AI Agents for Network Automation

Vertical AI Agents for Networks

IT networks are the backbone of modern infrastructure, but their growing complexity introduces challenges in incidence response, security, compliance, and cost. Addressing these challenges often requires stitching together insights from disparate systems, each using different tools for monitoring, observability, alerting, and more. These tools, built in isolation, force humans to bridge gaps in workflows, driving up operational complexity and costs.

In large-scale service provider, data center, and enterprise networks, the impact is significant: high SLA penalties, increased NOC expenses, and reduced productivity.

Automating Network Operations with AI Engineers

Nanites AI is addressing these challenges by building AI network engineers designed to automate a wide range of network operations tasks, including incident management, security engineering, and compliance.

Our first step in this ambitious journey is solving the most pressing network operations problem: troubleshooting alerts. Our system integrates with monitoring, observability, and NMS tools, consuming alerts from platforms network engineers rely on daily. While our system is already highly effective, we are working toward achieving over 90% automation for network alerts and incidents, minimizing the need for human intervention. Developed in collaboration with our customers, Nanites AI is built to independently and accurately perform network engineering tasks with lightning speed and high precision.

Deep Understanding of Network Systems and Tools

Network engineers rely on a variety of tools, including monitoring, configuration management, and observability tools, to diagnose and resolve issues. AI systems that work with these tools, using them as a human engineer would, must deeply understand:

• Interconnected systems and dependencies: AI must accurately synthesize information about devices, configurations, and traffic flows from monitoring systems, configuration management tools, and log analysis platforms.
  

• Handling scale and dynamic data: Networks generate vast amounts of constantly changing data. An effective AI system must parse only the most relevant data to minimize cost, latency, and noise.
  

To ensure precise and efficient data retrieval, our system employs proprietary prompt engineering techniques and advanced RAG to extract the most relevant network telemetry datasets for each task. When relationships between data points are deemed too complex or challenging to interpret, the AI dynamically chunks and then embeds the data into our vector store, enabling a deeper understanding of semantic relationships. Additionally, our system leverages knowledge graphs to map and contextualize intricate relationships between devices, configurations, and traffic flows, providing a structured foundation for advanced AI reasoning. This approach significantly increases accuracy while reducing data volume, compute costs, and context window limitations.

Dynamic Network Automation with AI

Dynamic networks require AI systems capable of real-time troubleshooting and remediation. Nanites AI excels in addressing challenges such as:

• Novel incidents: Networks often experience unique issues where pattern-matching falls short. Our AI system effectively handles both repeat and novel incidents.
  

• Causality determination: It removes noise from unrelated, coincidental events common in large-scale networks, enabling accurate root cause analysis.
  

• Continuous learning: Every network has distinct operational behaviors. Our agentic platform learns on the job, collaborating with humans and applying new knowledge across tasks and contexts.
  

• Executing complex actions: Beyond analysis, our system performs advanced tasks like executing CLI commands and device configurations.
  

Our platform consists of composable agents, each with specialized capabilities and human-like tool usage. These agents dynamically interact with APIs and physical devices, leveraging memory and historical data to enhance decision-making through rich contextual insights. They continuously learn and refine their processes to align with and achieve specified KPIs. Nanites AI includes specialized agents, such as protocol agents for BGP, VXLAN, DOCSIS, SNMP, gNMI, RADIUS, and more, enabling our system to efficiently manage and troubleshoot networks across diverse environments and use cases.

Our system integrates with top networking vendors, including Cisco, Juniper, Arista, Palo Alto Networks, and others, providing compatibility and support for diverse network environments. Additionally, we support open-source Linux-based network operating systems such as SONiC and OpenWrt, enabling flexibility for organizations leveraging disaggregated networking. The system uses infrastructure tools like Grafana, Prometheus, DataDog, PyATS, and many others, just as a human network engineer would do. Our cognitive architectures and flows enable the AI to establish direct, bidirectional communication with network devices, allowing real-time back-and-forth interactions. This capability ensures the system can efficiently gather data, execute commands, and validate results in parallel or sequentially, depending on the task at hand. By design, our platform is highly extensible, allowing integration with additional tools, protocols, and frameworks to meet evolving network requirements. By continually learning, our agents enhance their ability to handle novel incidents and apply their knowledge to new scenarios.

Data Privacy with Local Open-Source LLM Deployments

Our system supports local deployments of open-source LLMs, ensuring that sensitive data remains secure and never exposed to external APIs or commercial LLMs. This approach allows our system to maintain compliance with data privacy regulations while leveraging the power of AI to perform advanced troubleshooting and network management tasks. By keeping proprietary data within the customer’s infrastructure, we deliver cutting-edge AI capabilities without compromising security. This enables faster deployment timelines and accelerates time to value.

Distributed Deployment for Reduced Latency

Nanites AI is designed as a distributed system capable of being deployed on or near edge devices to minimize latency and optimize performance. By processing data closer to its source, our system ensures real-time troubleshooting and rapid decision-making, which is critical for time-sensitive network operations. To handle the complexity of modern networks, our platform employs a hierarchical architecture with specialized top-level agents and support agents working collaboratively. This design ensures efficient task execution, scalability, and alignment with a dedicated source of truth agent that maintains consistency across network operations.

The Future of Network Automation

While we are off to a great start, there are still a lot of hard and interesting problems to solve. We need to keep fine tuning specialized models and test across many more scenarios to effectively operate an increasing number of tasks and tools at scale, each of which brings its unique challenges and capabilities. To successfully build this system, we marshaled resources from multiple domains, combining deep AI expertise, networking knowledge, and operational experience. A critical part of our success was ensuring deterministic performance in a domain that inherently demands it, overcoming the challenges posed by the non-deterministic nature of AI. Through advanced prompt engineering, strict guardrails, rigorous validation, iterative refinement, and other carefully designed approaches, we delivered consistent and reliable outputs tailored to the high demands of real-world network operations. Our agentic AI will continue improving its reasoning abilities for increasingly complex tasks through better planning, learning, and orchestration. Finally, our customers are facing problems not just in incident troubleshooting, but a variety of network operations areas that need to be built out upon the same platform. The future is in self-healing, adaptive, and intelligent networks, where intent-based networking ensures seamless alignment between business objectives and network operations, transforming how businesses operate.