Agents build the network.
Engineers run the agents.

NetPilot's agent designs, builds, and validates a real multi-vendor network from your prompt — you own the intent, the judgment, and the sign-off.

Try It Free
app.netpilot.io

See It in Action

Watch the agent build a complete multi-vendor network lab from a single conversation.

What the agent does, step by step

Five steps from your prompt to a validated network — the agent runs each one, you review the result.

Step 1

Designs the topology from your prompt

Tell the agent what you need — OSPF multi-area, MPLS L3VPN, BGP peering, VXLAN EVPN, leaf-spine fabric, or any combination. It designs the full topology including IP addressing, interface mapping, and protocol relationships. No diagrams or templates required.

The NetPilot agent designing a network topology with 14 nodes from a chat conversation
The NetPilot agent auto-generating OSPF multi-area configuration for Cisco IOS devices
Step 2

Writes vendor-specific configs

The agent produces production-grade startup configurations for every device — Nokia SR Linux built in; Cisco IOL, Juniper cRPD, Arista cEOS, Palo Alto PAN-OS, and Fortinet FortiGate via BYOI (you supply the commercial image). Each config uses the correct vendor syntax, interface naming, and protocol semantics. No manual editing needed.

Step 3

Builds the lab on real device CLIs

Not a simulation. The agent deploys actual network OS instances to isolated cloud VMs. SSH into any device and run show commands, configure interfaces, debug protocols — exactly like production gear. Click any node in the topology to open a terminal.

NetPilot App - AI-powered network lab showing MPLS L3VPN configuration with live topology visualization
SSH: PE1 (1.1.1.1)
# Click any device to SSH
PE1# show mpls ldp neighbor
Peer LDP Ident: 3.3.3.3:0
TCP connection: 3.3.3.3:646
State: Oper
PE1# show ip route vrf CUSTOMER_A
B 192.168.2.0/30 [200/0] via 2.2.2.2
C 192.168.1.0/30 directly connected
PE1# _
NetPilot verification results showing all-green checks for OSPF, MPLS LDP, iBGP, and end-to-end routing
Step 4

Validates connectivity — you review it

After deployment, the agent checks routing tables, protocol adjacencies, end-to-end reachability, and VPN paths — and hands you the evidence instead of making you type every show command. If something fails, it diagnoses the issue and offers fixes for you to approve.

Step 5

Generates traffic to test performance

Validate throughput, latency, and failover behavior using built-in iperf3 traffic generation. Test across MPLS paths, VPN tunnels, and redundant links before anything touches production.

NetPilot iperf3 TCP throughput test across MPLS L3VPN network

You run the agent. The agent runs the build.

NetPilot keeps the engineer in command. The agent takes the repetitive work — design, config, deploy, validate — and hands every result back to you to verify and own.

You direct

Describe the change, topology, or what-if in plain English — or import sanitized production configs. You set the intent; you decide what gets built.

The agent does the toil

It designs the topology, writes per-vendor configs, builds a real multi-vendor lab in ~2 minutes, and runs first-pass validation — the repetitive work, done for you.

You verify & sign off

SSH into any device, check the real CLI, and take the proven change to production yourself. The agent proposes; the engineer decides and signs off.

The work that stays yours

The agent ends the toil — typing configs, wiring topologies, running show commands. Your role moves up, from operator to orchestrator: generation is solved; judgment, direction, and ownership are the craft.

Intent & architecture

Decide what the network should do — the design, the trade-offs, the business and security requirements the agent builds to.

Verification & judgment

Generation is solved; judgment isn't. Review the agent's design and validation evidence on the real CLI, and catch what it missed.

Change ownership

Approve the change, own the blast radius, decide what's safe for production. The agent proposes; you're accountable.

Orchestrating agents

Direct several agents in parallel — build, stage, validate, draft the rollback — and set the guardrails: what runs autonomously, what needs your sign-off.

The hard 20%

Novel designs, gnarly multi-vendor interop, ambiguous requirements, the debugging the agent can't finish — the work only you can do.

“Should we,” not just “can we”

Risk tolerance, compliance, and the judgment calls AI shouldn't own.

What makes the agent different

Not a chatbot that writes configs, and not a black box — an agent your engineers direct, with the real CLI as the trust layer.

Agentic, not a chatbot

Plans, builds, deploys, and validates a real multi-vendor lab end to end — then hands you the CLI to verify.

The alternative

A chatbot writes config text you still wire up, deploy, and debug yourself.

Engineers run the agents

You own the intent and the sign-off; the agent does the build-and-validate toil across every device.

The alternative

Hand-build each device and hand-check every change — the toil stays yours.

Real CLIs are the trust layer

SSH into the real NOS and confirm routing and protocol state — the agent's work is always verifiable, never a black box.

The alternative

Trust a model's text output with no real device state to check it against.

Multi-vendor from one prompt

One prompt yields correct per-vendor configs across Cisco, Juniper, Arista, and Nokia at once — no syntax lookups.

The alternative

Translate every change by hand, vendor by vendor, then wire it together.

What you can have the agent build

From a quick OSPF check to a full EVPN fabric — describe it and the agent designs, configures, deploys, and validates it on real multi-vendor network OSes.

Build a topology from a prompt

Describe the network in plain English and the agent designs nodes, links, IP addressing, and protocol relationships — agentic network engineering, no diagrams or templates.

