Inferensys

Glossary

Network Intent

A declarative, high-level specification of a desired network outcome or business objective—such as security posture or latency threshold—expressed independently of the underlying technical implementation details.
Security engineer implementing LLM guardrails on laptop, safety rules visible on screen, technical implementation session.
DECLARATIVE NETWORKING

What is Network Intent?

A network intent is a declarative, high-level specification of a desired business outcome, expressed independently of the technical mechanisms used to achieve it.

Network Intent is a declarative specification of a desired network state or business objective—such as a security posture or latency threshold—expressed independently of the underlying device-level configurations and vendor-specific syntax required to implement it. It represents the highest level of the policy continuum, focusing on what the business needs rather than how the network should be programmed.

Within an Intent-Based Networking (IBN) architecture, the declared intent is ingested by an intent engine, which algorithmically validates it for conflicts and translates it into low-level network configuration synthesis. A closed-loop assurance system then continuously monitors real-time telemetry collection to detect intent drift, automatically triggering remediation workflows to maintain continuous intent compliance with the original business objective.

DECLARATIVE DESIGN PRINCIPLES

Core Characteristics of a Network Intent

A network intent is a formal, declarative specification of a desired business outcome. It abstracts the 'what' from the 'how,' enabling autonomous closed-loop systems to handle the underlying technical implementation.

01

Declarative Abstraction

A network intent specifies the desired outcome without prescribing the specific configuration commands or device-level procedures required to achieve it. This decouples business policy from vendor-specific implementation details.

  • What, not How: An intent states 'Ensure HIPAA compliance for this traffic' rather than 'Configure ACL 101 on interface Gi0/1.'
  • Idempotency: The system continuously enforces the declared state, correcting any drift automatically.
  • Portability: The same high-level intent can be applied across heterogeneous hardware from different vendors.
100%
Abstraction from CLI
02

Measurable Service-Level Objectives

An intent must contain precise, quantifiable metrics that allow the closed-loop assurance system to validate compliance continuously. Vague requirements are not actionable.

  • Latency Thresholds: 'Intra-pod traffic must not exceed 5ms round-trip.'
  • Availability Guarantees: 'Service must maintain 99.999% uptime over a rolling 30-day window.'
  • Path Constraints: 'Traffic must never transit through geographic region X.'
  • Measurable KPIs: These SLOs serve as the ground truth for the intent assurance loop.
< 5ms
Typical Latency SLO
03

Continuous Closed-Loop Validation

A network intent is not a one-time configuration push. It establishes a persistent control loop that constantly compares the network's actual state against the declared intent.

  • Telemetry Ingestion: Streaming high-frequency metrics from all devices under management.
  • Drift Detection: Algorithms identify any divergence between the intended and operational state.
  • Automated Remediation: The system triggers pre-defined remediation workflows to restore compliance without a human ticket.
  • State Machine: The intent exists in a formal lifecycle, transitioning between states like 'Fulfilled,' 'Drifted,' and 'Remediating.'
Sub-second
Drift Detection Speed
04

Conflict-Free Policy Hierarchy

Multiple intents can coexist, but they must be structured within a policy continuum that defines precedence and resolves contradictions algorithmically.

  • Business Intent: Highest level, e.g., 'Prioritize real-time trading traffic.'
  • Operational Intent: Translates business goals into technical constraints.
  • Conflict Resolution: If two intents compete for the same bandwidth, a priority-based or negotiation-based arbiter decides the outcome.
  • Pre-Deployment Validation: The intent validation engine checks for logical inconsistencies before any configuration is synthesized.
3-Tier
Policy Continuum Layers
05

Vendor-Neutral Data Models

Intents are expressed using standardized, abstract data models (typically YANG-based) rather than proprietary command-line interfaces. This ensures the intent engine can synthesize configurations for any compliant device.

  • Schema-Driven: Intents conform to a strict, machine-readable schema.
  • Intent-Based APIs: Northbound interfaces allow business applications to declare requirements using RESTCONF/NETCONF.
  • Translation Layer: The intent translation function maps the abstract model to device-specific syntax for Cisco, Juniper, Arista, etc.
  • No Screen Scraping: Eliminates the fragility of CLI parsing.
YANG
Standard Data Model
06

Autonomous Lifecycle Management

A network intent has a formal, managed lifecycle from creation to decommissioning, governed by an intent state machine. It is not a static configuration file.

  • Creation & Validation: The intent is authored and checked for feasibility.
  • Fulfillment: The system orchestrates resources and pushes synthesized configurations.
  • Assurance: Continuous monitoring ensures the intent remains compliant.
  • Modification & Retirement: Intents can be dynamically updated or safely removed, with the system automatically cleaning up the underlying configurations.
6+
Lifecycle States
NETWORK INTENT

Frequently Asked Questions

Clear, technical answers to the most common questions about declarative network intent, its lifecycle, and its role in closed-loop automation.

Network intent is a declarative, high-level specification of a desired network outcome or business objective—such as a security posture, latency threshold, or application priority—expressed independently of the underlying technical implementation details. It works by feeding this abstract policy into an intent engine, which algorithmically validates, translates, and decomposes the intent into device-specific, low-level configurations. A closed-loop assurance system then continuously monitors streaming telemetry to verify that the operational state matches the declared intent, automatically triggering remediation workflows if drift is detected. This paradigm shifts network management from imperative, device-by-device programming to outcome-oriented automation.

Prasad Kumkar

About the author

Prasad Kumkar

CEO & MD, Inference Systems

Prasad Kumkar is the CEO & MD of Inference Systems and writes about AI systems architecture, LLM infrastructure, model serving, evaluation, and production deployment. Over 5+ years, he has worked across computer vision models, L5 autonomous vehicle systems, and LLM research, with a focus on taking complex AI ideas into real-world engineering systems.

His work and writing cover AI systems, large language models, AI agents, multimodal systems, autonomous systems, inference optimization, RAG, evaluation, and production AI engineering.