Messaging

Model event-driven systems with append-only topics and reactive subscriptions.

What it is

Messaging provides durable topic streams plus consumer semantics for fan-out processing, replay, and backpressure-aware handling.

When to use it

• Event-driven architectures with multiple independent consumers.
• Pipelines that require replay or sequence-aware processing.
• Systems needing bounded logs, compaction, and topic-level isolation.

When not to use it

• Tight DAG orchestration where step dependencies are primary.
• In-memory local state updates without message semantics.

Core primitives

• topic() for append-only streams.
• subscription() for tracked consumers and offsets.
• repeatPublish() for retries/re-delivery workflows.
• jobFlow() when stream processing and job semantics overlap.

Typical usage flow

1. Define a topic for your domain events.
2. Create one or more subscription consumers.
3. Process and commit messages, or retry with policies.
4. Monitor stream growth and consumer lag through companion stores.

Start here

• API:
  • topic()
  • subscription()
  • repeatPublish()
  • jobFlow()
• Recipes:
  • Tool Calls for Local LLMs
  • Reactive Data Pipeline