AI Agent Architecture for Network Digital Twin
draft-zhao-nmrg-ai-agent-for-ndt-00
This document is an Internet-Draft (I-D).
Anyone may submit an I-D to the IETF.
This I-D is not endorsed by the IETF and has no formal standing in the
IETF standards process.
| Document | Type | Active Internet-Draft (individual) | |
|---|---|---|---|
| Authors | Jing Zhao , Ran Pang , Shuai Zhang , Hongwei Shi , Chen Su | ||
| Last updated | 2026-03-02 | ||
| Replaces | draft-zhao-nmrg-ai-agent-for-dtn | ||
| RFC stream | (None) | ||
| Intended RFC status | (None) | ||
| Formats | |||
| Stream | Stream state | (No stream defined) | |
| Consensus boilerplate | Unknown | ||
| RFC Editor Note | (None) | ||
| IESG | IESG state | I-D Exists | |
| Telechat date | (None) | ||
| Responsible AD | (None) | ||
| Send notices to | (None) |
draft-zhao-nmrg-ai-agent-for-ndt-00
nmrg J. Zhao, Ed.
Internet-Draft R. Pang, Ed.
Intended status: Standards Track S. Zhang, Ed.
Expires: 3 September 2026 China Unicom
H. Shi, Ed.
C. Sun, Ed.
Purple Mountain Laboratories
2 March 2026
AI Agent Architecture for Network Digital Twin
draft-zhao-nmrg-ai-agent-for-ndt-00
Abstract
This document proposes an AI agent architecture for Network Digital
Twin (NDT) that integrates AI agents with digital twin technology.
Status of This Memo
This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF). Note that other groups may also distribute
working documents as Internet-Drafts. The list of current Internet-
Drafts is at https://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months
and may be updated, replaced, or obsoleted by other documents at any
time. It is inappropriate to use Internet-Drafts as reference
material or to cite them other than as "work in progress."
This Internet-Draft will expire on 3 September 2026.
Copyright Notice
Copyright (c) 2026 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents (https://trustee.ietf.org/
license-info) in effect on the date of publication of this document.
Please review these documents carefully, as they describe your rights
and restrictions with respect to this document. Code Components
extracted from this document must include Revised BSD License text as
described in Section 4.e of the Trust Legal Provisions and are
provided without warranty as described in the Revised BSD License.
Zhao, et al. Expires 3 September 2026 [Page 1]
Internet-Draft AI Agent Architecture for NDT March 2026
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 2
2. AI Agent Architecture for Network Digital Twin . . . . . . . 2
3. Architecture Components . . . . . . . . . . . . . . . . . . . 3
3.1. Network Digital Twin Management AI Agent . . . . . . . . 3
3.2. Functional Model AI Agent . . . . . . . . . . . . . . . . 4
3.3. Basic Model AI Agent . . . . . . . . . . . . . . . . . . 4
3.4. Data Repository AI Agent . . . . . . . . . . . . . . . . 4
4. Agent Interactions . . . . . . . . . . . . . . . . . . . . . 5
5. Intelligent Use Case Realization . . . . . . . . . . . . . . 5
5.1. Automated IP Network Configuration Generation . . . . . . 5
5.1.1. Intent Understanding and Policy Generation . . . . . 5
5.1.2. Multi-Level Simulation and Verification . . . . . . . 6
5.1.3. Model Evolution and Feedback Loop . . . . . . . . . . 6
5.2. Cutover Simulation (Scenario Construction) . . . . . . . 6
6. Security Considerations . . . . . . . . . . . . . . . . . . . 6
7. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 7
8. Informative References . . . . . . . . . . . . . . . . . . . 7
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 7
1. Introduction
Digital twins have emerged as a powerful paradigm for network
management, providing virtual representations of physical networks
that enable simulation, analysis, and optimization. However,
traditional digital twin architectures often lack the autonomous
decision-making capabilities needed for modern network environments.
This document proposes an architecture that combines digital twin
concepts with intelligent AI agents, creating a more dynamic and
responsive network management system.
The architecture is designed to be compatible with existing digital
twin architectures. This approach enables distributed decision-
making, adaptive behavior, and enhanced collaboration between digital
twin components.
2. AI Agent Architecture for Network Digital Twin
Based on the concept of the Network Management Agent (NMA)
[I-D.zhao-nmop-network-management-agent], we propose an AI Agent
architecture for Network Digital Twin (NDT)
[I-D.irtf-nmrg-network-digital-twin-arch]. This architecture extends
the traditional network digital twin by integrating AI agents into
each core component. While preserving the fundamental structure of
digital twins, the architecture introduces enhanced autonomous
capabilities and intelligent decision-making across the network
management lifecycle.
