Quickstart

From a fresh install to a diagnosed failure and an approved command, in roughly five minutes. You need one host you can log into over SSH and nothing else — no account, no API key yet, and no changes on the host. Everything here works during the 15-day free trial.

Step 1 — Install and start OpsPilot

1. Download the installer from subscription.nubestack.com/download.
2. Check its SHA-256 against the one on the download page, then run it. Install has the commands and explains the Windows publisher warning.
3. Start NubeStack OpsPilot. The first launch starts the free trial with no account and no email, and the title bar shows Trial · 15 days left.

The trial chip sits at the right of the title bar and opens Settings → License. The + new connection button you need next is at the foot of the connections list; on a new install the list holds only Local.

Step 2 — Create a connection

Click + new connection at the foot of the connections list. Pick SSH, then fill in a connection name, the host, port and username, and either a password or an SSH key.

The new connection dialog. Enable AI decides whether the AI may see this host at all, and Let AI Assistant open this session is a separate, stricter permission for external assistants. The two profile fields can stay on — inherit from group —.

Set these before you click connect:

• environment tags the connection as production or staging, or as an environment you add under Settings → Environments. Environments drive the colour-coding in the connections list, and they are how you scope safety policy later. Tag each connection for what it really is, so you do not have to reclassify a fleet afterwards.
• Enable AI controls whether the AI may see this session at all. Turn it on for this connection, since it is the one you will try the AI on. A connection with it off is invisible to every model and every assistant.
• Let AI Assistant open this session — leave it off for now.
• Save connection — leave it on so the entry persists. Credentials are stored encrypted, never in plaintext.

Click connect. From then on, double-click the saved connection to open it again.

Step 3 — Work normally

You now have a real terminal: full xterm emulation, GPU-accelerated rendering, search, copy and paste, and a session tab bar for working across several hosts at once. During the trial you can have up to 10 sessions open at once.

Nothing about your normal workflow changes, and nothing in this step involves the AI. OpsPilot is a complete terminal before any AI is configured.

A connected SSH session. The banner confirms the connection, its environment and ai: enabled. Here nginx fails to start, which is the kind of failure to hand to the AI in the next steps. The bar along the bottom is the system monitor for this host.

Step 4 — Turn AI on

First choose how the AI reaches you. Open Settings (the gear in the title bar), then AI Providers:

• Fastest to try: OpenRouter. It has free tiers suitable for evaluation, so you can see the loop working before committing to a provider.
• Best for a locked-down environment: Ollama. Install Ollama, run ollama pull llama3.3, then point OpsPilot at http://localhost:11434. Nothing leaves the machine. Small local models are less precise at diagnosis than the large cloud models.
• If you already pay for Claude, ChatGPT or Copilot: skip providers and connect it as an AI Assistant over MCP instead. Model usage is covered by that plan, with no API key; connecting it needs your OpsPilot trial or a subscription.

For a provider, click Test Connection to confirm the credentials work, then Save & Activate.

Settings → AI Providers. The provider in use sits at the top, marked Active. The others are listed below, each Not set up until you open it and enter its credentials.

Then check that AI is on for your connection. Its badge in the connections list should read AI. During the trial, AI can be on for 2 of your connections at a time, and the first two connections you save with Enable AI on take those two places, so your first SSH connection has AI. If its badge reads AI off · trial limit, click the badge: it takes a free place at once, or a dialog offers Use AI on … instead of … one of the two connections that have it. See Free trial and limits.

Step 5 — Ask the AI panel

Break something harmless first. In the SSH session, run a command that fails with no consequences:

systemctl status a-service-that-does-not-exist

Then type in the AI panel's Ask a question… box and press Enter:

why did that fail?

You get a diagnosis and, usually, a proposed next command on a card carrying a risk badge — Read-only, Low risk or High risk.

If a 🔒 badge appears on the turn, redaction fired before anything left the machine. The number beside the padlock is how many values were replaced; hover it to see which categories were scrubbed.

Asked why nginx does not start, the AI says what it will check and proposes nginx -t, which tests the configuration without changing it. The card carries a Read-only badge and waits for approve & run or dismiss.

