One Platform. Four Solutions and Counting. Zero Lock-In.
Your data. Turned into answers and action.
Your network and AI-infrastructure data is already spread across switches, controllers, observability stacks, and clouds. AAAP turns it into answers and action, private, vendor-neutral, and programmable, unlike single-vendor copilots or SaaS tools that ship your telemetry to someone else's model.
One platform, four solutions today
Network Copilot
Enterprise NetOps
AI NOC
AI Factories & AI Grids
SONiC AI Assistant
Open-Networking Support
Custom Agents
Built By Your Team
Ingest. Reason. Act.
Ingest
Connectors stream config, telemetry, flows, logs, and tickets into a governed, project-scoped data layer, indexed for retrieval.
Reason
A private LLM reasons over your data, grounded by retrieval, not guessing, to answer questions and find root cause.
Act
Agents generate reports, open tickets, and run checks, every state-changing step routes through your approval gates.
Interactive and scheduled background agents automate recurring operational tasks, while state-changing actions remain subject to human approval.
A platform you shape and govern
Competitors bolt a chatbot onto a fixed product. AAAP gives you customizability without code, guardrails on every action, and engineering most copilots skip.
Console
One central place to manage the platform, dashboards for usage and health.
What you manage here
- Resources: data connectors, devices, knowledge base assets.
- AI: LLMs, agents, background jobs, reasoning settings.
- People: users, roles, authentication, organization preferences.
- Monitoring: CPU, memory, disk, GPU utilization, and diagnostic logs.
Conversational UX
Ask in plain language, get answers that stream in real time.
What's included
- In-line artifacts: charts and sortable tables, exportable to PDF, CSV, or Excel.
- Persistent history: pick up any past conversation where you left off.
- Feedback loop: prompt suggestions and thumbs-up/down keep it sharp.
Efficiency by Design
Optimizations most copilots skip, engineered to keep private-LLM answers fast and grounded on your own GPUs.
What's under the hood
- Efficient retrieval: grounds answers in relevant operational evidence.
- Prompt preprocessing: interprets intent and prepares execution plans.
- Token Compression: reduces LLM context consumption, with model-level savings visibility.
- LLM Management: configure models and reasoning effort to optimize performance.
Accessibility & Global Reach
Built to be usable by every operator, in every region, a requirement for public-sector and enterprise deployments alike.
What's included
- Voicebox: hands-free voice interaction with speech-to-text and text-to-speech.
- Multilingual operations: voice and conversational interaction across 18 languages.
- Rich visualizations: interactive charts, heatmaps, network graphs, and dashboards.
Enterprise Controls
Use AI in the most regulated, compliance-demanding environments, control lives at every layer.
What's enforced
- Identity: enterprise directory authentication and role-based access.
- Access: project- and organization-scoped resource controls.
- Guardrails: human approval for state-changing operations.
- Agent Trace: visibility into agents, tools, execution steps, and sources.
- Audit Logging: searchable activity records with export and SIEM forwarding.
Projects
Scoped workspaces of connectors, files, and members, each team gets exactly the sources it needs.
How teams use it
- Scoped resources: knowledge base documents, LLMs, and data connectors per project.
- Shared conversations: members see the same live chat and inject their own prompts.
- No more copy-paste: one context, not isolated threads per person.
29+ Data Connectors. Open-Ended Extensibility.
Connect networking, cloud, security, observability, ITSM, and AI infrastructure through built-in integrations and MCP servers—without replacing your existing tools or migrating operational data.
