Public Preview: NetBox Agents Automate Infrastructure Workflows, With You in Control

NetBox Agents are now in public preview. They investigate, report, and propose change against the infrastructure model you already maintain in NetBox, and every write runs under an approval policy you control.

Putting an AI tool anywhere near your infrastructure raises one question before any other: what, exactly, is it allowed to change? Nothing else gets approved until that question has an answer.

That is where NetBox Agents start. In public preview today, they work from the infrastructure data model your team maintains, and they propose change rather than making it.

• Grounded across every vendor you have modeled. Agents draw on your own topology, addressing, and intent, including your own customizations like plugins and custom fields.
• Useful on the first run. Pre-built agents for common infrastructure management workflows ship ready to run, and you can tune their instructions when you want them to match your own conventions.
• Autonomous on your terms. Describe an agent in natural language, then decide when it should go to work: on a schedule, when a condition you describe is met, on a NetBox event, or on demand.
• Every write behind your policy. You decide what an agent can do on its own and what waits for a person. Every action is logged and attributable.

Why we’re building this

Network and infrastructure teams exert heroic effort to manage increasingly complex and critical environments. Automation has long promised relief, but it demands new skillsets like coding and DevOps, holding infrastructure teams back. We believe AI agents offer a critical unlock for these teams – automation, built the way infrastructure teams really think: with natural language describing workflows and processes, deployed where they create leverage for operators.

At the same time, agents – like any automation – are only as good as the data they act upon. NetBox has long enabled teams to model their intent in a structured knowledge graph, which over the last several years has become the critical context layer for AI and agents across the industry. At NetBox Labs, that’s afforded us the vantage to understand – and now deliver – what agents and teams need to actually succeed in practice.

Grounded data is critical for infrastructure agents, but it isn’t enough. Errant changes to infrastructure aren’t like bugs in code: they take down entire operations and can’t just be reverted. Every infrastructure change must be made safely. Much of what makes NetBox Agents special comes from the guardrails inherent in the NetBox Labs ecosystem, earned over years as the system of record for mission-critical infrastructure management. That means a rigorous real-world data model, branching and change management workflows with approval gates, pre-change validation and policy controls, and safe-by-default human in the loop approvals with comprehensive audit trails on every data change.

What’s in the preview

Pre-built agents for the work you already do

Pre-built agents lower the barrier to starting. The first three cover workflows common to nearly every infrastructure team, with more on the way, and each one is tunable, so you can review and edit its instructions to match how your team actually operates.

• Reporting posts scheduled digests covering capacity, drift, hygiene, and change activity, all read-only.
• Troubleshooting traces documented cable paths, correlates recent changes, and maps blast radius for a connectivity question, also read-only.
• IPAM watches addressing events and keeps allocation consistent, carving subnets and flagging conflicts, with every write staged behind your approval.

The best place to start is Reporting. It runs on a schedule against your own estate, so you can see what an agent produces on real data before anything is at stake.

Three pre-built agents, plus a Custom Agent option.

Custom agents built with natural language, not code

Beyond our pre-built agents, your team describes a recurring task in natural language, attaches the context it should consider, picks its toolkits, and sets what kicks it off. No script, no developer, no backlog ticket.

Agents run on a schedule, on a NetBox event, when a condition you described in natural language is met, or on demand, so the work happens between conversations.

Every change behind a gate

You set the approval policy by action type. That decides what an agent handles alone, what needs a named approver, and what arrives as a branch diff you review before anything merges.

Each agent acts under a platform service account you scope, so the RBAC you already maintain governs what it can reach. There is no second permission system to run. Every agent action is logged and attributable, whether a person approved it or your policy allowed it. One console shows what ran, what each agent did and why, and what is waiting on a decision.

An agent’s proposed write, waiting for a decision.

How it works

Agents connect to your NetBox Labs platform on their own. There is nothing to install and no schema to map. They read the infrastructure data model you already maintain, including the plugins and custom fields your team added, and they gain reach as you extend it. That data model describes what your infrastructure is supposed to be, not what one vendor’s telemetry reports it is.

NetBox Labs hosts the AI model, so there is no client to connect and no key to rotate. More AI model hosting options will be available at general availability. Agents carry context between runs and keep your team’s corrections, so local conventions stick.

Agents and NetBox Copilot are different tools on the same platform. Agents work on their own schedule against your data. NetBox Copilot answers questions for a person in the UI. Both reach the platform the same way.

How to get hands-on

You will need the SaaS edition running NetBox 4.5 or later, with Branching and Change Management available. Ask your account executive or customer success manager to enable NetBox Agents for your organization.

Because agents work from what you have modeled, the more complete your NetBox, the more useful they are on day one. Most teams start on a non-production instance.

Agents are a commercial capability of the NetBox Labs platform, not part of open source NetBox, and will be tiered at general availability. Any feedback you have during the preview goes straight to the team building it.

Not on the SaaS edition yet? Request a demo and we will walk through Agents alongside the rest of the platform.