Glossary

Agent Event Bus

Name: Fleet
Author: Fleet

An agent event bus is a shared messaging system that lets AI agents publish structured events and subscribe to events published by other agents, enabling asynchronous coordination without direct agent-to-agent calls.

Direct agent-to-agent calls are the simplest coordination model: agent A finishes and calls agent B directly. This works for two-agent systems but becomes brittle as the team grows. Each agent must know the addresses and interfaces of every agent it needs to hand off to. Adding a new agent type requires modifying every upstream agent.

An event bus decouples producers from consumers. Agent A publishes a pr_created event; it does not need to know that agent B is subscribed to that event type. Agent B subscribes to pr_created events; it does not need to know which agent produced them. The bus handles routing, delivery, and durability.

For agent coordination specifically, the event schema is important: events should carry enough context for the consuming agent to take action without needing to query back to the producer. A pr_created event should include the PR number, branch, associated issue, and repository — not just a notification that a PR was opened.

How this relates to Fleet

Fleet's event bus is called the fabric. It accepts JSON events from any agent via a simple publish command, stores them in SQLite, and matches them against agent subscriptions. When a match is found, the watcher daemon starts the subscribed agent with the event payload available as an environment variable. The fabric can operate over a Unix socket or fall back to direct SQLite writes when the socket is unavailable.

Frequently asked questions

What is the difference between an event bus and a message queue?

A message queue is point-to-point: a producer sends to a specific queue that a specific consumer reads. An event bus is broadcast: producers publish to a topic, and multiple consumers can independently subscribe to the same topic. For multi-agent systems, the bus model is more flexible because new agent types can subscribe to existing events without modifying producers.

Does an agent event bus need to be a separate infrastructure component?

Not necessarily. Fleet implements its fabric event bus in SQLite, running on the same machine as the agents. For small-to-medium teams, this is sufficient and eliminates an external dependency. Larger deployments might use Kafka, Redis Streams, or a cloud pub/sub service for higher throughput and multi-machine distribution.

Run your first agent fleet

One binary. Five minutes. See every agent, coordinate every handoff, and keep a full audit trail of what your fleet did.

See how it works Install Fleet

← Back to the AI agent glossary