Updated 2026-09-29

SSH from your phone is painful: troubleshoot your server from Telegram instead

You're on a train, at dinner, or half asleep, and something on your server is wrong. You could open an SSH app on your phone and start typing docker logs --since 30m --tail 200 … on a glass keyboard. This guide shows another way to fix server problems from your phone, or at least find out what's wrong: in OpsMate's Telegram bot, you describe the problem in a sentence, the AI runs the checks over SSH, and you approve anything risky yourself. It covers what you need to set up first, what happens to each command, what's reasonable to do from a phone, and when to put the phone down and go back to a real computer. Getting alerts into Telegram in the first place is a separate topic, covered in Server alerts in Telegram. This guide starts where that one leaves off: you know something is wrong, and all you have is your phone.

The bot replies in English or Chinese based on your OpsMate account language or, if you haven't set one, your Telegram language, and you can switch with /language. In AI chat, the AI replies in your current language. This guide uses the English labels.

Why SSH from your phone is painful

Phone SSH apps such as Termius or Blink Shell work, and for a single uptime they're fine. Real troubleshooting is where it gets painful:

• The keyboard. Pipes, dashes, quotes, ~, Tab and Ctrl are buried behind extra keys, and autocorrect "fixes" journalctl for you.

• Long commands. A useful log check is 60 characters of flags and a container name you have to spell exactly.

• Output that doesn't fit. Log lines wrap into an unreadable column, and scrolling back through 200 lines on a phone means losing your place.

• Dropped sessions. Switch apps or lose signal in a tunnel and the session may be gone, along with whatever was running in it. tmux or mosh helps, if you set it up beforehand.

• A private key on your phone. Your SSH key has to live in the app on a device that's easy to lose.

If you're comfortable with all that, a phone SSH client is a perfectly good tool. The rest of this guide is for when you'd rather say what's wrong in one sentence and read a summary.

Before you start: alerts need no cloud credentials, remote SSH does

These are two separate things, and they have different requirements:

This is the trade-off, stated plainly. By default, the OpsMate desktop app keeps your SSH credentials on your own machine. The Telegram bot runs in the cloud, so to reach your server from Telegram, the key for that server has to be stored in the cloud too. Decide per server whether that's acceptable. A few practical options:

• Keep production servers on local-only credentials, and use Telegram SSH only on servers where a cloud-stored key fits your security policy.

• Give the cloud a dedicated key that you can revoke on its own, not the key you use everywhere else.

• Use an account on the server that has only the rights you're comfortable using from a phone.

Either way, you can still receive alerts in Telegram for every server. Only the remote SSH part needs the cloud key.

1. Enter AI SSH mode and describe the problem in one message

In your private chat with the OpsMate bot, send /servers, pick the server, and tap 🤖 AI chat. That puts you in AI SSH mode. From then on, you don't type shell commands; you send ordinary messages to the AI. That's the difference from a plain Telegram SSH bot, which relays whatever you type straight to a shell: here you describe the problem, the AI picks the commands, and each command goes through a safety check first.

If you came from an alert, note that the alert card's 📋 View and handle button does not take you into AI SSH mode. It opens the issue's details (the issue, severity, server, status, evidence and when it was last updated) with the buttons Recollect, Snooze, Intervene, AI advice and Mark resolved. AI advice gives text suggestions only and runs no commands. Intervene is for describing the situation in your own words, and you finish it with /done or /cancel. Neither of them is AI SSH mode. The alert card is where you look at the issue; the AI chat you open from /servers is where commands run.

A good first message names the symptom, the service and roughly when it started, and says what you don't want done yet:

The API has been returning 502 since about 08:40. Check the app container and Nginx. Don't restart anything yet.

Short is fine; you're typing on a phone. "Site is slow since 9pm, check load, memory and disk" gives the AI enough to start.

2. What happens to each command: run directly, or wait for your approval

The AI decides which commands to run. Not every command needs your approval. Each one goes through a safety check first:

To change a held command, tap ✏️ Edit. The edited command goes through the same safety check again:

• If it passes, it runs and its output arrives as a 📟 Terminal output message. As with an approved command, the AI doesn't analyse it until you send your next message.

