Trial
Free · 15 days
Install it and start working. No card, no quote, no sales call.
- The complete product
- AI on 2 sessions at a time
- Up to 10 concurrent sessions
- Every connection type
OPSPILOT
A co-pilot that never touches the controls
It reads the session with you, works out what broke, and writes the next command out in full — then waits. Flying the aircraft stays your job, and there is no setting that changes that.
An AI-powered terminal, for the infrastructure AI normally cannot reach.
A terminal, a file manager, a remote desktop client and an AI assistant in one window. Secrets are stripped on your machine before anything is sent, and what comes back is an explanation and a proposal — never an executed command.
Two inversions
Everything else assumes the opposite of what your estate actually is
What the rest assume
- The machine you’re working on can reach the internet
- Giving the model a shell is the point
- An agent gets installed on every host
- Regulated estates are somebody else’s problem
What OpsPilot assumes
- Only your workstation needs a route out — with a local model, not even that
- The model proposes. You execute
- Nothing is installed. If you can SSH to it today, you’re done
- Banking, defence, healthcare and air-gapped networks are the point
OpsPilot runs on the one machine that already touches both sides. The servers make no outbound call and never learn an AI was involved.
The execution boundary
The AI is never given a shell
Every proposal lands in one of three tiers, and turning one into a real command takes a click you make.
- Read-only Runs on its own, if you’ve turned that on. Off by default.
- Low risk One explicit click, every time.
- High risk A typed written justification, then the click.
Your own pattern list can promote a command to dangerous. Nothing can demote one the model already flagged.
What leaves the machine
Secrets are removed before anything is sent
Redaction runs locally, on the single path between the terminal and the AI.
- Ten redaction categoriesKeys, tokens and passwords always. Hostnames, IPs and emails when you want them — strict in production, relaxed in the lab.
- Ten AI providers, including a local oneClaude, ChatGPT, Copilot, Gemini and the rest — or Ollama on your own machine, and nothing leaves it.
- Use the subscription you already pay forConnect Claude, ChatGPT or Copilot directly. A command proposed there hits exactly the same gate.
- Operate it from your phoneYour workstation stays on the VPN. You don’t. Approval still waits at the desk.
Plans
One price, every feature
Subscription
$5 / user / month
Lifts both caps and keeps the AI layer on.
- Unlimited sessions, AI on all of them
- All ten providers, local and self-hosted
- Assistant integrations and remote operation
- Updates and support
Windows, macOS and Linux, built from the same codebase. When a trial ends the terminal keeps working — only the AI layer switches off, and your profiles are waiting when you subscribe. Inference is billed by whichever provider you choose, and a local model makes that nothing.
Common questions
Worth asking early
No. No agent, no daemon, no outbound firewall rule. OpsPilot connects the way you already do, from a workstation that already reaches both sides.
Yes, with a local model. Point it at Ollama or a self-hosted endpoint and nothing leaves the machine — the interface ships every font and icon locally and never calls a CDN.
It has no way to run anything. A proposal becomes a command only through your click, and a destructive one needs a written reason first.
Only the provider you configured, and only after local redaction. There is no cloud backend, no account system and no sync service.
Documentation
Everything, written down
Install, connections, the AI layer, the safety model and the full reference tables — a complete user and product manual.
Fifteen days, the whole product, no card
Install it, point it at something genuinely broken, and see whether the diagnosis is worth having.
