Build realtime apps
without building realtime infrastructure.

Without Messaging Platform

• A WebSocket service, with a polling fallback for networks that block it
• Authentication, API keys and short-lived browser credentials
• Reconnect and resume without losing or duplicating messages
• Channels, rooms and who may join them
• Persistence, ordering and replay from an offset
• Presence: joins, leaves, and noticing the ones that never say goodbye
• Shared state that late joiners can catch up on
• WebRTC signaling, plus STUN/TURN for the peers that cannot connect directly
• Rate limits, quotas and usage metering
• Client libraries for every language you ship

With Messaging Platform

const agent = new AgentConnection();

// messages and events
agent.addEventListener('message', e => render(e.response.data));
// presence
agent.addEventListener('agent-connect', e => showOnline(e.agentName));
agent.addEventListener('connect', () => {
  agent.sendMessage({ content: 'hello, room' });
  // shared state that late joiners catch up on
  agent.storagePut({ storageKey: 'board', content: board });
});

agent.connect({ api, apiKey, channelName: 'room-123',
                channelPassword, agentName, autoReceive: true });

// peer-to-peer media and data over the same channel
const rtc = new WebRtcHelper(agent);

Real SDK calls, not pseudocode. Run your first one →

What it is for

• Streaming an agent's answer to the browser as it is generated
• Human approval: the agent asks, a person approves, the agent continues
• Tool progress and status events while long work runs
• Several agents, and several people, in one room
• A Python agent and a Java service exchanging work without a queue to run

An agent in a few lines of Python

agent = AgentConnection.with_api_key(
    API, os.environ["MESSAGING_PLATFORM_API_KEY"])
agent.connect("support-room", ROOM_PASSWORD, "assistant")

class Assistant(AgentConnectionEventHandler):
    def on_message_events(self, events):
        for e in events:
            if e.get("from") == "assistant":
                continue  # skip our own replies
            if e.get("type") == "chat-text":
                reply = answer(e.get("content"))  # your model
                agent.send_message(reply)

agent.receive_async(Assistant())

Messaging Platform is not a model provider; it is the channel your model talks through.