In this Networking Field Day session, Aviz Networks introduced Network Copilot (NCP), their private AI platform built for NetOps, emphasizing their vision of "Networks for AI, AI for Networks" and enabling open networking with SONiC and LLMs. The demonstration showcased NCP as a self-hosted software solution that integrates with existing network data sources through various data connectors, such as Cisco Catalyst Center, Nexus Dashboard, IP Fabric, Elastic, Splunk, and Zendesk. The primary goal of NCP is to centralize disparate network data, allowing users to query and correlate information through a natural language chat interface, thereby streamlining operations and reducing the need for manual data aggregation from various tools or complex scripting.

The demo highlighted NCP's ability to perform tasks like inventory analysis and hostname validation. Users can create projects within NCP to focus on specific troubleshooting sessions or events, inviting collaborators and selectively enabling relevant data connectors to avoid data pollution. A key feature is the ability to upload static contextual information, like naming conventions or CVE lists, as files, which NCP can then use to validate device configurations or generate compliance reports. The presenters stressed that NCP doesn't "train" on operational data in the traditional sense; instead, it uses a pre-trained LLM (like Llama 70B), fine-tuned for networking, to interpret questions and leverage AI agents to retrieve, process, and summarize data from connected sources.

While acknowledging that some functions could be replicated with scripting, the true value of NCP lies in its abstraction layer, enabling network engineers to manage diverse multi-vendor, multi-NOS environments without needing deep knowledge of every CLI or proprietary system. This empowers junior engineers by providing suggestions for troubleshooting and allows for more efficient audit reporting and capacity planning. Aviz Networks emphasized that NCP is not a CLI replacement, nor does it push configurations, but it can provide insights and facilitate data-driven decisions. The platform's self-hosted nature with GPU requirements (like NVIDIA A100 or H100) ensures data privacy and offers a quicker ROI by automating tedious tasks, freeing up valuable engineering time.

Presented by Thomas Scheibe, Chief Product Officer, and Madhu Paluru, Director, AI Engineering. Recorded live at Networking Field Day 38 in Silicon Valley on July 9, 2025. Watch the entire presentation at https://techfieldday.com/appearance/aviz-networks-presents-at-networking-field-day-38/ or visit https://techfieldday.com/event/nfd38/ or https://AvizNetworks.com for more information.