The usual way to get help on a server still looks like this: you reproduce the error in a terminal, copy some logs, open a chat, and type a briefing. The model has no idea which host you are on, what user you became, or what you already tried, unless you tell it.

I got tired of that loop, so I built Ark SSH.

Ark is an AI-native SSH client and terminal for real servers. It is a desktop app for macOS and Windows. The important part is not that it has a chat panel. The important part is that the Agent starts from the SSH session you already opened.

Not a chat you paste into

Ark is a complete SSH client on its own. You get a familiar CLI with splits, history, and completion; SFTP browsing and transfers beside the terminal; and a view of server state — identity, resources, services, ports, and watch items — before you decide what to do. You can import hosts from your existing SSH config. Unlimited connected hosts are included on the free plan.

The Agent is optional. Nothing in the client requires you to use it.

When you do use it, it does not wait for a pasted background. It starts from the same connection: the host, the identity, the current directory, running services, listening ports, and recent terminal output. CLI, files, state, and the Agent share that one real SSH connection.

Control stays with you

I do not want an agent that mutates a production host behind my back.

Before a material change, Ark shows the server, the identity, the target, the full command, the expected impact, and how the result will be verified. A confirmation authorizes only that operation, not the steps after it. After the command runs, execution success, technical verification, and business health are reported separately. If the Agent is not the right tool, you stop and finish in the CLI.

SSH private keys and passwords stay on your device. They are not uploaded to Ark. For models, you bring your own API key. Today that means an OpenAI-compatible API, which also covers locally hosted models such as Ollama. The key is sent only to the endpoint you configure.

A short product film on the site walks through a 401 investigation: inspect server signals, review a command, and separate findings from unknowns.

Platforms and price

Ark runs on macOS 12 or later (Apple Silicon and Intel) and Windows 10 or later (x64). macOS is also on the Mac App Store outside mainland China.

  • Free ($0): SSH/CLI, SFTP, unlimited hosts, local history, multi-device sign-in without sync.
  • Basic ($1.99/month): the Agent, native Harness, a single-server cockpit, and custom models.
  • Pro ($4.99/month): inspections and scheduled tasks on your device while Ark is running, multi-device sync, and an iOS companion. An Android companion is in development; some cross-server features are still planned.

Accounts registered by 30 November 2026 receive Basic through a promotional period that does not auto-renew or charge. Without a paid subscription, the account returns to Free when that period ends.

If you already live in SSH — backend, test, GPU boxes, or DevOps/SRE — I'd like to hear whether starting from live server state helps, and where the confirmation step helps or gets in the way.

Site and downloads: https://arkapp.ai