A Manifesto
A framework for building AI agents that can discover, communicate with, and trust one another — across platforms, protocols, and organizations.
The Framework
Inspired by the 12-Factor App methodology — designed for the age of multi-agent AI systems.
FACTOR 01
Every agent must publish a standard, machine-readable Agent Card — a self-describing document declaring its identity, capabilities, endpoints, and trust metadata.
IdentityFACTOR 02
Agent capabilities must be published in a standard, machine-readable format — discoverable by any orchestrator without prior coordination.
DiscoveryFACTOR 03
Agents must not be coupled to a single communication protocol. Support for A2A, MCP, and emerging standards should be composable, not competing.
ProtocolsFACTOR 04
Authorization must be declared, scoped, and verifiable. An agent should never inherit ambient trust — every capability grant must be explicit.
TrustFACTOR 05
Each agent interaction must carry all the context it needs. Shared mutable state across agent boundaries is a coordination failure waiting to happen.
ArchitectureFACTOR 06
Every action an agent takes must emit structured logs, traces, and metrics. Silent agents are unauditable agents — and unauditable agents are ungovernable.
ObservabilityFACTOR 07
Agents must expose discrete, independently invokable capabilities — not monolithic endpoints. Composability is what enables true multi-agent orchestration.
ComposabilityFACTOR 08
Every autonomous action chain must expose well-defined escalation points. Humans should be able to intercept, redirect, or halt any agent workflow at any stage.
GovernanceFACTOR 09
Agents must fail safely. When a dependency is unavailable or a tool call fails, agents should communicate failure clearly — never silently propagate errors downstream.
ResilienceFACTOR 10
Agent standards must be developed in the open. Interoperability is a collective good — no single vendor, platform, or protocol should own the agentic internet.
OpennessGo Deeper
Read the complete rationale behind each factor — with implementation guidance, protocol references, and the thinking behind the framework.
Read the Manifesto →