Generate multi-vendor configs

Ask for "eBGP between the Cisco edge and the Juniper transit" — the agent writes correct IOS, Junos, EOS, and SR Linux syntax for every device at once.

Test a change before production

Import sanitized configs, let the agent build a digital twin, then run the change against it and review the validation evidence before it touches prod.

Stand up OSPF, BGP & MPLS

Multi-area OSPF with summarization, eBGP/iBGP with route reflectors and route-maps, MPLS L3VPN with VPNv4 and customer VRFs — built and adjacency-checked.

Spin up a VXLAN/EVPN fabric

A leaf-spine data-center fabric with an eBGP underlay and EVPN overlay — the agent wires the fabric and confirms the overlay comes up.

Stage firewall & security policy

Palo Alto or Fortinet zones, security rules, and NAT between Cisco segments (via BYOI) — staged and reachability-tested in an isolated lab.

Validate connectivity & protocol state

After the build the agent checks routing tables, adjacencies, end-to-end reachability, and VPN paths, then hands you the report — you SSH in and verify on the real CLI.

Verify multi-vendor interop

Put Nokia SR Linux (built in) alongside Cisco, Juniper, and Arista (via BYOI) in one topology and have the agent reproduce and confirm cross-vendor behavior before you commit to it.

More built-in capabilities

Beyond the core loop — everything else the agent brings to your lab environment.

Digital Twin Import

Import your production network configurations to create an exact virtual replica. Test changes in a sandbox that mirrors your live environment — before they touch prod.

Bring Your Own Images

Upload your own vendor device images (BYOI) for commercial NOSes. The agent configures them within your topologies just like the built-in images.

MCP-Native Platform

Built on an MCP-native agent that connects to the tools your team already runs (inventory, config repos, ticketing). Under the enterprise plan, NetPilot's team builds the MCP/API integration into your stack with you.

Multi-Vendor Native

The agent understands each vendor's CLI syntax, interface naming, and protocol semantics natively — so multi-vendor interop is a prompt, not a week of syntax lookups.

Isolated Cloud Environments

Every lab runs in a dedicated cloud VM. No shared infrastructure, no conflicts between labs. Test sensitive scenarios without exposing production.

Template Library

Save any lab as a reusable template. Duplicate and modify to create variations — perfect for repeatable team demos, runbooks, or onboarding environments.

Packet Tracer Export

Export complete topologies as .pkt files for Cisco Packet Tracer — a portable interchange format for sharing or archiving a lab outside NetPilot.

Works with any vendor

NetPilot is built on ContainerLab. Nokia SR Linux, FRR, and Linux are built in; commercial NOSes run via bring-your-own-image. If a device image runs in a container, the agent can deploy and configure it.

Nokia SR Linuxbuilt inFRRbuilt inLinuxbuilt inCisco IOLBYOIJuniper cRPDBYOIArista cEOSBYOIPalo Alto PAN-OSBYOIFortinet FortiGateBYOI+ more on request

Frequently Asked Questions

How an agent your engineers run fits the way your team already works

It's more useful than that label suggests — and less threatening. NetPilot is an agentic AI your engineers run: you describe a change, topology, or what-if in plain English, and the agent designs the topology, generates per-vendor configs, builds a real multi-vendor lab, and runs first-pass validation — then you SSH in, verify on the real CLI, and sign off. The engineer stays in command; the agent does the toil. Think 'an agent between you and the network,' not a replacement.
Agentic. A chatbot answers and writes config text you still have to wire up, deploy, and verify yourself. NetPilot plans, acts, observes, and iterates: it designs the topology, generates the configs, deploys a real multi-vendor lab on cloud-hosted ContainerLab, validates connectivity and protocol state on live CLIs, and fixes what's broken — closing the loop. You direct it conversationally ("add an Arista spine, move OSPF to area 0.0.0.1") and it updates every affected device.
No — engineers run the agents. NetPilot takes the repetitive toil (build the lab, write per-vendor configs, deploy, run first-pass validation) so engineers move up to design intent, judgment, and owning the change. The agent proposes; you decide, verify on the real CLI, and sign off. Nothing reaches production without an engineer in command.
No. NetPilot builds an isolated lab on cloud-hosted ContainerLab — it never connects to or changes your production devices. You stage changes, reproduce issues, and run automation against the lab's real NOS CLIs, then take what you've proven back to production yourself. For air-gapped or compliance teams, an on-prem deployment is available on the enterprise plan.
9+ network OSes and growing. Nokia SR Linux, FRR, and Linux are built in; commercial NOSes — Cisco IOL, Juniper cRPD, Arista cEOS, Palo Alto PAN-OS, and Fortinet FortiGate — run via bring-your-own-image (BYOI): upload once and NetPilot bakes it in. Additional variants (e.g. Cisco IOS-XE), SONiC, and other custom NOS images are built for you on the enterprise plan. The agent handles vendor-syntax differences automatically — ask for "eBGP between the Cisco edge and the Juniper transit" and it writes correct Cisco and Junos at the same time.
Always — but as the verification layer, not the job. The agent does the building; the real CLI is how you check its work and take over for the cases you want to drive yourself. SSH into any device, run the real vendor NOS, confirm the routing tables and protocol state match what the agent reported. Agent-only would be a black box a network team can't trust — the CLI is the trust layer. The toil is gone; the verification, judgment, and sign-off stay yours.

Put an agent on the toil. Keep your engineers in command.

Try It Free