Aviz Networks Network Copilot Agent SDK details agent architecture and MCP governance

Aviz Networks’ Network Copilot (NCP) Agent SDK outlines an on-premises approach for building AI-driven network agents that can query multi-vendor telemetry, correlate context, and execute governed actions. For enterprise IT and security leaders, it offers a reference architecture for operational automation with audit and access controls.

Research Overview

The blog frames traditional NetOps as fragmented and manual, citing troubleshooting that requires combining CLI output, SNMP traps, log files, and multiple dashboards. It says NCP Agent SDK is designed to let network teams pose network questions in natural language to receive correlated insights and, in some cases, automated remediation.

It positions the update around a private AI architecture and an SDK for creating custom agents, with stated emphasis on vendor-neutral data integration, on-premises deployment, and open integration practices. The blog also connects the platform to compliance, troubleshooting, and operational monitoring use cases.

Key Findings

The NCP architecture combines a fine-tuned large language model with a multi-vendor data integration layer, and it uses a knowledge base approach via Retrieval-Augmented Generation. The blog says all data remains in the user environment for security and that it uses cost-effective open-source LLMs.

It describes multiple agent categories, including interactive assistant agents, proactive monitoring agents, compliance agents, troubleshooting agents aimed at root-cause analysis, and performance/capacity agents. The blog also says built-in capabilities cover compliance monitoring, upgrade verification, security auditing, performance analysis, and business-intent analytics, with customization supported through the SDK.

Technical Breakdown

The blog describes modular components: data ingestion and normalization feed into a unified data store, and an embeddings-driven knowledge base supports RAG. It says an agent runtime connects custom logic with AI reasoning and that connectors use open standards across the stack.

For telemetry and configuration access, it explains that NCP normalizes differences in vendor data models so agents receive structured results. It lists supported collection methods including gNMI streaming, sFlow/NetFlow flow records, direct API calls, SNMP, and existing telemetry agents, as well as integrations with time-series and log systems such as Elastic and Splunk and InfluxDB.

Operational Impact and Governance

To enable autonomous actions, the blog says agents use a tool framework that abstracts network operations into callable tools rather than requiring direct SSH or SQL usage. It describes examples of tools for running device CLI commands, querying telemetry databases, and creating ServiceNow tickets or sending Slack alerts.

On governance, the blog states the platform supports role-based access control for agent deployment and triggering, with policy validation and audit logging for each agent action. It also says the Model Context Protocol (MCP) mediates agent-tool interactions through a mediation layer that enforces authentication and access control while maintaining logged interactions and results.

The blog also describes a conversational user interface in a web portal, with Slack integration mentioned as an option for querying in a collaboration channel. It says the UI supports contextual conversations, streaming responses, conversation export, regeneration of responses, user feedback, and multi-language support.

Blog Signals brief is a fact-based summary of the vendor blog. It consolidates the described NCP Agent SDK architecture for on-premises, vendor-neutral agent development, outlines the agent types and tool framework, and highlights stated security and governance mechanisms such as RBAC, audit logging, and MCP-based mediation.