Ask
You type in plain English. A permission gate checks the scopes your message touches before anything else runs.
Admin Chat AI · operate and engineer
Chat with your backend. Run it and engineer it.
Drop the Admin Chat AI into your backend and just talk. Ask billing, tickets and CRM in plain English, then have it run real operations like refunds, orders and due dates as signed, approval-gated cards. Connect a server and the same chat becomes a hands-on engineer. It SSHes in, reads and edits your code, runs commands and migrations, watches server health, and scans for security holes. Every change is backed up first, waits for your yes, and rolls back in one click. It never invents an operation, and every move is written down.
Signed connector actionsSSH server engineerBacked up & reversibleEvery move audited
Refund invoice #4021 for Dana and email her the confirmation
Found #4021 · $180 · Dana Okafor, status Paid. Here is the refund for your ok.
refund_invoiceneeds approval
ClientDana · #5140
Amount$180.00
permission ✓HMAC ✓
ConfirmCancel
Checkout has 500'd since last night's deploy. Find the cause.
✓
grep -R "Fatal error" storage/logs | taildonereading CheckoutController.php:142 run_commandmedium
php artisan migrate --force
DenyAllow once
what is this error screen telling me?
I read your screenshot. That is a 502 from nginx because php-fpm is down. Checking the pool now.
✓
systemctl status php8.2-fpminactive
Safe by design
An AI that can act, without acting up.
Operatebilling, tickets, CRM, connectors
EngineerSSH, code, health, security scans
5 modesplan, ask, accept, auto, bypass
Reversiblebacked up and fully audited
See it work
One question. One command. One signed, audited action.
This is the live admin console: ask in plain English, get a grounded answer with sources, then watch a real operation turn into a signed card that waits for your approve, and lands in the audit log.
What's overdue this week, and refund Adam's last invoice?
you · admin
Two invoices are overdue, #8412 ($1,460) and #8390 ($240), both 40+ days. Adam's last invoice is #8421 · $240, currently Paid.
📊 Billing · 2 overdue👤 Adam #5512cited from live data
refund_invoiceneeds approval
ClientAdam · #5512
Amount$240.00
Invoice#8421 · last paid
permission ✓ ownerHMAC-SHA256 ✓reversible
✓ Refunded $240 to Adam · executed via connector, logged to AI History (prompt · response · result).
Nothing ran until a human clicked Confirm. The AI proposed; you approved; the connector executed; AI History recorded it.
Built-in engineer
It reads the code, writes the fix, and ships it signed.
Give it a server and the same chat turns into a real engineer. It greps the codebase, opens the file, writes the patch, runs the build, then checks the page is healthy again. Every write is backed up first, waits for your yes, and rolls back in one click.
Checkout has 500'd since last night's deploy. Find it and fix it.
OpsIQ agent is working...
The deploy renamed the cart key, so
items_total comes back null and total() throws. I will grep the call, read the method, and guard it.✓ Searching
"items_total"CheckoutController.php:142✓ Reading
CheckoutController.phpdoneapp/Http/CheckoutController.phpedit_file
141 public function total(array $cart): float {
142 return $cart['items_total'] + $cart['tax'];
142 return ($cart['items_total'] ?? 0) + ($cart['tax'] ?? 0);
143 }edit_file · CheckoutController.phpneeds approval
backed up ✓reversibleHMAC ✓
DenyAllow once
✓ Applied, reloaded php-fpm, and
/checkout returns 200. Logged to AI History, one click rolls it back.Ask OpsIQ about this admin area↵
Mode
Ask permissions1
Accept edits2
Plan mode3
Auto mode4 ✓
Bypass permissions5
Add or connect
🔌Connect server (SSH)Add and manage the servers the agent can work on.
🛡Security scansSchedule read-only vulnerability and health reports.
📊Server monitorsDaily, weekly or monthly performance and health reports.
Work on server
StandardGeneral assistant, everything except servers
web-01your connected server✓
AI model
DefaultWorkspace default model
gpt-5.4OpenAI✓
gpt-5.5OpenAI
claude-opus-4-8Anthropic
claude-sonnet-4-6Anthropic
gemini-3.5-flashGoogle Gemini
self-hostedyour endpoint
Thinking Auto
FasterSmarter
How hard the AI thinks before answering. Auto uses your AI Config default.
