AI Agents

Agent Skills

Teach NetBrain agents your network’s unique language. Custom circuit IDs, internal WAN naming conventions, tribal procedures, and site-specific troubleshooting logic encoded directly into NetBrain’s AI layer, with no model retraining required. Skills are the enabler that makes every NetBrain agent operate precisely on your specific environment rather than a generic network abstraction. 

What Skills do 

• Encode site-specific naming, circuit IDs, and escalation paths so the agent speaks your network’s language. 
• Capture senior-engineer playbooks (the steps an expert would take for a specific class of incident) as deterministic procedures. 
• Compound over time: every Network Intent encoded and every solved past problem makes Skills more reliable. 
• No model retraining required. Skills sit on top of the foundation model. 

Why Skills matter for accuracy 

Enterprise AI needs to be 99.99% accurate, not 80%. The difference is the part of the network that doesn’t appear in any vendor’s training data: your circuit naming, your WAN convention, the senior engineer’s mental rule that says “if HSRP fails on these three sites, check this upstream first.” Skills turn that institutional knowledge into agent behavior. 

How Skills work 

• Read once at the start of every agent execution — before any reasoning begins. 
• Matched by description: the agent reads skill descriptions and selects the ones relevant to the current task. 
• Act as guardrails, not scripts. Short skills that nudge the agent in the right direction outperform long prescriptive ones. The agent handles the reasoning; the skill defines the boundaries. 

AI Document

AI-generated network documentation, automatically, from live operational data. NetBrain’s AI agents summarize network activities, translate operational findings into narrative documentation, and export Observability Dashboard content directly to Word or PDF. The agent reads the context-aware digital twin and the activity log, then writes the documentation the team would have written if they’d had the time. 

• Auto-generate documentation for network diagrams, device configurations, and operational processes. 
• Export in editable formats including Microsoft Word and Microsoft Visio. 
• Include design and inventory data, diagnosis details, full configuration files, and routing tables. 
• Translate Observability Dashboard snapshots into a narrative report ready for leadership. 
• Sharable, audit-ready, current. 

Deep Diagnosis Agent

Fully autonomous, end-to-end root cause analysis. Triggered directly from a ServiceNow ticket, an alert, or a webhook, the Deep Diagnosis Agent runs a complete root cause analysis without human hand-holding. It navigates the network topology, executes relevant tests, synthesizes findings, and delivers a structured diagnosis ready for engineer review or automated action. 

How the agent decides what to do 

At the start of every session, the agent reads the descriptions of all enabled MCP tools. It matches those descriptions against the prompt using the underlying language model — no hardcoding, no routing rules. The right tools are called automatically based on what the task requires. 

From event to diagnosis 

• ITSM ticket opens or webhook fires. Triggered Automation Framework activates the Agent. 
• The Agent reads available tools, pulls incident context, queries connected monitoring systems, and runs the relevant diagnostic intents. 
• Output appears in NetBrain Incidents, the Incident Portal, and the originating ticket, with a Summary and View Full Diagnosis link. 
• Engineer reviews and either acts on the recommended remediation or escalates. 

Audit trails preserved end to end. No unsupervised remediation: action requires human review or pre-approved automation. 

Runbook Companion Agent

Conversational intelligence over your existing runbook library. Engineers ask natural-language questions about runbook results, explore alternative remediation paths, and update runbooks with lessons learned from live incidents. Institutional knowledge becomes a living resource rather than a static document. 

Change impact analysis, before you push 

Before any change executes, the Companion reviews live configuration across every in-scope device and issues a push or do-not-push verdict per device — with the reason. It identifies which applications would be affected, drawing on both path data and configuration context. Engineers see a clear recommendation for every device before a single command is sent. 

Post-change verification at scale 

After execution, the Companion verifies success for every device in the change scope — condensed into a single line per device. What would otherwise be an unreadable diff across hundreds of devices becomes a concise, actionable summary. 

Conversational intelligence over your runbooks 

• Ask natural-language questions about runbook results and get direct answers, not raw output. 
• When a runbook doesn’t resolve the issue, the Companion recommends the next automation to apply. 
• Generates concise summaries with initial conclusions — shareable without cleanup. 

Example prompts 

• “Summarize the alerts in this runbook.” 
• “Show batch ping results as a table.” 
• “Why is voice quality poor between these endpoints?” 
• “What changed on these devices in the last 2 hours?” 

Share context across the team 

• Pin frequent prompts as Goals so the team starts from a shared context, not a blank chat. 
• Share chat threads across teams — runbook intelligence is no longer locked in one engineer’s session. 

MCP Bidirectional Integration

The connective tissue of the agentic platform. Model Context Protocol (MCP) support exposes NetBrain’s network context to external AI tools and pulls signals from APM, observability, and ITSM systems into NetBrain agents.