AI Infrastructure & GPU
5 connectorsvLLM, NVIDIA Dynamo, DCGM, Base Command Manager, Run:ai
Observability & Infrastructure
9 connectorsCisco Nexus/Catalyst, NVIDIA NetQ, Forward, IP Fabric, SolarWinds, NetBox, Zabbix, Prometheus
Collaboration & ITSM
5 connectorsServiceNow, Zendesk, DevRev, Slack, Rocket.Chat
Log & Metrics
5 connectorsSplunk, ELK, Loki, InfluxDB, Syslog
Cloud & Storage
3 connectorsAmazon S3, Azure Blob, Google Cloud Storage
Extend with MCP
No-codeAttach any external MCP server, the catalog becomes open-ended
See the full connector catalog (29+)
| Category | Connector | What it's for |
|---|---|---|
| Aviz Native | Aviz ONES | Multi-vendor fabric management and telemetry |
| Cloud & Storage | Amazon S3 | Pull configs, logs, and datasets from AWS S3 |
| Cloud & Storage | Azure Blob Storage | Ingest documents, logs, and dumps from Azure |
| Cloud & Storage | Google Cloud Storage | Read files, logs, and datasets from GCP buckets |
| Log & Metrics | Splunk | Query logs and operational metrics on demand |
| Log & Metrics | ELK / Elasticsearch | Search logs and events in the ELK stack |
| Log & Metrics | Loki (syslog only) | Retrieve syslog from Grafana Loki |
| Log & Metrics | InfluxDB (flow logs only) | Pull network flow logs from InfluxDB |
| Log & Metrics | Syslog | Forward platform audit events to SIEM systems |
| Data Warehousing | Snowflake | Query operational and business datasets |
| Data Warehousing | Google BigQuery | Run large-scale analytics queries |
| Observability | Cisco Nexus Dashboard | Data-center fabric operations and telemetry |
| Observability | Cisco Catalyst Center | Campus and enterprise network management |
| Observability | NVIDIA NetQ | Fabric validation for SONiC and Cumulus networks |
| Observability | Forward Enterprise | Network digital twin for path and change analysis |
| Observability | IP Fabric | Automated network assurance and topology snapshots |
| Observability | SolarWinds | Network performance monitoring and alerts |
| Observability | NetBox | Source-of-truth IPAM and DCIM inventory |
| Observability | Zabbix | Open-source monitoring of network and infra health |
| Observability | Prometheus | Natural-language metrics queries and visualizations |
| AI Infra & GPU | vLLM | Inference telemetry, latency, throughput, KV-cache, tokens |
| AI Infra & GPU | NVIDIA Dynamo | Inference serving-framework metrics and workload signals |
| AI Infra & GPU | NVIDIA DCGM | Per-GPU telemetry, utilization, VRAM, power, thermals, XID |
| AI Infra & GPU | NVIDIA Base Command Mgr | Cluster and compute-node health across your GPU fleet |
| AI Infra & GPU | NVIDIA Run:ai | GPU workload orchestration, projects, quotas, jobs |
| Security | Palo Alto NGFW | Next-gen firewall policy, rules, security telemetry |
| Security | Fortinet FortiGate | Firewall configuration and security event data |
| ITSM | ServiceNow | Incidents, changes, and ticket workflows |
| ITSM | Zendesk | Support tickets and customer service records |
| ITSM | DevRev | Access ticketing and support workflows through MCP |
| ITSM | Slack | Team messaging context and notifications |
| ITSM | Rocket.Chat | Self-hosted team messaging and collaboration |
| Extend with MCP | Generic MCP connector | Attach any external MCP server, no code required |
New MCP integration options: Cisco Catalyst Center, Cisco Nexus Dashboard, and Aviz ONES.
Full, current list: Data Connectors documentation.
KNOWLEDGE
RAG Ingestion
Built-in ingestion of runbooks, design docs, PDFs, and datasets makes your agents aware of your processes, schedules, and operational context, no code required to enforce or extend it.
COLLECTION
Data Ingestion
Connectors pull telemetry over SNMP/gNMI; collectors gather syslogs and ingest offline configs and dumps for air-gapped environments a live connection can't reach.
Build Your Own Agents. In Python. It's Your IP.
The Aviz Agent SDK (ncp-sdk on PyPI) is what makes the platform programmable, not a walled garden of prepackaged use cases, but a real SDK your engineers write against.
- Author in Python: define an Agent, wrap your own functions as tools with a decorator, configure models and memory in a few lines.
- Compose multi-agent workflows: delegate to sub-agents for multi-step, cross-domain tasks.
- Consume MCP-native tools: connect to MCP servers over stdio, SSE, and streamable HTTP, multiple servers concurrently.
- Ship with a CLI: initialize, validate, package, deploy, and test agents in a local playground before production.
- Publish through the UI: role-based governance so what one engineer builds, every operator can safely run.
- Agent SDK v1.4.0: Build more contextual, collaborative agents with agent memory, handoffs, and reusable skills for complex operational workflows.
Open reference agents on GitHub
Capture a senior engineer's playbook once, let every operator run it. Browse the working code at github.com/AvizNetworks/ncp-sdk-agents.
code.aviznetworks.com, a dedicated home for Aviz's open reference agents and SDK examples.Your Data. Your GPUs. Your Agents. Your IP.