Nothing touched the server until you approved. It backed the file up first, made the change, proved the page was healthy, then wrote it all down.
It sees the whole box
Ask it to check the server. It scans, scores and tells you what to fix.
One line, "scan web-01 and check its health", runs a full security audit and a health check, each scored, with the risks ranked and a fix-this-first list. No hand-run commands, no guessing.
security_scan17 checks
72/100
C+posturereport saved
- Root SSH login is enabledhigh
- Port 3306 is exposed to the worldhigh
- TLS cert expires in 6 daysmedium
- 3 world-writable files in web rootmedium
server_monitor94 / 100 ▲
TLS valid12 updates0 OOM eventsno auth failures
▲ Up 6 points since the last check, disk is the one to watch.
roadmap Fix first: disable root SSH, close 3306, renew TLS. It can open a card for each, backed up and reversible.
Both reports are saved to the workspace, so you can open the full findings any time, not just the summary here.
How a message becomes a safe action
Five steps, every single turn.
The same path runs behind every message, whether it is a refund or a code change, so power never costs you control.
Ground
It reads what it needs, your live billing, tickets and CRM, or your files, logs and server health, filtered to what you are allowed to see, and answers from real data.
Prepare
For anything that changes something, it lines up the exact move: a registered action card, or a file edit, command or migration. It backs up first and cannot invent an operation.
Approve
Your permission mode decides. Ask queues a signed card for your Confirm, accept and auto apply low-risk work and still card the risky, plan changes nothing at all.
Audit
It runs, then the whole turn, prompt, reasoning, the diff or parameters, and the result, goes to an owner-only AI History. One click rolls a server change back.
What it can run on your platform
Real operations, not just answers.
Dozens of built-in operations, plus every action your connectors register. Stripe, Shopify, WHMCS or OpsIQ SaaS get auto-discovered into the catalog the moment a connector is enabled.
💳
Billing
Create, refund, cancel and mark invoices paid or unpaid, add credit, capture payment, change due dates.
🧾
Orders & quotes
Accept, create or cancel orders, or spin up a quote, straight from the conversation.
🛠️
Services
Suspend, unsuspend, terminate or upgrade a service, and change its price or due date.
🌐
Domains
Toggle the registrar lock, push a renewal date, run a live WHOIS or availability check.
👤
Clients
Reset a password, update an email, add an account note, look a client up by anything.
🎫
Tickets & comms
Draft and send a reply, open or close a ticket, email a client, prepare a campaign.
📝
Content
Draft a knowledge-base article, or publish a server or business announcement.
🔌
Your connectors
Every action a connector registers, from Stripe, Shopify, WHMCS or OpsIQ SaaS, joins the catalog automatically.
What it can build and fix
Point it at a server. It works like an engineer.
Connect a box over SSH and the same chat becomes a hands-on server, web and security engineer. Read-only work runs on the spot. Anything that changes the server is backed up first and follows your permission mode.
🔐
SSH & shell
Runs commands on your connected server. Read-only checks come back inline, anything that changes it waits for approval.
🔎
Read & search code
Greps the whole codebase, opens files, walks the tree, so it finds the real cause before it changes a thing.
✏️
Edit & write files
Surgical find and replace, or a full rewrite. Backed up first, reversible in one click.
⚙️
Commands & jobs
Installs, builds, migrations and backups run in the background as a job you watch to the exit code.
🌿
Git & cron
Status, diff, commit, push, checkout, and scheduled jobs, handled as clear first-class moves.
📈
Server health
A full health check with a score out of 100 and a trend: CPU, memory, disk, services, TLS, updates, log errors.
🛡️
Security scan
A 17-check audit with a posture score, a letter grade, and a fix-this-first roadmap you can open.
🖥️
Real terminal
Opens an interactive terminal for wizards like certbot, answering each prompt step by step.
More than a command runner
It sees what you paste, shows its work, and remembers.
The parts that make it feel like a teammate, not a text box. It reads your screenshots, narrates each step as it goes, and keeps what it learns.