• If it's still held, a new ⚠️ Command pending confirmation message appears with ✅ Approve and ✏️ Edit.

• If it isn't allowed at all, the bot tells you the command didn't pass the safety policy, and it doesn't run.

A few rules that matter when you're deciding on the move:

• Approval expires after 30 minutes. If you leave a held command alone, it doesn't run later by surprise. If you still want it after that, ask again.

• Only the Telegram user who started the session can approve its commands. A teammate can't approve on your behalf.

• After an approved or edited command runs, send a message. Something as short as "what does that show?" is enough: the AI then analyses the output it kept, together with what you wrote.

• Rounds, not a command count. For each message you send, the AI runs at most 3 rounds of commands, and each round can include several commands. A typical pattern: round 1 checks uptime, df -h and docker ps -a together; round 2 reads the logs of the container that looks wrong; round 3 checks the thing those logs point at. After 3 rounds, it won't run more commands for that message. To keep going, send another message or tap a ▶️ suggestion; either one starts a new turn with a fresh 3 rounds.

• Read the held command before you tap anything. The reason tells you why it was held. If you don't understand the command, don't approve it; you can come back to it at a computer.

The raw 📟 Terminal output messages are your evidence. The AI's summary is built from them, so when the summary says "the disk is full", scroll up and find the df -h line that shows it.

3. Read the evidence, then pick the next step

Each AI reply comes with next-step suggestion buttons such as ▶️ 1. …, listing what it would check or do next. Tap one to continue with that step instead of typing it out. A tap counts as a new message: the AI starts a new turn, the 3-round limit starts over, and it uses AI calls just like a message you type.

If you'd rather run a command yourself, tap ⏹ Exit AI mode to go back to manual SSH in the same chat. A few short commands are worth knowing by heart for exactly this situation:

uptime
df -h
free -h
docker ps -a

uptime shows load averages (compare them with the number of CPU cores), df -h shows whether a disk is full, free -h shows available memory (read the available column), and docker ps -a shows which containers are up, restarting or exited. None of them tells you why something broke; they tell you where to look next.

Can you fix a server from your phone?

Sometimes, if the fix is small and easy to undo. Reasonable from Telegram:

• Finding out what's wrong: which service is down, whether the disk or memory is full, what the first error in the logs was.

• Restarting one failing service or container, once you've seen why it failed and copied the evidence somewhere.

• Deciding it can wait until morning, and writing down what you found.

For anything bigger, the phone is the wrong tool, even when the AI suggests the step and the button is right there.

When to go back to a real computer

Put the phone down and open a laptop when:

• The change is big. Editing configuration files, database migrations, package or kernel upgrades, firewall rules, deleting data. You want to see the whole file, make a backup and read the diff.

• You're rolling back. A rollback touches code, config and possibly the database at once, and you need to know whether the migration can be reversed. That isn't a decision for a small screen at 3 a.m.

• You need to watch for a while. Confirming that memory has stopped climbing, or that errors stay gone for an hour, means keeping graphs and logs open over time. A chat thread isn't made for that.

• The output is long or you're going in circles. If several messages and their rounds haven't narrowed it down, a full terminal is faster.

• You'd have to paste a secret. Never type passwords, tokens or keys into the chat.

• You can't reach the server at all. If SSH doesn't work, the bot can't connect either. Go to your cloud provider's console.

On a computer, OpsMate's web app puts the SSH terminal and the AI on the same server page, so you can pick up where the phone left off; see AI and SSH commands in one workspace. If the whole site is down and you're on your own, the first-15-minutes order in Site down at 2 a.m. works from a phone as well as a laptop.

Boundaries

• AI SSH mode is mainly for troubleshooting. Dangerous commands are blocked. What you approve is your decision; for what OpsMate does and doesn't do on its own, see the FAQ.

• An AI summary is a starting point, not proof. Check it against the 📟 Terminal output messages before you act on it.

