Deploy AI agents on any infrastructure.

Fleet is a lightweight server framework for hosting agents with built-in orchestration, tool integration (MCP), and Agent2Agent (A2A) communication.

Features

• Multi-Framework Support — Express and Hono adapters (with edge runtime compatibility on the way).
• A2A Protocol Compliant — Full JSON-RPC 2.0 implementation for agent interactions
• MCP Integration — Connect to Model Context Protocol servers for tool access
• Pluggable Storage — In-memory, SQLite, or custom storage backends
• Custom Middleware — Intercept and transform requests/responses
• Multi-Tenant Ready — Built-in user isolation for SaaS deployments

Installation

npm install @artinet/fleet openai @modelcontextprotocol/sdk @a2a-js/sdk

Requirements: Node.js >= 18.9.1

Quick Start

1. Launch a Fleet

Set an OPENAI_API_KEY in your environment variables, then start your server.

Express:

npm install express

import { fleet } from '@artinet/fleet/express';

fleet().launch(3000);

Hono:

npm install hono

import { fleet } from '@artinet/fleet/hono';

fleet().launch(3000);

That's it. You now have:

• POST /deploy — Deploy agents
• POST /test — Test agent deployments
• GET /agentId/:id/.well-known/agent-card.json — Agent metadata
• POST /agentId/:id — JSON-RPC agent interaction

2. Deploy an Agent

Prelaunch (ship before launch):

import { fleet } from '@artinet/fleet/express';

const myFleet = await fleet().ship([
    {
        config: {
            uri: 'my-agent',
            name: 'My Agent',
            description: 'A helpful assistant',
            modelId: 'gpt-4o',
            instructions: 'You are a helpful assistant.',
            version: '1.0.0',
            skills: [],
            capabilities: {},
            defaultInputModes: ['text'],
            defaultOutputModes: ['text'],
            services: [],
        },
    },
]);

myFleet.launch(3000);

Post Launch (ship to a running server):

Use the ship command, it uses zod to verify agent configurations before sending them to fleet.

import { ship } from '@artinet/fleet';

await ship('http://localhost:3000', {
    config: {
        uri: 'my-agent',
        name: 'My Agent',
        description: 'A helpful assistant',
        modelId: 'gpt-4o',
        instructions: 'You are a helpful assistant.',
        version: '1.0.0',
        skills: [],
        capabilities: {},
        defaultInputModes: ['text'],
        defaultOutputModes: ['text'],
        services: [],
    },
});

With MCP Tools:

import { ship } from '@artinet/fleet';

await ship('http://localhost:3000', {
    config: {
        uri: 'tool-agent',
        name: 'Tool Agent',
        description: 'An agent with access to tools',
        modelId: 'gpt-4o',
        instructions: 'You are a helpful assistant with tool access.',
        version: '1.0.0',
        skills: [],
        capabilities: {},
        defaultInputModes: ['text'],
        defaultOutputModes: ['text'],
        services: [
            {
                type: 'mcp',
                uri: 'everything-server',
                info: {
                    implementation: {
                        version: '0.0.1',
                        name: 'everything',
                    },
                },
                arguments: {
                    command: 'npx',
                    args: ['-y', '@modelcontextprotocol/[email protected]'],
                },
            },
        ],
    },
});

3. Talk to Your Agent

Via curl:

curl -X POST http://localhost:3000/agentId/my-agent \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "message/send",
    "params": {
      "message": {
        "messageId": "hello-id",
        "kind": "message",
        "role": "user",
        "parts": [{ "kind": "text", "text": "Hello!" }]
      }
    }
  }'

Via the SDK:

import { createMessenger } from '@artinet/sdk';

const messenger = createMessenger({
    baseUrl: 'http://localhost:3000/agentId/my-agent',
});

// Send a message
const response = await messenger.sendMessage('Hello!');
console.log(response);

// Or stream the response
for await (const update of messenger.sendMessageStream('Tell me a story')) {
    console.log(update);
}

Documentation

| Document | Description | | ---------------------------------- | ---------------------------------- | | Settings | Complete settings reference | | Storage | Storage backends and configuration | | Middleware | Request/response interception | | API Reference | Endpoints and JSON-RPC methods | | Deployment | Docker and production deployment |

API Reference

Endpoints

| Method | Path | Description | | ------ | ------------------------------------------ | -------------------- | | POST | /deploy | Deploy a new agent | | POST | /test | Test an agent | | GET | /agentId/:id/.well-known/agent-card.json | Get agent card | | POST | /agentId/:id | JSON-RPC interaction |

JSON-RPC Methods

| Method | Description | | ---------------- | ----------------------------- | | message/send | Send a message, get response | | message/stream | Send a message, stream events | | task/get | Get task status | | task/cancel | Cancel a running task | | resubscribe | Stream events |

Architecture

@artinet/fleet
├── /express     # Express adapter
├── /hono        # Hono adapter
└── /sqlite      # SQLite storage adapter

Depends on:
├── @artinet/armada   # Core business logic
├── @artinet/sdk      # A2A protocol client/server
├── orc8              # Agent/Tool orchestration
├── agent-def         # Standardized Agent Definitions
├── openai            # OpenAI API Client
└── @mcp              # @modelcontextprotocol/sdk

Testing

npm test

License

Apache-2.0

Contributing

Contributions welcome! Please open an issue or PR on GitHub.