👁️
Reads your screenshots
Paste an error screen, a dashboard, a chart or a log. It reads the image and reasons about what it actually shows.
⚡
Shows its work live
It streams each step as it happens, the command, the check, the result, so you watch it think instead of waiting on a spinner.
↩️
Obeys your corrections
Say "not that, do this" halfway through and it drops the old track at once and does exactly what you asked instead.
🧠
Remembers what matters
Per workspace and per server: your corrections, log paths, stack versions and restart commands, so it stops re-asking.
⚡
One-click questions
A row of chips for what you ask daily: revenue, overdue invoices, open tickets, expiring services, errors, activity.
📝
Writes KB articles
Ask for a help article and it drafts a full, structured piece, ready to review and publish to your knowledge base.
Safe by design
Power an admin can trust. Guardrails an owner can prove.
An AI that can write to your business, and to your servers, has to be governed like one. OpsIQ holds every change to the same discipline: permission-scoped, signed or backed up, approved by your rules, then logged. The model proposes, the rules decide.
It can't invent operations. Every platform call must match a registered, role-checked action contract, and an unknown key is rejected.
Backed up before every write. Server file changes are saved and drift-checked first, so one click restores them, in reverse order.
You hold the mode. Plan, ask, accept, auto or bypass decides what runs on its own and what waits for your Confirm. Writes are signed, and a tampered card is rejected.
Owner-only AI History. Every turn's prompt, command, diff and result is logged, searchable and exportable.
0off-policy ops
2 checksrender + execute
Fail-closedmulti-tenant isolation
SSRF-guardedpinned, vetted endpoints
Owner-gatedhistory non-owners can't see
Lives where you work
One snippet. Inside the backend you already use.
The Admin Chat AI is a widget you embed, whether that is your own admin dashboard, your OpsIQ workspace, or cross-origin into WHMCS, cPanel, WordPress or a custom panel. Identity is resolved through the connector: the loader carries an HMAC-signed token your platform mints, and the connector confirms which admin it is and what they're allowed to do. No host database credentials, no copy-pasting context.
- One-line loader. Drop it on the page and it mounts in your backend, same-origin or cross-origin.
- Identity through the connector. WHMCS, for example, confirms the admin via its own API, so the AI acts as the right person.
- Runs anywhere. Your admin dashboard, your OpsIQ workspace, remote and self-hosted, signed cross-origin embeds.
Your model, your data
Bring the brain you trust.
Your modelClaude, GPT, Gemini, Grok or self-host
Per surfacedifferent model for admin vs chat
Never trainedon your data, BYOK keys stay yours
2 wayscloud or self-hosted, same path
Full feature list
Everything in Admin Chat AI.
| Feature | What it does |
|---|---|
| Ask & ground | |
| Plain-English ops queries | Ask billing, tickets, CRM, orders, domains and visitors in natural language. |
| Cited sources | Every fact points back to the invoice, ticket or contact row it came from. |
| Live re-read | Context is re-read on every question, so real numbers every time. |
| Correction memory | The brain learns your corrections per workspace and applies them next time. |
| Act safely | |
| Registered action contracts | Dozens of built-in operations plus every connector-registered action. |
| Approval cards | Writes become signed cards; a human Confirm executes, Cancel dismisses. |
| Permission scopes & role gating | Nine scopes; data and tools filtered to what the admin may see and do. |
| HMAC-signed payloads | Tampered requests are rejected at the signature check. |
| Idempotency guard | Warns if the same action already ran on the same entity moments ago. |
| Engineer your servers NEW | |
| SSH server engineer | Connect a server and the chat runs commands, reads and edits code over SSH. |
| Read, search & edit code | Grep the whole codebase, open files, surgical find-and-replace or full writes. |
| Background jobs | Long installs, builds and migrations run detached with a job you watch to the exit code. |
| Git & cron | Status, diff, commit, push, checkout, and scheduled tasks as first-class operations. |
| Interactive terminal (PTY) | A real terminal for step-by-step wizards like certbot or mysql_secure_installation. |
| Security scan | A 17-check audit with a 0 to 100 posture score, a grade and a remediation roadmap. |
| Health monitoring | CPU, memory, disk, services, TLS and updates, scored with a trend against the last check. |
| Permission modes | Plan, ask, accept, auto or bypass govern how much runs without your Confirm. |
| Backup & one-click rollback | Every server write is saved first and reverts in one click, in reverse order. |
| Server memory | Remembers log paths, stack versions and restart commands for each server. |
| Parallel investigation | Fans out read-only helper agents to look into several things at once. |
| Deploy & govern | |
| Embed loader | One snippet mounts the widget in your backend, same-origin or cross-origin. |
| Identity via connector | The connector confirms the admin and permissions through a signed token. |
| Owner-only AI History | Exact prompt, response, result, provider, model and duration, searchable and exportable. |
| Bring your own model | Claude, GPT, Gemini, Grok or self-hosted, switchable per surface; never trains on your data. |
| Cloud or self-hosted | Identical code path, run on our cloud or your own servers. |
FAQ
Admin Chat AI, answered.
