Skip to main content

Mastra Framework

Use the Mastra framework with the gateway by selecting a gateway-backed model such as mastra/openai/gpt-5-mini. Then pass memory.thread and memory.resource when you stream to enable observational memory.

Set your API key​

Set your API key in the environment variable MASTRA_GATEWAY_API_KEY to authenticate with the gateway.

GNU Bash.env
MASTRA_GATEWAY_API_KEY="sk-..."

Set the gateway model ID​

Set the agent model to a gateway model ID. You do not need to wire custom headers or map request context manually.

TypeScriptsrc/mastra/agents/weather-agent.ts
import { Agent } from '@mastra/core/agent'

export const weatherAgent = new Agent({
	id: 'weather-agent',
	name: 'Weather Agent',
	instructions: 'You are a helpful weather assistant.',
	model: 'mastra/openai/gpt-5-mini',
})

Stream with server-enriched memory​

Pass the same thread and resource values on each request. The gateway enriches each call with stored observations for that conversation automatically.

TypeScriptsrc/mastra/run.ts
import { weatherAgent } from './agents/weather-agent'

const memory = {
	thread: 'weather-thread-1',
	resource: 'user-42',
}

const result = await weatherAgent.stream('My name is Alex and I prefer concise answers.', {
	memory,
})

for await (const chunk of result.textStream) {
	process.stdout.write(chunk)
}

const followUp = await weatherAgent.stream('What is my name?', {
	memory,
})

for await (const chunk of followUp.textStream) {
	process.stdout.write(chunk)
}
// "Alex"
  • Features: Observational memory, streaming, BYOK, and gateway tools
  • Vercel AI SDK: Use @ai-sdk/openai-compatible and @ai-sdk/anthropic
  • API reference: Complete endpoint documentation