Engineered for teams that cannot, or will not, send operational data to someone else's cloud. The reasoning layer is a private, self-hosted LLM on your own GPUs. The platform is LLM-agnostic: run the model you choose, where you choose.
On-prem
Docker Compose for PoC, Kubernetes for production, or a virtual appliance. Your data center, your rules.
Air-gapped
Fully disconnected operation with an offline model-delivery path, for classified and sovereign facilities.
Hybrid
Run the platform in your environment while Aviz hosts the LLM, faster value when GPUs are scarce.
Bring your own LLM
Point the platform at the model endpoint you already operate. No AI tax, no forced model.
Governed autonomy. Not blind faith.
Security is the reason teams choose a platform they own, it runs through every layer of AAAP.
Private LLM, your GPUs
Prompts and telemetry never leave your environment or train anyone else's model.
Enterprise identity
Directory-based SSO, certificate login, RBAC, project-level data isolation.
Guardrails on action
Human approval gates on anything state-changing, with full audit visibility.
Compliance posture
Engineered to meet the controls your auditors care about, actively hardening further.
Agent Trace & Explainability
See which agents, tools, and data sources contributed to every response, with execution status and timing for greater transparency and trust.
Enterprise Activity & Audit Logs
Track agent, LLM, and tool activity through searchable logs, export records in CSV, JSON, or NDJSON, and forward audit events to external SIEM platforms.
One foundation. Purpose-built solutions.
Everything below shares the same connectors, governance, and private-LLM core.
| Solution | What it is | Status |
|---|---|---|
| Network Copilot™ | Conversational NetOps for multi-vendor troubleshooting, observability, security audits, and operational automation | Available today |
| AI NOC | Conversational AI infrastructure operations across GPU, inference, cluster, and network fabrics | Early access |
| SONiC AI Assistant | SONiC support, reinvented with AI, delivered as an Aviz cloud service | Coming soon |
| Your custom agents | Workflows your team authors with the Aviz Agent SDK and publishes with governance | Build today |
| GRC Copilot · Agent Studio | Governance/risk/compliance and agent-building solutions | Roadmap |
You license a solution, not the platform.
AAAP itself isn't sold standalone, you license the solution built on it (Network Copilot, AI NOC, SONiC AI Assistant, or your own custom agents), and everything you deploy shares the same connectors, governance, and private-LLM core.
Network Copilot
Three tiers by network size, Base (<100 devices), Advanced (100–1,000), Performance (1,000+). Same use cases and connectors at every tier.
AI NOC
Early access, licensed per engagement, not yet tiered or metered.
SONiC AI Assistant
Launching through early access, packaging and pricing shared with participants first.
Custom Agents
Build on any tier with the Aviz Agent SDK, or add an agent pack scoped to your use case.
Proven, and backed
Best in Show
ONUG AI Networking Summit, Fall 2025
Silver Winner
2025 Merit Awards, Best AI Startup
Common questions
The most-asked questions from customers evaluating AAAP for enterprise and sovereign deployments.
Does my data leave my environment?
No. Deploy on-prem or fully air-gapped with a private LLM served on your own GPUs. Prompts and telemetry stay inside your boundary and never train an external model.
Am I locked into Aviz, or into any vendor?
No. The platform is vendor-neutral across hardware and NOS, LLM-agnostic across models, and programmable through an open Python SDK. Agents you build are your IP.
Do I have to use the Aviz LLM?
No. Run a private model on your own GPUs, use the hybrid option with an Aviz-hosted LLM, or bring your own LLM endpoint, and change your mind later.
Can it run fully air-gapped?
Yes. Air-gapped deployment is a first-class mode, including an offline model-delivery path, designed for classified, sovereign, and disconnected environments.
Is the platform certified for my compliance framework?
Aviz does not claim certifications. The platform is engineered to help you meet the controls enterprise and government frameworks require, and is actively hardening to meet them. Ask us how it maps to your environment.
Can my team build its own agents?
Yes, that's the point of the platform. Author agents in Python with the Aviz Agent SDK, test in a local playground, and publish through the UI with role-based governance.
How is it priced?
You license a solution built on the platform, Network Copilot, AI NOC, SONiC AI Assistant, or custom agents, not the platform on its own. Pricing is on value, not metered per device, flow, or agent run.