A chat widget you embed in your own backend: your admin dashboard, your OpsIQ workspace, or cross-origin into WHMCS, cPanel, WordPress or a custom panel. You talk to it in plain English. In standard mode it runs your platform, so billing, tickets, CRM and any action your connectors register. Point it at a connected server and it becomes a hands-on engineer that reads and edits code, runs commands, monitors health and scans for security issues.
Yes. Connect a server over SSH and the same chat remotes in. It reads files and greps the codebase, writes and edits code, runs commands and long jobs in the background, manages git and cron, opens a real terminal for wizards, runs a full security scan and a health check with a score, and remembers what it learns about the box. Read-only work runs on the spot. Anything that changes the server is backed up first and follows your permission mode.
Yes. Paste a screenshot straight into the chat, an error screen, a dashboard, a chart, a stack trace, a config, and it reads the image and reasons about what it shows, then acts on it. You can send an image with no text at all and just ask what is wrong. It is the fastest way to hand it a problem you are looking at.
That is the whole design. Every write is backed up before it runs and rolls back in one click. You pick the permission mode: plan looks only, ask queues an approval card for each change, accept or auto apply the low-risk changes and still card the risky ones, and bypass runs freely. Nothing state-changing happens unless your mode allows it, and every prompt, command and result lands in an owner-only AI History.
Read your live billing, tickets, CRM, orders, services, domains, products and logs, then run the actions your connectors register: refunds, invoices, orders, service changes, due dates, password resets, ticket replies, KB articles and more. Stripe, Shopify, WHMCS, OpsIQ SaaS and any other connector are auto-discovered, and writes come back as signed cards you approve.
Identity is resolved through the connector. The loader carries an HMAC-signed token your platform mints, and the connector (for example WHMCS) confirms the admin and their permissions, so the AI always acts as the right person with no host database credentials.
Yours: Anthropic Claude, OpenAI GPT, Google Gemini, xAI Grok or a self-hosted endpoint, switchable per surface. OpsIQ never trains models on your data, and it runs the same on our cloud or self-hosted.
You add a server connection once, an SSH host and key, and pick it from the composer's server selector. From then on the chat is remoted into that one box and works from your web root. You can connect more than one and switch between them. It never touches a server you have not connected.
Every file it writes is backed up first. One rollback restores every file it changed in that session, in reverse order. For a command-only change, like a service restart, it verifies the result and will work the fix with you if the check fails. You are never left guessing whether it worked.
Yes. Long jobs, installs, builds, migrations, backups, run detached in the background and return a job id. It watches the job to the exit code and reports what actually happened, so a slow task never times the chat out.
No. Who the admin is comes from the connector's signed token, and server access is the SSH connection you set up. There are no host database passwords to hand over and nothing to copy and paste into the chat.
That is the permission mode, and you set it per session. Plan looks only and changes nothing. Ask queues a card for every change. Accept and auto apply the low-risk changes and still card the risky ones. Bypass runs freely for a trusted, hands-on session. You can dial it up or down at any point.
In the owner-only AI History. Every turn is recorded with the exact prompt, the reasoning, the commands and diffs, the result, the model and how long it took. It is searchable and exportable, so an audit is a filter, not a favour.
No. That is the Client Chat AI, for your visitors and customers. This one lives in your backend for your team, and it can run operations and engineer your servers. Different audience, different powers.