Every command, seen before it runs.

Smart SSH connects you to your servers with an AI copilot that reads logs, diagnoses problems, and suggests the fix — but it never touches your server without your approval first.

macOS, Windows, Linux · Apache 2.0

How it works

  1. 01

    Tell it what's wrong

    Connect to a server and describe the problem in plain language — a failing service, a full disk, a slow query.

  2. 02

    Review what it wants to run

    The assistant proposes a specific command. You see the exact text — editable, never hidden — before anything happens.

  3. 03

    Approve, edit, or reject

    Nothing executes until you say so. Routine commands you trust can be whitelisted to run automatically — everything else waits for you.

Built around one rule:
nothing runs unseen.

Example command rules
PatternVerdictScope
systemctl status * ALLOW tag: web
rm -rf * BLOCK all servers
apt install * CONFIRM web-01

Rule-based command filter

Every proposed command is checked against your rules before it can run.

ls -la ALLOW
&&
rm -rf /backup BLOCK
Result BLOCK

Chaining can't sneak past a rule

A chained command is evaluated piece by piece.

AWS_SECRET_ACCESS_KEY=wJalrXUtnFEMI/K7MDENG/bPxRfiCY

sent to provider

AWS_SECRET_ACCESS_KEY=••••••••••••••••••••••••••••••

Secrets are redacted before they leave your machine

Command output is scanned for keys, passwords, and tokens, and masked before it reaches any AI provider.

Risk indicators
Commands and file writes are flagged for server risk and data risk.
Host key protection
A changed host key gets a deliberately alarming warning, not the routine first-connection prompt.
Any provider you choose
OpenAI, Anthropic, OpenRouter, any OpenAI-compatible endpoint, or a local model through Ollama.
Per-server context notes
A running memory of each server: versions, paths, decisions.
File transfer with a real diff
File edits arrive as a line diff, with a backup taken before anything is overwritten.
Generated documents
Ask for a summary, get Markdown or Word back.
Platforms
macOS, Windows, Linux — a native build each
Auth
Password, key, agent, certificate — credentials in the OS keychain
Organization
Nested groups and tags, scoped rules and shared notes
License
Open source, licensed under Apache 2.0.

A closer look

Drawn illustrations, not screenshots — they show the shape of each flow rather than a pixel-exact capture of the app.

Illustration Illustration of a proposed shell command held in a confirmation panel, with Run, Edit and Reject as the only ways forward.
01

The confirmation stop

The proposed command, the rule that caught it, and the only three ways forward.

Illustration Illustration of a rule list, each row pairing a command pattern with an allow, confirm or block verdict and the servers it applies to.
02

The rules that decide

Your patterns, their verdicts, and the servers each one applies to.

Illustration Illustration of a line-by-line diff of a config file, added lines in green and removed lines in red, with a backup noted underneath.
03

File edits, as a diff

A line diff against what is on the server right now, with a backup underneath.

The same request, two different endings

An AI terminal that executes on its own

  1. prompt
  2. command
  3. executed
  4. you read about it afterwards

Smart SSH

  1. prompt
  2. command
  3. rule check
  4. your approval
  5. executed

Works with Claude Code and any MCP agent — through the same policy engine.

  • Claude Code
  • Claude Desktop
  • Cursor
  • Built-in chat
Smart SSH policy engine
systemctl restart nginx CONFIRM
rm -rf /var/lib BLOCK

…and a confirmation you have to click.

your server

The agent proposes. You still confirm.

Claude Code, Claude Desktop and Cursor can talk to Smart SSH as an MCP server. Every command such an agent suggests goes through the same filter rules and the same confirmation step as a command from the built-in chat — no separate, looser path for external tools.

Give an agent raw SSH through its Bash tool instead, and none of that applies: no rule check, no risk indicator, no stop before execution. Here the agent keeps what it is good at — reading logs, correlating, proposing — while execution stays tied to a confirmation you can see.

Pricing

Do-it-yourself free, convenience paid.

Available now

Free

0 €

Everything the app does today. No sign-up, no account, no time limit.

Coming

Pro

9 € / month

Managed AI without your own key, hosted sync, searchable history and reports.

Coming

Team

24 € / seat / month

Centrally enforced policies, signed audit export, shared inventory.

Self-hosting and air-gap on request. Local features never require an account.

Full details and roadmap

Tell me when Pro launches

One email when Pro launches. Nothing else.

Almost there — check your inbox.

We sent you a confirmation email. Click the link inside and you are on the list.

Your address is used only for this one announcement and for occasional news about Smart SSH. Unsubscribe anytime with the link in every email.

Download

Smart SSH is in 0.x Early Access — usable every day, but still moving. Expect changes between versions.