Fleet Overview

Fleet is Astonish’s multi-agent system. Instead of relying on a single agent, Fleet lets you define teams of specialized AI agents that collaborate on complex tasks — each with its own role, tools, and workspace.

Architecture

Fleet uses an event-driven, hub-and-spoke message routing architecture:

• A central router receives messages from all agents and routes them to the appropriate recipient.
• Agents communicate through pairwise conversation threads — dedicated channels between each pair of agents that needs to interact.
• Each agent has its own session, tools, and optional workspace, keeping concerns separated.

When to Use Fleet

Single agent is the right choice for most tasks — one conversation, one agent, full tool access. Use it whenever the work fits naturally into a single workflow.

Fleet is the right choice when:

• A project requires multiple specialized roles (e.g., developer + QA + PM).
• You need parallel workstreams running simultaneously.
• The mission is long-running and benefits from structured coordination.

Key Concepts

Concept	Description
Template	Defines the team composition — agent roles, models, and capabilities.
Plan	A configured instance of a template for a specific mission, with credentials, channels, and settings.
Session	A running execution of a plan, with real-time agent communication.
Thread	A pairwise conversation between two agents within a session.

Triggering Fleet Plans

Fleet plans can be started from multiple entry points:

• Studio chat — /fleet or /fleet-plan slash commands.
• Telegram — /fleet or /fleet_plan commands.
• CLI — astonish fleet activate.
• External sources — automated polling (e.g., GitHub Issues).