About SAMP
Learn about SAMP, its purpose, protocol scope, current Internet-Draft status, participation model and author.
The Simple Agent Management Protocol (SAMP) is an experimental, vendor-neutral management-plane protocol for heterogeneous AI agents.
SAMP defines a common operational interface through which management systems can discover and enroll agents, inspect their operational state and capabilities, receive events, subscribe to event streams, apply controlled configuration, and invoke explicitly exposed operations under policy control.
Inspired by established network-management principles, SAMP approaches AI agents as managed operational entities rather than treating them exclusively as reasoning systems or task-execution components.
Why SAMP Exists
AI-agent ecosystems are rapidly developing protocols for communication, collaboration, task delegation, tool access, and context exchange.
These capabilities do not, by themselves, provide a consistent operational management plane for heterogeneous agents throughout their lifecycle.
Operators still need reliable ways to answer practical questions:
- Which agents exist in the environment?
- What identities, capabilities, tools, and resources do they expose?
- Are they reachable, healthy, trusted, and operating within policy?
- How can configuration changes be applied safely?
- How should operational events, failures, and uncertain execution outcomes be handled?
- Which management operations are authorized, and how is that authorization enforced?
- How can operational evidence support verification, accountability, and recovery?
SAMP provides a common protocol foundation for addressing these requirements across different agent frameworks, runtimes, and deployment environments.
Scope and Boundaries
SAMP is not an agent-to-agent communication protocol, an AI-agent framework, or a tool-use protocol.
It complements technologies such as A2A and MCP by addressing a distinct relationship: operational management between a management authority and managed AI agents.
SAMP draws from network and infrastructure management principles while introducing requirements relevant to AI-agent environments, including:
- Dynamic agent profiles and capability declarations.
- Explicit enrollment and trust management.
- Independent trust and liveness states.
- Management-impact autonomy classifications.
- Policy-controlled configuration and execution.
- Authorization bound to specific consequential operations.
- Runtime evidence and operational accountability.
- Failure-aware management and execution recovery.
SAMP does not prescribe how an AI agent reasons, plans, remembers, or collaborates.
It defines how that agent can be observed, managed, and controlled within an operational environment, subject to applicable security and authorization policies.
Current Status
SAMP version 0.2 is documented in:
draft-efstathiou-samp-agent-management-04
Published: 26 October 2026
Intended status: Experimental
Read the current Internet-Draft
SAMP is currently an individual Internet-Draft. It has not been adopted by an IETF Working Group, is not endorsed by the IETF, and has no formal standing in the IETF standards process.
The specification is being evaluated through technical review, experimental implementations, security analysis, and interoperability testing.
The Agent Management System (AgMS) provides an Alpha-stage experimental implementation environment for exploring SAMP management concepts with heterogeneous AI agents.
Participation
SAMP is an open technical effort welcoming independent review, criticism, implementation experience, and interoperability research.
Contributions from network and security engineers, protocol designers, AI developers, researchers, infrastructure operators, and independent implementers are encouraged.
Collaboration with complementary protocols and security mechanisms is particularly welcome where it can improve interoperability without introducing unnecessary dependencies.
Technical contact:
Security reports:
General enquiries:
Author
SAMP was initiated and authored by Stelio E. Efstathiou, an independent Network, Security and Infrastructure Architect with more than three decades of professional experience across telecommunications carriers, data centres, enterprise infrastructure, and European institutional environments.
The project was publicly launched in August 2026 as an open technical initiative addressing the operational management of heterogeneous AI agents.
Its development combines established network-management engineering principles with practical experimentation in AI-agent infrastructure.
The objective is straightforward: bring consistent, interoperable, and accountable operational management to heterogeneous AI-agent environments.