• Command output is sent to cloud AI for analysis. Commands that pass the safety check run and are analysed without a stop in between; the output of commands you approve or edit stays in the session and is analysed with your next message. So keep secrets out of reach: don't ask for commands that print config files, .env files or connection strings, and if a held command would print them, edit it or don't approve it. The output also stays in your Telegram chat history. Redact anything sensitive before you forward or screenshot it.

• Remote SSH from Telegram needs an SSH key stored in the cloud. Receiving alerts doesn't. The desktop app keeps credentials on your own machine by default; the Telegram SSH route is the exception, by design.

• If SSH to the server fails, the bot can't connect either. Use your cloud provider's console.

Illustrative examples

These three scenes are illustrative examples, not real customer cases. Names, services and times are made up.

On the train (indie developer). Leo runs a small SaaS on one VPS and is on a 40-minute commute when a user writes that the dashboard shows a 502. He sends /servers, picks the VPS, taps 🤖 AI chat and sends: "dashboard 502 since 08:40, check the app container and nginx, don't restart anything." Round 1 runs docker ps -a and ss -ltnp together: the app container is Restarting, and nothing is listening on port 3000. Round 2 reads the app's recent logs: the first error is a refused connection to Redis. Round 3 shows the redis container exited. The AI's suggestion to restart redis arrives as a held ⚠️ Command pending confirmation message with its risk level and reason. Leo has already copied the log lines into his notes, so he taps ✅ Approve for that one container. The restart output arrives as a 📟 Terminal output message, and the AI waits for him, so he sends "restarted, check that app stays up now". The AI reads the restart output together with that message and runs a new round of checks. Why Redis exited is a question for his laptop that evening.

No laptop (founder). Mia, a non-technical founder, is at a family dinner when a customer says checkout is very slow. She doesn't know the commands, so she writes: "customers say checkout is very slow, what's going on? don't change anything." The checks that pass run on their own: load is normal and memory is fine, but df -h shows the data disk at 97%. The AI proposes deleting old files to free space, and that command arrives as held, with the reason. Mia doesn't understand what would be deleted, so she doesn't approve it; it simply expires after 30 minutes. She copies the summary and the df -h output, removes the server's address, and sends it to the contractor who set up the server. Not approving is a valid answer.

Woken at night (on-call teammate). Sam is on call for a five-person team and gets an alert at 3 a.m. that a background worker keeps dying. From bed, he opens the alert's 📋 View and handle to read the issue details, then goes to /servers, picks the server, taps 🤖 AI chat and asks the AI to check the worker container and memory. Over three rounds, the evidence points to the worker being killed for running out of memory, with memory climbing since yesterday's deploy. The restart policy already brought the worker back and the queue is draining. The real fix, rolling back yesterday's deploy or raising the memory limit, is a big change that needs watching afterwards. So Sam writes three lines in the team chat (symptom, evidence, proposed change and how to undo it) and leaves the decision for the morning at a computer. He can't hand approvals in his session to the team lead anyway: only the person who started a session can approve its commands.

AI diagnostic prompt

Short enough to type on a phone:

Something's wrong with [service] since about [time]. Check it without changing anything: load, memory, disk, the service's status and its recent errors. Tell me what's confirmed and what's a guess. If a change is needed, suggest the smallest step and how to undo it, and wait for me.

Next checks

• Setting up alerts, or reading an alert card before acting: Server alerts in Telegram: investigate when you are away from your desk.

• Whole site down and you're the only one on call: Site down at 2 a.m.: the first 15 minutes when you're the only one on call.

• Back at a computer: AI and SSH commands in one workspace.

Try OpsMate

• Website: https://www.itops.sh/en/?utm_source=guide&utm_medium=cta&utm_campaign=sprint_w1&utm_content=fix-server-from-phone-telegram

• Sign up: https://app.itops.sh/login?screen=register&lang=en&utm_source=guide&utm_medium=cta&utm_campaign=sprint_w1&utm_content=fix-server-from-phone-telegram

500 free AI calls per month and unlimited servers. The desktop app keeps your SSH credentials on your own machine by default; AI SSH mode in Telegram is the exception and needs an SSH key stored in the cloud for that server.

Troubleshooting guides