Mastra

Run Mastra agents in TypeScript on this API with a custom OpenAI-compatible model.

Mastra takes a custom OpenAI-compatible endpoint as the agent's model: an object with id, url and apiKey. Set url to https://test-demo.pinerouter.ai/v1, the base URL, not the chat endpoint.

Install

npm install @mastra/core zod

pnpm add @mastra/core zod

bun add @mastra/core zod

Configure API keys

.env

API_KEY=sk-...

Load it with node --env-file=.env or your framework's own .env support. Keep the key out of your code.

Run

Text

import { Agent } from "@mastra/core/agent";

const model = {
  id: "custom/gpt-6-luna",
  url: "https://test-demo.pinerouter.ai/v1",
  apiKey: process.env.API_KEY,
} as const;

const agent = new Agent({
  id: "assistant",
  name: "Assistant",
  instructions: "You are concise.",
  model,
});

const res = await agent.generate("Name three rivers in Europe.");
console.log(res.text);

Stream

const stream = await agent.stream("Write a haiku about rivers.");
for await (const chunk of stream.textStream) {
  process.stdout.write(chunk);
}

Tool calling

import { createTool } from "@mastra/core/tools";
import { z } from "zod";

const weatherTool = createTool({
  id: "get-weather",
  description: "Current weather for a city",
  inputSchema: z.object({ city: z.string() }),
  execute: async ({ city }) => ({ city, forecast: "Sunny, 21 C" }),
});

const weatherAgent = new Agent({
  id: "weather",
  name: "Weather",
  instructions: "Answer weather questions with the tool.",
  model,
  tools: { weatherTool },
});

console.log((await weatherAgent.generate("What is the weather in Paris?")).text);

Specifics

In id, the part before / is your own label and the part after it is our model id, so custom/gpt-6-luna calls gpt-6-luna.
Custom url models use Chat Completions by default. Add api: "responses" to use /v1/responses instead.
Mastra also accepts AI SDK providers as model, so the Vercel AI SDK provider works here too.
For images, video and other media call Jobs with the HTTP helper inside a tool.
Tool calling depends on the model. Test with the model you plan to use.