AI & Local Model Integration
Relevant source files
The following files were used as context for generating this wiki page:
Overview
The AI Network Operations Platform serves as a self-hosted "AI Network Engineer" designed to assist network teams with infrastructure discovery, troubleshooting, and management. By integrating Large Language Models (LLMs) directly into the operational workflow, the platform allows engineers to interact with their network using plain English commands rather than complex CLI syntax.
Sources: CLAUDE.md:1-12, README.md:1-5
Key Problems Solved
- Multi-Vendor Complexity: Simplifies operations across different vendors like Cisco, Juniper, Arista, and Fortinet by providing a unified interface.
- Data Silos: Consolidates information from various sources (SNMP, SSH, APIs) into a single topology and knowledge graph.
- Audit and Security: Ensures every AI-suggested change is reviewed by a human and every decision is logged for auditing.
Design Philosophy
The platform follows a "local-first" approach. It is designed to be self-hosted within a company's own data center or private cloud to ensure that sensitive network configurations and security policies never leave the internal environment. This is achieved primarily through the use of Ollama, which runs AI models locally.
Sources: CLAUDE.md:30-38, deploy/docker/README.md:94-100
Supported Vendors and Protocols
The platform uses a standardized connector system for each vendor. This ensures that whether the platform is talking to a Cisco router or an Infoblox DDI appliance, the data is normalized into a consistent format.
| Category | Supported Vendors / Protocols |
|---|---|
| Routing & Switching | Cisco (IOS, IOS-XE, NX-OS), Juniper JunOS, Arista EOS |
| Security | Palo Alto PAN-OS, Fortinet FortiOS, F5 BIG-IP |
| DDI (DNS/DHCP/IPAM) | BlueCat, Infoblox, AWS Route53 |
| Cloud & Virtualization | AWS, Azure, VMware |
| Protocols | SNMP, SSH, LLDP, CDP, BGP, OSPF |
Sources: CLAUDE.md:14-28, README.md:18-22, backend/app/agents/documentation/agent.py:1-15
Specialist Agents
The platform routes user questions to "Specialist Agents," which can be thought of as specialized team members who excel at specific tasks.
- Master Architect: The supervisor that listens to your request and decides which specialist to assign the task to.
- Troubleshooting Agent: Diagnoses why something is broken (e.g., "Why is BGP down on edge-1?").
- Discovery Agent: Scans the network to find and list devices and neighbors.
- DDI Agent: Manages DNS and DHCP records and troubleshoots name resolution.
- Security Agent: Audits firewall policies for overly-permissive rules or security gaps.
- Documentation Agent: Automatically generates network diagrams, inventories, and runbooks.
Sources: CLAUDE.md:40-51, backend/app/llm/prompts/init.py:190-250
Architecture
The platform is composed of seven core components that work together to provide the AI-assisted experience. Each component is responsible for a specific part of the network operations lifecycle.
Sources: README.md:10-17, deploy/docker/docker-compose.yml:21-170
Platform Components
- Frontend: The operations console where you chat with the AI, view the topology map, and approve changes.
- API Server: The brain that handles your requests, checks your permissions, and manages the AI agent sessions.
- Worker: The background engine that runs discovery jobs, takes configuration backups, and captures packets without blocking your current session.
- Postgres: The system of record that stores device lists, user accounts, and historical audit logs.
- Neo4j: A living topology map that tracks how every device, interface, subnet, DNS zone, and application connects to everything else.
- Redis: A fast message broker that coordinates tasks between the API and the Worker.
- Ollama: The local AI engine that performs model inference, allowing the platform to "think" without using the public internet.
Deployment Options
The platform can be deployed in two primary ways depending on the environment requirements.
- Docker Compose: Best for laboratory environments or small teams. It runs the entire stack on a single server using a single command.
- Kubernetes (Helm): Designed for production use. This setup provides high availability, better resource management, and integrates with enterprise security tools like
cert-managerfor encrypted traffic.
Sources: deploy/docker/README.md:1-10, deploy/kubernetes/netops/README.md:1-15
Change Approval Workflow
Every change requested through the AI follows a strict lifecycle to prevent unauthorized or accidental modifications to the network.
- Draft: The specialist agent prepares a "Change Request" (CR) based on your English command.
- Pending Approval: The CR appears in the dashboard for a second human engineer to review.
- Approved: A human clicks "Approve" after verifying the logic and the pre-change backup.
- Executing: The Automation Agent takes the approved CR and applies it to the device.
- Completed/Rolled Back: If the change fails, the platform can use the backup to restore the previous state.
Sources: README.md:2-5, backend/app/agents/ddi/agent.py:16-25, backend/app/agents/security/agent.py:18-25
Diagrams
System Overview
This diagram shows how the user interacts with the platform and how the platform connects to your infrastructure and AI providers.
Visual representation of user-to-platform interaction and backend connectivity to various vendor families and security services.
Sources: CLAUDE.md:14-28, README.md:10-22
Agent Routing
The Master Architect acts as a dispatcher, ensuring your question reaches the right specialist.
The workflow from initial user query to routing by the Master Architect, leading to either immediate data retrieval or an approval requirement for network changes.
Sources: backend/app/llm/prompts/init.py:190-250, backend/app/agents/documentation/agent.py:46-70
Change Approval Flow
The lifecycle of a network change, ensuring human-in-the-loop verification.
The sequential steps involved in moving from an AI-drafted change to human approval and final execution by the Automation Agent.
Sources: README.md:1-8, backend/app/agents/ddi/agent.py:16-25
Topology Data Model
What the platform "knows" about your network and how those items relate.
A mapping of the relationships stored in the Neo4j knowledge graph, showing how physical hardware connects to logical network constructs like subnets and DNS records.
Sources: CLAUDE.md:53-61, backend/app/agents/documentation/agent.py:30-40
Deployment Options
Comparison of the two primary ways to run the platform.
High-level view comparing the single-node simplicity of Docker Compose with the multi-node resilience and scaling of Kubernetes.
Sources: deploy/docker/README.md:1-15, deploy/kubernetes/netops/README.md:1-25
Conclusion
The AI & Local Model Integration allows network engineering teams to transition from manual, CLI-heavy workflows to an intent-based system. By utilizing specialist agents and a local-first AI architecture, the platform provides a secure way to leverage LLMs for complex network tasks while maintaining total human control over infrastructure changes.