Skip to content

OpsPilot

AI on your terminal — including the terminals AI can't normally reach.

Get started Download Pricing Why OpsPilot

The OpsPilot workspace: saved connections with their AI badges on the left, an SSH session in the centre where a broken nginx configuration was fixed, and the AI panel on the right listing the three commands that ran, each with its risk tier, followed by a summary

OpsPilot is a desktop operations workbench — a terminal, a file manager, a remote-desktop client and an AI assistant in one application. You connect the way you always have — SSH, RDP, VNC, FTP, S3, serial, or a local shell — and in SSH and local shell sessions OpsPilot watches the session with you. Ask it a question and it reads the scrollback, strips the secrets out locally, and brings back an explanation plus a proposed command.

The AI helps you operate your infrastructure. It never operates your infrastructure itself.

That command does not run on its own. It appears in front of you, labelled with a risk tier and showing the exact text you are about to execute. You decide.

Start here

  • New to OpsPilot


    What it is, why it exists and how the loop works.

    Overview

  • Ready to install


    Requirements, installers, first run, and a five-minute quickstart.

    Getting started

  • Trying and buying


    The 15-day free trial, subscribing, and activating online, offline or for a whole site.

    Licensing

  • Connecting to things


    Ten connection types, hypervisor consoles, groups and credentials.

    Connections

  • Bringing your own AI


    Ten providers, a fully offline option, or the subscription you already pay for.

    AI

  • Signing it off


    The execution boundary, risk tiers, redaction and the security model.

    Safety & security

  • Rolling it out


    Role workflows, administration, upgrades and troubleshooting.

    Operate

What makes it different

Most AI tooling assumes the machine you are working on can reach the internet, and assumes that giving the model a shell is the point. OpsPilot assumes the opposite on both counts.

Your servers can stay disconnected: only the workstation needs to reach the AI, and with a local model not even that. And the model is never given a shell — it can propose, it cannot execute. Together those two properties make OpsPilot usable in estates where AI assistance has not been an option: regulated banking, defence, healthcare, industrial control, classified networks, and production environments where a model running a command directly would not pass review.

Nothing is installed on the target hosts — no agent, no sidecar, no daemon and no outbound firewall rule. Any host you can already reach is a host OpsPilot can work against.

Private or VPN-only is not the same as air-gapped

With a cloud provider, your workstation still talks to the internet — that is private or VPN-only operation, and it is what most regulated teams actually need. For a genuinely air-gapped deployment, point OpsPilot at a local Ollama or self-hosted model, and license it with an offline license file or an organisation-wide deployment license. The loop then closes entirely: nothing leaves the machine, licensing included.

The safety model

  • The AI never gets a shell. Proposals become commands only through OpsPilot's own approval path. This is architectural, not a setting.
  • Secrets are removed locally, before anything leaves the machine. Every route to a model redacts first.
  • You decide what runs without asking. Each Command Safety profile sets which commands may run without a click — none, read-only ones, or everything except dangerous ones — and a command that counts as High risk always waits for your click. By default your pattern list can only make things stricter: a pattern can promote a command, and nothing demotes one the model already flagged.

See also