AI-SSH-Pro is a local, multi-session SSH desktop client with a built-in, Cursor-style AI ops assistant. The AI proposes commands; you decide how they run: Suggest only, Confirm before run, or Auto run.
- Website: https://guoshuaipeng.github.io/ai-ssh-pro/
- Download: https://github.com/guoshuaipeng/ai-ssh-pro/releases
- Source (MIT): https://github.com/guoshuaipeng/ai-ssh-pro
The UI ships in both English and Chinese — switch the language in Settings. Screenshots below use the English UI with the built-in demo host (simulated data; some sample AI replies are in Chinese).
Who it's for
- Ops / SRE engineers watching several machines at once who want one place to see everything.
- Backend developers who regularly SSH in to read logs or restart services, and want help structuring the troubleshooting.
- Small-team leads who want AI in the loop, with per-host control over what it may execute.
- Xshell / FinalShell users looking for familiar tabs and a session tree.
Ops home: every host at a glance
The app opens on an ops dashboard:
- Online hosts and open alerts at the top.
- One card per host with live CPU, memory and disk.
- One click into Terminal, SFTP or Monitor.
- Quick actions for common playbooks, such as a rolling nginx restart or disk cleanup.
Multi-session SSH, done properly
- Tabs, split panes and session groups, with search and
#tagfiltering. - Jump hosts (multi-level ProxyJump), port forwarding and host-key verification.
- Import from
~/.ssh/config; reconnect to recent sessions in one click. - Production tabs carry a prominent banner, so you always know where you are.
The AI panel
The assistant lives in a panel to the right of the terminal, so you never switch windows mid-incident.
- Suggested commands are grouped into one compact block with Copy / Run and a risk label (low / medium / high).
- Each reply shows how long it took and can be copied in one click.
- The input row holds everything: model, execution policy, attach terminal output, send.
- Connect any OpenAI-compatible provider (OpenAI, DeepSeek, Qwen, …). Your API keys stay on your machine.
Three execution policies
Pick one in Settings → AI:
| Policy | Behavior |
|---|---|
| Suggest only | Runs read-only commands; changes are shown as suggestions only |
| Confirm before run (default) | Read-only commands run; changes need confirmation; high-risk needs a second confirmation |
| Auto run | Low- and medium-risk commands run automatically; high-risk still asks |
- Under Auto run you can additionally allow high-risk commands to run automatically. It's off by default.
- The policy menu in the input row can override the policy for the current tab only.
Per-host safety
Every host profile has a Safety card:
- Protection level: Normal / Production / Read-only.
- AI policy for this host: follow the global setting, or pick one.
- The card shows the policy actually in effect, e.g. "In effect: Suggest only (limited by read-only guard)".
Production hosts are not capped: your chosen policy applies as-is, and the banner reminds you of the environment. Only read-only hosts block AI writes.
Conversations bound to tabs
- Each terminal tab is its own conversation; two tabs on the same host give you #1 and #2.
- + New chat archives the current conversation instead of deleting it.
- History is grouped by host — click to restore.
- With no terminal bound, the assistant is plain Q&A and never runs anything.
SFTP and the rest of the toolbox
- Dual-pane SFTP: local and remote side by side, drag-and-drop, transfer queue.
- Monitoring: resource charts, multi-host dashboard, alert rules and an alert workbench.
- Toolbox: processes, ports, services, logs, Docker containers and Compose.
- Productivity: snippets, playbooks, batch execution, broadcast input, session recording and replay.
- Audit: sensitive operations are logged, filterable by time and exportable.
Security by design
Safety is layered, not a single switch:
- Main-process enforcement: before the AI writes to a terminal, the main process re-checks host protection. On a read-only host, the AI cannot write even if you click approve.
- One rule everywhere: settings, host card, AI input and background execution all compute the effective policy the same way — no "UI says confirm, backend just runs".
- Bounded overrides: per-tab overrides live in memory only, reset on close or disconnect, and still respect read-only.
- Credentials stay local: passwords, key passphrases and API keys are encrypted with the OS secure store and never uploaded to any product cloud.
- Traceable: AI-initiated commands and dangerous operations are recorded in the audit log.
Get started
- Download the latest installer from GitHub Releases.
- Launch it and try the built-in demo host — all data is simulated, nothing touches a real server.
- Add your own model provider in Settings → AI.
If you want a comfortable multi-session SSH client where the AI proposes and you decide how it executes, give AI-SSH-Pro a try. Stars and issues are very welcome.
Top comments (1)
how do you sandbox the auto‑run mode to avoid a malicious suggestion escalating privileges across sessions?