@reminix/vercel-ai

Reminix agents for the Vercel AI SDK. Serve AI agents as a REST API.

Supports both:

• ToolLoopAgent - Full agents with tools and automatic tool loop handling
• Model - Simple completions via generateText/streamText without tools

Ready to go live? Deploy to Reminix Cloud for zero-config hosting, or self-host on your own infrastructure.

Installation

npm install @reminix/vercel-ai ai @ai-sdk/openai

This will also install @reminix/runtime as a dependency.

Quick Start with Model

import { openai } from '@ai-sdk/openai';
import { VercelAIChatAgent } from '@reminix/vercel-ai';
import { serve } from '@reminix/runtime';

const model = openai('gpt-4o');
const agent = new VercelAIChatAgent(model, { name: 'chat-agent' });
serve({ agents: [agent] });

Quick Start with ToolLoopAgent

For agents with tools, use ToolLoopAgent:

import { ToolLoopAgent, tool } from 'ai';
import { openai } from '@ai-sdk/openai';
import { z } from 'zod';
import { VercelAIChatAgent } from '@reminix/vercel-ai';
import { serve } from '@reminix/runtime';

const weatherTool = tool({
  description: 'Get the current weather for a city',
  inputSchema: z.object({
    city: z.string()
  }),
  execute: async ({ city }) => {
    return { city, temp: 72, condition: 'sunny' };
  }
});

const toolAgent = new ToolLoopAgent({
  model: openai('gpt-4o'),
  instructions: 'You help users check the weather.',
  tools: { getWeather: weatherTool }
});

const agent = new VercelAIChatAgent(toolAgent, { name: 'weather-agent' });
serve({ agents: [agent] });

Quick Start with VercelAIThreadAgent

For agents that need a full thread of messages (including tool call and tool result messages) in their output, use VercelAIThreadAgent. It accepts a LanguageModel and an array of Reminix Tool objects, and runs the Vercel AI SDK tool loop via generateText with stopWhen:

import { openai } from '@ai-sdk/openai';
import { VercelAIThreadAgent } from '@reminix/vercel-ai';
import { tool, serve } from '@reminix/runtime';
import { z } from 'zod';

const getWeather = tool('getWeather', {
  description: 'Get the current weather for a city',
  inputSchema: z.object({ city: z.string() }),
  handler: async ({ city }) => ({ city, temp: 72, condition: 'sunny' }),
});

const agent = new VercelAIThreadAgent(openai('gpt-4o'), {
  tools: [getWeather],
  name: 'weather-thread',
  maxTurns: 5,
});

serve({ agents: [agent] });

Your agent is now available at:

• POST /agents/{name}/invoke - Execute the agent

API Reference

new VercelAIChatAgent(modelOrAgent, options)

Create a Vercel AI chat agent for use with Reminix Runtime.

| Parameter | Type | Default | Description | |-----------|------|---------|-------------| | modelOrAgent | ToolLoopAgent \| LanguageModel | required | A ToolLoopAgent or language model | | options.name | string | "vercel-ai-agent" | Name for the agent (used in URL path) | | options.description | string | "vercel ai chat agent" | Description shown in agent metadata | | options.instructions | string | — | System instructions (used when passing a model, not a ToolLoopAgent) | | options.tags | string[] | — | Tags for categorizing/filtering agents | | options.metadata | Record<string, unknown> | — | Custom metadata merged into agent info |

Returns: VercelAIChatAgent - A Reminix chat agent instance

new VercelAIThreadAgent(model, options)

Create a Vercel AI thread agent for use with Reminix Runtime. The thread agent returns the full message history (including tool calls and tool results) rather than a single text output.

| Parameter | Type | Default | Description | |-----------|------|---------|-------------| | model | LanguageModel | required | A Vercel AI SDK language model | | options.tools | Tool[] | required | Array of Reminix Runtime tools | | options.name | string | "vercel-ai-thread-agent" | Name for the agent (used in URL path) | | options.maxTurns | number | 10 | Maximum number of tool-loop turns | | options.description | string | "vercel ai thread agent" | Description shown in agent metadata | | options.instructions | string | — | System instructions passed to generateText | | options.tags | string[] | — | Tags for categorizing/filtering agents | | options.metadata | Record<string, unknown> | — | Custom metadata merged into agent info |

Returns: VercelAIThreadAgent - A Reminix thread agent instance

Using Different Providers

The Vercel AI SDK supports multiple providers:

import { openai } from '@ai-sdk/openai';
import { anthropic } from '@ai-sdk/anthropic';
import { google } from '@ai-sdk/google';
import { VercelAIChatAgent } from '@reminix/vercel-ai';
import { serve } from '@reminix/runtime';

const gpt = new VercelAIChatAgent(openai('gpt-4o'), { name: 'gpt' });
const claude = new VercelAIChatAgent(anthropic('claude-sonnet-4-5-20250929'), { name: 'claude' });
const gemini = new VercelAIChatAgent(google('gemini-pro'), { name: 'gemini' });

serve({ agents: [gpt, claude, gemini] });

When to Use Each Option

| Option | Use Case | |--------|----------| | VercelAIChatAgent + ToolLoopAgent | Agents that need tools, multi-step reasoning, automatic tool loop | | VercelAIChatAgent + Model | Simple completions, providers without dedicated packages (Google, Mistral, etc.) | | VercelAIThreadAgent | Agents that need full message history with tool calls and results in the output |

Endpoint Input/Output Formats

POST /agents/{name}/invoke

Execute the agent with a prompt or messages.

Request with prompt:

{
  "input": {
    "prompt": "Summarize this text: ..."
  }
}

Request with messages:

{
  "input": {
    "messages": [
      {"role": "user", "content": "Hello!"}
    ]
  }
}

Response (VercelAIChatAgent):

{
  "output": "Hello! How can I help you today?"
}

Response (VercelAIThreadAgent):

{
  "output": [
    {"role": "user", "content": "What is the weather in Paris?"},
    {"role": "assistant", "content": "", "tool_calls": [{"id": "call_1", "type": "function", "function": {"name": "getWeather", "arguments": "{\"city\":\"Paris\"}"}}]},
    {"role": "tool", "content": "{\"city\":\"Paris\",\"temp\":72,\"condition\":\"sunny\"}", "tool_call_id": "call_1"},
    {"role": "assistant", "content": "The weather in Paris is 72F and sunny."}
  ]
}

Streaming

For streaming responses, set stream: true in the request:

{
  "input": {
    "prompt": "Tell me a story"
  },
  "stream": true
}

The response will be sent as Server-Sent Events (SSE).

Note: Streaming is supported by VercelAIChatAgent only. VercelAIThreadAgent does not support streaming.

Runtime Documentation

For information about the server, endpoints, request/response formats, and more, see the @reminix/runtime package.

Deployment

Ready to go live?

• Deploy to Reminix Cloud - Zero-config cloud hosting
• Self-host - Run on your own infrastructure

Links

• GitHub Repository
• Vercel AI SDK Documentation

License

Apache-2.0