageModelV3 object with:
  • specificationVersion: "v3"
  • provider: "aixyz/fake"
  • modelId: "aixyz/fake"
  • doGenerate() and doStream() that call your transform and report zero token usage

​
Examples

Simple echo: 
import { fake } from "aixyz/model";

const model = fake((input) => `You said: ${input}`);
Using full prompt context: 
import { fake } from "aixyz/model";

const model = fake((input, prompt) => {
  const turn = prompt.filter((m) => m.role === "user").length;
  return `Turn ${turn}: ${input}`;
});
Wiring into an agent: 
import { fake } from "aixyz/model";
import { ToolLoopAgent } from "ai";

export const model = fake((input) => `Echo: ${input}`);

export default new ToolLoopAgent({
  model,
  instructions: "You echo back whatever the user says.",
});

​
Prompt

Type alias for LanguageModelV3Prompt from @ai-sdk/provider. This is an array of messages where each message has a role and content: 
import type { Prompt } from "aixyz/model";
Use this type when you need to reference the prompt shape in your transform function or tests: 
import { fake, type Prompt } from "aixyz/model";

const model = fake((_input: string, prompt: Prompt) => {
  return `${prompt.length} messages in history`;
});

​
Testing with fake()

The fake model makes every test deterministic and CI-safe. Export the model from your agent file so tests can call doGenerate() directly: 
// app/agent.test.ts
import { describe, expect, test } from "bun:test";
import { model } from "./agent";
import type { Prompt } from "aixyz/model";

describe("agent (fake model)", () => {
  test("echoes the user message", async () => {
    const prompt: Prompt = [{ role: "user", content: [{ type: "text", text: "hello" }] }];
    const result = await model.doGenerate({ prompt });
    expect(result.content).toEqual([{ type: "text", text: "Echo: hello" }]);
  });
});
See the Testing guide and the Fake Model Agent template for complete examples.