Step 6 — Approve the proposal

Read the command on the card: it is the exact text that will be typed into the session. Click approve & run to run it, or dismiss to drop it. The command runs in the real terminal, and its output, redacted, feeds the next turn of the conversation.

A High risk card says why the command was flagged and asks why it is needed. Type a reason of at least ten characters, then click confirm & run. The reason is kept with the command.

A High risk card. The note under the command says why it was flagged: it matches the rm -rf pattern in the Default profile. confirm & run works only once a reason is typed in the box.

The command did not run until you clicked. A model's output is a proposal, and only OpsPilot's own code types it into the session: after your click, or under a rule you set in advance in your Command Safety profile, which is the next step. No provider, prompt or assistant can make it run any other way.

Step 7 — Set your safety policy

Open Settings → Security → Command Safety. Clicking the Auto-run off chip in the AI panel's composer opens the same page. The three rows there set the rules for the Default profile, which every group and connection uses until you give it a profile of its own:

1. Run commands without asking me — ships as Ask me every time, so nothing runs until you click. Leave it there until you have watched the AI work for a while. Only read-only commands lets commands that only look at things run by themselves; Everything except dangerous ones runs whatever the AI suggests unless it counts as dangerous.
2. What counts as a dangerous command — ships as The AI's warning and my list. Keep it: a command is then dangerous if the AI says it is destructive or if it contains one of your patterns. Only my list ignores the AI's warning and makes your list the only judge.
3. Make me type a reason for dangerous commands — on as shipped. Keep it on. A dangerous command always needs a click; this adds the typed reason.

Under the rows, What this means right now spells out the result in plain sentences. Click Save Security Settings to keep your choices; saving a looser combination first shows Check these rules before saving.

Command Safety as it ships. What this means right now turns the three rows into plain sentences, here: nothing runs by itself, a command is dangerous if the AI warns about it or it matches one of the 21 shipped patterns, and a dangerous command needs a click and a typed reason. How OpsPilot runs a command is marked Always enforced: it is not a setting.

Then check the dangerous patterns: choose Manage profiles… and edit the Default profile's list of dangerous words. The shipped list already covers the obvious ones (rm -rf, dd if=, mkfs, drop table, shutdown and more), so your job is to add what is specific to your estate.

Estate-specific patterns worth adding

Start from the things that are irreversible in your environment rather than in general:

terraform destroy

Then your own: the cluster-wipe script, the deployment tool's teardown subcommand, the storage command that reinitialises an array. Patterns are plain, case-insensitive substrings — not regexes — so a fragment of the command is enough.

With The AI's warning and my list, matching a pattern can only promote a command to High risk. Nothing on your list can demote a command the model already flagged, so adding patterns can only make the rules stricter.

Step 8 — Subscribe when you are ready

The trial runs for 15 days from the first launch, and the title bar counts them down. When you are ready:

1. Subscribe at subscription.nubestack.com/opspilot, or choose Subscribe under 1 Get a subscription in Settings → License. Each person gets a license key that works on up to 2 devices.
2. Activate this computer in Settings → License, under 2 Activate this computer: Activate online with your license key, or Activate offline with a request code and a license file, which needs no network on this computer.

Subscribing lifts the trial's limits, and AI comes back by itself in your open sessions, except where you turned it off. If you do not subscribe, nothing is lost: after the trial the terminal keeps working with your first 10 saved connections and up to 10 sessions open at once, AI is off, and all your configuration is kept. See Subscribe and Activate OpsPilot.

Settings → License during the trial. The card at the top gives the trial's limits and the day it ends. 1 Get a subscription has Subscribe and the address to open on another computer; 2 Activate this computer offers Activate online, Activate offline and Import license file.

Where to go next

You have the loop working on one host. The next decisions are about scope: which connections get AI enabled, which groups carry which Command Safety profile, and whether a Data Handling profile needs tightening for your data.

See also

• Execution boundary & risk tiers — read this before rolling OpsPilot out to a team
• AI Providers — every provider that ships, and what each needs
• Offline with Ollama — the fully offline route in detail
• Licensing — the trial, subscribing and activation in full