The fastest way to understand AgentWorld and navigate the stack.
Start Here
AgentWorld is infrastructure for businesses run by coordinated agents on Solana.
The system is designed around one idea: autonomy only works when roles, state, policy, and settlement are all explicit. A single model call can generate output. It cannot safely run a business by itself.
What AgentWorld provides
a runtime for bounded agents
a workflow engine for multi-agent coordination
durable state and replayable execution records
policy checks and approval checkpoints
Solana-based settlement for money-moving actions
Mental model
event -> workflow -> task -> agent -> tool call -> policy -> state change -> approval or settlement
That sequence is the core of the platform.
Every useful action should fit somewhere in that chain. If it does not, it is probably hidden logic that will become hard to inspect later.
Recommended reading order
Overview and Thesis for the product model.
System Architecture for the runtime layers.
Agent Runtime for the execution contract.
Coordination and Workflows for orchestration.
Solana Execution Layer for onchain settlement.
State, Memory, and Data Model for persistence.
Security and Governance for trust boundaries.
Developer Guide for implementation patterns.
Reference Business Patterns for concrete examples.
What makes the system business-grade
A business-grade agent platform needs more than generation quality.
It needs deterministic control around model behavior, clear policy at every risk boundary, durable system state, and hard separation between reasoning and signing authority. AgentWorld is organized to preserve those properties as workflows become more autonomous.
If you are deciding what to build first, start with a narrow workflow that has repeatable inputs, clear success criteria, and limited downside per action.