Zhao, et al. Expires 3 September 2026 [Page 2]
Internet-Draft AI Agent Architecture for NDT March 2026
+-------------------------------------------------------------------------------------------+
| Network Digital Twin Management AI Agent |
| |
| - Resource & Health Monitoring - Intent Parsing & Logic Translation |
| - NDT Instance Lifecycle Management - Multi-Agent Session Orchestration |
| - Virtual-Physical State Sync Control - Business Intent & Policy Derivation |
+-------------------------------------------------------------------------------------------+
| |
| |
+---------------------------------------+ +----------------------------------+
| Functional Model Agent | | Data Repository Agent |
| | | |
| - Service Model Optimization |<------------->| - Real-time Telemetry Ingestion |
| - Automated Configuration Synthesis | | & Anomaly Detection |
| - Multi-Vendor Syntax Mapping | | |
| - Hierarchical Simulation Sandbox | | - Historical Data Intelligence |
| (Compliance, Logic, Impact) | | & Pattern Mining |
| - Scenario-specific Modeling | | |
| - Verification Feedback Loop | | - Knowledge Base Management |
+---------------------------------------+ | (Vendor Docs/Best Practice) |
| | |
| | - Adaptive Data Retrieval |
+---------------------------------------+ | & Conflict Resolution |
| Basic Model Agent | | |
| |<------------->| - Standardized Validation |
| - Network Element Digital Models | | Report Storage |
| (Config, State, Env) | +----------------------------------+
| - Topology & Connectivity Models |
| - Real-time Model Self-Updates |
+---------------------------------------+
Figure 1: AI Agent Architecture for Network Digital Twin
The following figure illustrates the hierarchical interaction between
the Management, Functional, Basic, and Data agents.
3. Architecture Components
3.1. Network Digital Twin Management AI Agent
The Network Digital Twin Management Agent serves as the central
coordination and management component, providing the following key
functionalities:
* Resource Monitoring: Continuously tracks and monitors the status,
performance metrics, and operational health of all resources
within the digital twin environment.
Zhao, et al. Expires 3 September 2026 [Page 3]
Internet-Draft AI Agent Architecture for NDT March 2026
* Lifecycle Management: Governs the complete lifecycle of NDT
instances, encompassing instantiation, configuration, state
synchronization, maintenance, and termination.
* Session Control: Orchestrates communication sessions and
interactions among various AI agents to ensure coherent operation.
* Intent Translation & Policy Derivation: Derives executable
policies from high-level business intents through semantic parsing
and internal logic models.
* Virtual-Physical Synchronization Control: Manages bidirectional
data flow between the NDT and the physical network to ensure
accurate representation.
3.2. Functional Model AI Agent
The Functional Model Agent is responsible for advanced service
modeling, configuration generation, and optimization. It
autonomously invokes required functional models per validation
policies and refines them via historical data analysis.
* Service Model Optimization: Refines models through performance
analysis and adaptive learning algorithms.
* Automated Configuration Synthesis: Generates vendor-specific
configurations or intermediate policy representations based on the
intent model.
* Hierarchical Simulation Sandbox: Provides a multi-stage
environment to verify configurations across compliance, logic, and
business impact layers.
TBD.
3.3. Basic Model AI Agent
The Basic Model Agent maintains fundamental network element and
topology representations. It is capable of updating digital models
in real-time based on physical network changes to ensure the accuracy
of validation.
3.4. Data Repository AI Agent
The Data Repository AI Agent serves as the intelligent data
governance and provisioning component. It autonomously manages the
data lifecycle with the following capabilities:
Zhao, et al. Expires 3 September 2026 [Page 4]
Internet-Draft AI Agent Architecture for NDT March 2026
* Real-time Data Collection: Implements multi-protocol ingestion for
streaming telemetry while autonomously detecting data anomalies.
* Historical Data Intelligence: Curates structured data and
knowledge graphs to support pattern mining, and model training.
* Adaptive Data Services: Provides context-aware data retrieval with
intelligent caching and conflict resolution.
* Knowledge Base Integration: Stores network configurations, vendor
documents, and best practices to support incremental model
updates.
* Validation Reporting: Generates standardized, machine-readable
validation reports including rule IDs, configuration items, and
evidence.
4. Agent Interactions
The architecture employs bidirectional Agent-to-Agent (A2A)
communication: the Functional and Basic Model Agents interact with
the Data Repository Agent for data synchronization, while the
Management Agent centrally orchestrates these interactions to
maintain a coherent workflow.
