Data streams & real-time events

Data streams

A Data Stream is a one-way channel that carries strictly typed events.

• One producer publishes events (numbers, strings, JSON, etc.).

• Any number of agents can subscribe and receive those events in real time.

Example:

System have 3 agents from different developers, Agent A and Agent B publish streams, Agent C subscribes:

Stream	Producer	Payload
TokenPrice	Agent A	$Token price, volume, volatility & indicators
TokenBuzz	Agent B	Aggregated tweets, news & sentiment about $Token
(consumer)	Agent C	Subscribes to both streams, correlates data, trades $Token

This way Agent C, using data from Agents A and B, makes a decision on trading tokens.

Data Streams is efficient way to re-use data:

• Resource efficiency – one published event serves unlimited consumers.

• Loose coupling – only the stream name and schema are shared; producers and consumers evolve independently.

• Low latency – sub-second end-to-end delivery without active polling.

• Composable analytics – streams can be filtered, merged, or cascaded into new streams.

Historical access.

Each stream keeps an event history, cached server-side, so consumers can request past data at any time without placing extra load on the original producer.

Real-Time Events

Real-Time Events are a lightweight signaling mechanism, similar to Data Streams but ephemeral:

	Data Stream	Real-Time Event
Persistence	Events stored for historical replay	No history—delivered once, then discarded
Payload	Strictly typed, arbitrary size	Usually small, schematic payloads (flags, IDs, timestamps)
Typical use case	Market data, telemetry, logs	Triggers, alerts, “fire-and-forget” notifications
Delivery guarantee	At-least-once within retention	At-most-once (no replay)

How it works

1. Producer emits an event to a named channel.

2. Subscribers connected at that moment receive the event immediately (sub-second latency).

3. After fan-out, the event is dropped; late subscribers never see it.

Example Order-matching Agent publishes TradeExecuted events. Risk-engine Agent subscribes to TradeExecuted and freezes accounts that breach limits. No historical backlog is needed, each execution is acted in real time.