A2B is an AI agent governance protocol with a reference implementation architecture. It defines how an agent proves who it is, how policy decides what it may do before execution, and how the outcome is preserved as durable evidence.
Identity → Policy-before-execution → Immutable Ledger
The system knows which agent is requesting an action, for whom it acts, and which capabilities it holds.
Every proposed action is evaluated before it reaches the target system: allow, deny, or require approval.
The request, decision, approval and actual result are preserved in a tamper-evident, reviewable trace.
The protocol applies wherever an agent can create a business side effect.
Email sends, support replies, CRM updates and scheduling proposals.
Control before an agent affects a customer or business record.
Reads, writes, exports and tool calls constrained by identity and scope.
Explicit boundaries around what an agent may access and change.
Destructive, financial or infrastructure actions that need stronger policy and approval.
A reviewable decision path before irreversible execution.
A2B is the umbrella protocol and architecture. A2B Operator is the first hosted product being built to expose its policy, approval and traceability controls through a usable interface and API.
The concept, documentation and implementation work are active. Operator is not presented here as a finished public SaaS subscription.
View the Operator direction →Read the implementation notes and review the public code as the protocol becomes a working control product.