Inter-agent interactions SHOULD support state rollback mechanisms to
ensure the virtual state remains synchronized with the physical
network during failed intent decompositions.
5. Intelligent Use Case Realization
5.1. Automated IP Network Configuration Generation
This use case demonstrates how the AI Agent architecture automates
the end-to-end lifecycle of network configuration.
5.1.1. Intent Understanding and Policy Generation
* Intent Parsing: The Management Agent receives declarative intents,
identifying network objects and resolving logic conflicts between
multiple intents.
* Config Generation: The Functional Model Agent produces vendor-
specific configurations. It maintains context awareness by
retrieving current states from the Basic Model Agent.
Zhao, et al. Expires 3 September 2026 [Page 5]
Internet-Draft AI Agent Architecture for NDT March 2026
* Vendor-agnostic Abstraction: The system uses an intermediate
policy representation to ensure functional consistency across
heterogeneous hardware.
5.1.2. Multi-Level Simulation and Verification
Before deployment, configurations MUST undergo a multi-stage
verification process:
* Semantic Consistency Verification: Ensures that synthesized
configurations result in deterministic network behavior, bridging
the gap between probabilistic AI generation and deterministic
network operations..
* Hierarchical Validation: The Functional Model Agent executes
simulations in a sandbox layer-by-layer: 1. Compliance: Detecting
syntax errors and policy violations. 2. Functional Correctness:
Verifying reachability and protocol convergence via the Basic
Model Agent. 3. Service Impact: Evaluating potential performance
degradation using traffic patterns from the Data Repository Agent.
5.1.3. Model Evolution and Feedback Loop
* Closed-loop Optimization: If verification fails, the Management
Agent feeds error reports back to the generation models of the
Functional Model Agent for iterative optimization.
* Incremental Learning: Experts can manually correct AI outputs,
which are stored in the Data Repository Agent to fine-tune future
generation and verification models.
5.2. Cutover Simulation (Scenario Construction)
* Process Reproduction: The architecture simulates the complete
cutover lifecycle, including device startup/shutdown, routing
adjustments, and configuration delivery.
* Risk Mitigation: By monitoring link status and business capacity
in the NDT, the Network Digital Twin Management AI Agent and
Functional Model Agent jointly identify plan loopholes and
optimize emergency response sequences before physical execution.
6. Security Considerations
AI-Generated Risks: Specific checks must be implemented to detect
"hallucinated" commands or non-compliant security policies generated
by large language models.
Zhao, et al. Expires 3 September 2026 [Page 6]
Internet-Draft AI Agent Architecture for NDT March 2026
7. IANA Considerations
TBD.
8. Informative References
[I-D.irtf-nmrg-network-digital-twin-arch]
Zhou, C., Yang, H., Duan, X., Lopez, D., Pastor, A., Wu,
Q., Boucadair, M., and C. Jacquenet, "Network Digital
Twin: Concepts and Reference Architecture", Work in
Progress, Internet-Draft, draft-irtf-nmrg-network-digital-
twin-arch-12, 27 February 2026,
<https://datatracker.ietf.org/doc/html/draft-irtf-nmrg-
network-digital-twin-arch-12>.
[I-D.zhao-nmop-network-management-agent]
XingZhao, Wang, M., Wu, B., Ceccarelli, D., Zheng, H., and
J. Zhou, "AI based Network Management Agent(NMA): Concepts
and Architecture", Work in Progress, Internet-Draft,
draft-zhao-nmop-network-management-agent-04, 26 February
2026, <https://datatracker.ietf.org/doc/html/draft-zhao-
nmop-network-management-agent-04>.
Authors' Addresses
Jing Zhao (editor)
China Unicom
Beijing
China
Email: zhaoj501@chinaunicom.cn
Ran Pang (editor)
China Unicom
Beijing
China
Email: pangran@chinaunicom.cn
Shuai Zhang (editor)
China Unicom
Beijing
China
Email: zhangs366@chinaunicom.cn
Zhao, et al. Expires 3 September 2026 [Page 7]
Internet-Draft AI Agent Architecture for NDT March 2026
Hongwei Shi (editor)
Purple Mountain Laboratories
Nanjing
China
Email: shihongwei@pmlabs.com.cn
Chen Su (editor)
Purple Mountain Laboratories
Nanjing
China
Email: suchen@pmlabs.com.cn
Zhao, et al. Expires 3 September 2026 [Page 8]