Let an assistant work out what is wrong with a computer, and hand you a fix worth approving.

Cortt puts a small agent on the machines you look after. Your assistant asks it real questions. What is running, what the logs say, what the network is doing. It reads the answers and comes back with the one fix that helps. You say yes or no. It cannot change anything before you do.

  • Any assistant
  • Windows, macOS, Linux
  • Answers in seconds

Three things to set up, once.

Install the agent

A small program, installed like any other. It runs as a service, uses almost nothing, and needs no inbound access.

Deploy it the way you deploy anything else.

Bring your own assistant

Claude, ChatGPT, Copilot, or a model running on your own hardware. Cortt is not tied to any of them and does not sell you one. Connect whichever your team already uses.

Claude ChatGPT Copilot Your own model

Decide what it may run

You write the list of what is allowed: restart this service, clear that queue. Reading is never on the list, because reading changes nothing. Anything that changes a machine waits for a person to say yes.

Start from ours and edit it, or write your own.

What it actually does

The work a good technician does in the first twenty minutes of a ticket, done in the time it takes to read the ticket.

Investigates on its own

What is running and what stopped. What the logs said overnight. Disk, memory, updates, network, who signed in. Reads only, so nothing on the machine changes.

Explains in plain terms

Not a wall of output. What broke, when it started, why it looks intermittent, and what it would do about it, ready to paste into the ticket.

Fixes once you approve

Restart the service, clear the queue, remove the bad record. You see the exact command before it runs, and it only runs if it is on your list.

Works across the whole site, not one box at a time

Ask about every machine at a site at once. Faults that only show up as a pattern across machines get found, not just the one that raised the ticket.

Remembers what it learned

Site details worth keeping, like addresses, quirks and change windows, stay with the site, so the next person does not start from nothing.

In the chat you already work in, and the channel you already watch.

In your assistant
cortt Reading only, nothing changed

Asked DNS-02 who answers for the domain, and got back 10.0.4.9, a machine that was retired in June. It still answers, then refuses, so every sign-in waits thirty seconds on it.

Removing that one entry should fix it for everyone at the site. It is on the list you wrote and it can be put back, but it changes a server, so it needs your yes.

Yes, do it No

Output from a machine is always shown as output from that machine, kept separate from what Cortt tells you.

In Slack or Teams
Cortt 9:41 Waiting on you

Approval needed: remove the dead server entry on DNS-02

Fourteen machines at that site are taking 94 seconds to sign in. One server, one entry, reversible.

Yes, do it No Open in Cortt
Proposed 9:41 → Waiting on you → Approved 9:42 → Done 9:42

Nothing disappears. Declines stay too, with the reason, so months later you can see why something was not done.

The assistant suggests. You decide. The record stays.

An assistant given a technician's password can do anything a technician can. Cortt gives it a much smaller set of keys, and you cut them.

On the machine

The agent answers questions using the operating system's own interfaces. The only thing it ever launches is a command that was already on your list.

In the middle

Your list decides what may run, and a shorter list underneath it, covering password stores, security tooling and anything that wipes a disk, stays off limits whatever your list says.

In the chat

The assistant never gets a password or a remote session. It gets a set of questions it may ask, fixed to the sites the person who connected it could already reach.

Servers are treated as servers

Domain controllers and hypervisors are detected automatically and held to a stricter standard than a laptop.

Say yes in advance, with an end date

Rebooting kiosks overnight for the next month, say. It expires on its own, and you can pull it back at any point.

Every machine has its own identity

Issued on the machine, never shared, and never a name someone can type in to pretend to be it.

The log cannot be quietly edited

Every entry is tied to the one before it, so a line taken out of the middle shows up as missing.

You can keep your own copy

Cortt signs the state of your log every hour and posts it somewhere you own, so the proof does not only live with us.

A report you can hand over

What was looked at, what was proposed, who approved it, and when. Per site, per month.

An assistant that can look at the machine, and still has to ask you.

Join the waitlist and you will hear from us when it is ready to try.