We have an announcement post for Sentinel (thread 253) and one for the Telegram bridge (thread 243). What people keep asking in tickets is narrower: what does a real job look like, from the prompt to the result? So here are six, written the way they actually happened.
Two things first. OpsAI drives your own agent — Claude Code, Codex CLI, Gemini CLI or Cursor Agent — installed from the panel and signed in with your account; we do not run a model for you. And the agent is given the panel's own API (an index of 1388 endpoints across 93 categories), so it changes things the same way the interface does instead of guessing at config files.
1. "example.com returns 502. Find out why and fix it."
It reads that domain's error log and PHP-FPM log through the panel, finds the pool refusing connections at its process limit, raises the limit and reloads PHP-FPM, then reports what it changed with the log lines that led it there.
Check afterwards: the new pool value under Domains → PHP Settings — and whether the limit was the cause or a symptom. If one slow script filled the pool, raising the ceiling only buys time, and the agent usually says so.
2. "Move this account from our old cPanel server."
Pick the Migration expert first — it appends a role brief to the agent's system prompt, so it follows Panelica's pipeline instead of inventing one: discovery, credentials, files, databases, mail, SSL, verification. Database password hashes transfer as hashes, so applications keep working without anyone resetting passwords.
3. "Why has the server been slow since this morning?"
With the Performance expert you get a ranked answer — which account, which process, which query — rather than a wall of statistics.
4. From your phone: "disk is filling up, what can I safely remove?"
The Telegram bridge: you message your own bot, the agent answers there and the conversation resumes where it left off. Backups still inside their retention window come back as not safe to delete. The useful moment for server work is rarely when you are at a desk.
5. A page that only misbehaves in a browser
A checkout failing on step three, a redirect loop that needs a cookie — none of it is in the logs. OpsAI ships a real Chromium on the server, shown inside the panel, and the agent can open the page and click through it. For "works with curl, fails in a browser", that is the difference between a guess and an answer.
6. The job you do not ask for: Sentinel
Same agent, but it is not waiting in a chat window. It is the guard on the dashboard's bottom bar, and the rule is: the AI never polls. The panel watches — 40 triggers across panel events, thresholds, log patterns and five-minute probes — and an agent run happens only when one fires. A quiet server costs nothing.
Two runs from our own test server, with the timings as measured:
How much you let it do
Three levels, and the limit is enforced by the agent's own command line, not by asking it politely in a prompt:
Cost
You pay your agent provider, and only for runs that actually happen — no Panelica charge for the model, no per-token markup, no background polling inflating the bill.
Getting started
Tools → Ops AI, pick an agent, press Install (a background job with a live step list — on a clean server it takes a few minutes), then sign in. Sentinel is the first tab of the bar at the bottom of the dashboard; it starts switched off and asks you to choose a level before it does anything.
The long version, with the same examples: https://panelica.com/blog/opsai-cookbook-six-real-server-jobs-from-prompt-to-result
If you want a setup for a specific case — which triggers, which level, which expert — reply here with the case and we will answer with the concrete configuration rather than generalities.
Two things first. OpsAI drives your own agent — Claude Code, Codex CLI, Gemini CLI or Cursor Agent — installed from the panel and signed in with your account; we do not run a model for you. And the agent is given the panel's own API (an index of 1388 endpoints across 93 categories), so it changes things the same way the interface does instead of guessing at config files.
1. "example.com returns 502. Find out why and fix it."
It reads that domain's error log and PHP-FPM log through the panel, finds the pool refusing connections at its process limit, raises the limit and reloads PHP-FPM, then reports what it changed with the log lines that led it there.
Check afterwards: the new pool value under Domains → PHP Settings — and whether the limit was the cause or a symptom. If one slow script filled the pool, raising the ceiling only buys time, and the agent usually says so.
2. "Move this account from our old cPanel server."
Pick the Migration expert first — it appends a role brief to the agent's system prompt, so it follows Panelica's pipeline instead of inventing one: discovery, credentials, files, databases, mail, SSL, verification. Database password hashes transfer as hashes, so applications keep working without anyone resetting passwords.
3. "Why has the server been slow since this morning?"
With the Performance expert you get a ranked answer — which account, which process, which query — rather than a wall of statistics.
4. From your phone: "disk is filling up, what can I safely remove?"
The Telegram bridge: you message your own bot, the agent answers there and the conversation resumes where it left off. Backups still inside their retention window come back as not safe to delete. The useful moment for server work is rarely when you are at a desk.
5. A page that only misbehaves in a browser
A checkout failing on step three, a redirect loop that needs a cookie — none of it is in the logs. OpsAI ships a real Chromium on the server, shown inside the panel, and the agent can open the page and click through it. For "works with curl, fails in a browser", that is the difference between a guess and an answer.
6. The job you do not ask for: Sentinel
Same agent, but it is not waiting in a chat window. It is the guard on the dashboard's bottom bar, and the rule is: the AI never polls. The panel watches — 40 triggers across panel events, thresholds, log patterns and five-minute probes — and an agent run happens only when one fires. A quiet server costs nothing.
Two runs from our own test server, with the timings as measured:
- An
[emerg]line in nginx's error log — queued 2 seconds after it was written, diagnosed and closed 23 seconds later. The agent noticed the line was an injected test and lowered the severity itself. - A probe found 12 failed systemd units — 48 seconds later a report was waiting with a proposed action. Waiting, not done, because the level was inform. Steps outside the allow-list were listed as suggestions; one button hands exactly those back to the agent for a single authorised run.
How much you let it do
Three levels, and the limit is enforced by the agent's own command line, not by asking it politely in a prompt:
- inform — read-only diagnosis; anything it wants to do waits for your approval.
- safe — read-only diagnosis, and it may request actions from a fixed allow-list which the panel executes and then verifies.
- full — the agent acts itself.
Cost
You pay your agent provider, and only for runs that actually happen — no Panelica charge for the model, no per-token markup, no background polling inflating the bill.
Getting started
Tools → Ops AI, pick an agent, press Install (a background job with a live step list — on a clean server it takes a few minutes), then sign in. Sentinel is the first tab of the bar at the bottom of the dashboard; it starts switched off and asks you to choose a level before it does anything.
The long version, with the same examples: https://panelica.com/blog/opsai-cookbook-six-real-server-jobs-from-prompt-to-result
If you want a setup for a specific case — which triggers, which level, which expert — reply here with the case and we will answer with the concrete configuration rather than generalities.