A sealed workspace
Each client gets its own isolated workspace: its own files, memory, skills and logins. Clients can never see each other's data.
Know your stack is working. Every day.
Everything your automations need to run for years, not weeks
StackProbe takes what you've built, wherever you built it, and gives it an always-on home with rules, records and people watching. Here's the full picture.
Run
Your jobs run on schedule in a workspace that's yours alone, whether or not anyone's computer is on.
Skills: saved jobs that run themselves
A skill is a job StackProbe knows how to do, such as a morning brief, inbox triage, quote follow-ups, an unpaid-invoice chaser, meeting prep, a weekly business pulse or a low-stock watch. Skills run on a schedule or when someone asks.
Skills the AI suggests
When the AI notices a task repeating, it can draft a new skill for it. Drafts wait for a person to approve them before they ever run.
Company memory
What the AI knows about your business, in one place. Staff or the AI can suggest a new fact; it only counts as true once a person approves it, and every fact records where it came from.
Managed scripts
Already have code? Python, Node, file transfers, finance exports: they keep running as code, on schedule, with the same monitoring and the same record as everything else.
Cost control
Quick questions go through a light, cheap lane; real work goes through a full one. Caps per company and per person stop spending from running away.
Connect
StackProbe plugs into the systems you already use, with your own accounts. Nothing to install.
- Microsoft 365
- Google Workspace
- Shopify
- QuickBooks
- Stripe
- Jobber
- Slack
- GitHub
- SuperOps
- Google Search Console
- GA4
Your systems
Connections use your own accounts, so access stays yours. See the list above; more are added as clients need them.
Three doors, one AI
Staff reach StackProbe from Microsoft Teams, by email (send a request to your company's AI mailbox), or from their own Claude app, signing in with their work account.
Everyone's AI knows the same things
Every door reaches the same memory and the same skills, so the answer in Teams matches the answer by email.
Govern
Every action the AI takes is checked against your rules before it runs. You decide how much rope to give it.
Rules, per system and per action
Written in plain English and enforced on every action. For example: Shopify may create draft orders but never cancel one. Read-only skills can never send or change anything.
Approvals where people already look
A Teams message with Approve, Edit and Reject. A draft waiting in the mailbox. Or a daily digest. Nothing goes out in Assisted mode until a person says so.
Least access
Each skill only touches the systems it says it needs, and nothing else.
A tamper-evident record
Every run, every action and every change made in another system is written to an audit ledger. Entries are chained together, so a deleted or edited line can be detected.
A kill switch
One click stops everything for your company.
Loosening a rule is a big deal
Making any rule less strict takes an extra confirmation and raises an alert, so it can't happen quietly.
Watch
Our team, and the system itself, check on your stack constantly.
Health checks every five minutes
Deeper checks twice a day.
Alerts to us, not you
Failed jobs, spending near a cap, broken connections and backup trouble all alert our team. We respond and fix them.
Encrypted off-site backups
Every few hours.
Secrets stay secret
Passwords and keys are encrypted and unlocked only for the moment a job needs them. They never appear in chat, in files or on a laptop.
You hear about it in the morning
A morning digest of what happened, and a monthly summary of what ran and what it cost.
Your stack, on one screen
The client console is a secure web app where you see everything StackProbe runs for you. It works on a phone.
Home. Is everything healthy? Green, amber or red for every job and every connection.
Jobs6
Waiting for you2
Actions recorded today148
Spend, month to date46% of cap
| Job | Status | Last run |
|---|---|---|
| Morning brief | Healthy | Ran today, 7:00 |
| Inbox triage | Healthy | Running now |
| Quote follow-ups | Healthy | Ran today, 8:15 |
| Low-stock watch | Needs attention | Ran today, 6:00 |
| Unpaid-invoice chaser | Healthy | Next: Monday, 9:00 |
| Weekly business pulse | Healthy | Next: Friday, 16:00 |
Activity. What ran, when, what it did, and what it cost.
Approvals. Everything waiting for a person, with Approve, Edit and Reject.
Waiting for you3 items · Assisted mode: nothing goes out until you approve it
- Quote follow-ups8:15 todaySend email to a customerFollow-up on quote #Q-1182 (bathroom reno), 3 days since sent.Via Microsoft 365
- Unpaid-invoice chaser9:00 todaySend reminder emailFriendly reminder for invoice #4471, 14 days overdue.Via QuickBooks + Microsoft 365
- Low-stock watch6:00 todayCreate draft purchase order40 × ¾" copper elbows from the usual supplier. Draft only; nothing is sent.Via QuickBooks
Record. The audit ledger, searchable. Who or what did this, and when?
RecordEvery run, action and change. Entries are chained; an edited or deleted line would show here.
| Time | Who / what | Action |
|---|---|---|
| 08:15:04 | Quote follow-ups | Drafted email to customer (quote Q-1182) |
| 08:15:06 | Quote follow-ups | Sent for approval → Dana R. |
| 08:22:41 | Dana R. | Approved: send email (quote Q-1182) |
| 08:22:42 | Quote follow-ups | Email sent to customer |
| 07:00:12 | Morning brief | Read 14 calendar events, 31 emails |
| 07:00:58 | Morning brief | Posted brief to #ops |
| 06:00:03 | Low-stock watch | Blocked: "cancel purchase order" not allowed by rule |
Rules. What the AI may do, per system and per action, in plain English.
RulesWhat the AI may do, per system. Loosening a rule needs a second confirmation and raises an alert.
- QuickBooksMay
- Read invoices and customers
- Create draft purchase orders
- Send invoice reminders (with approval)
Never- Delete or void anything
- Change bank details
- ShopifyMay
- Read orders and stock
- Create draft orders
Never- Cancel an order
- Issue a refund
- Microsoft 365May
- Read mail and calendar
- Draft replies
- Send email (with approval)
Never- Delete mail
- Change mailbox rules
Memory. What the AI knows about your company, plus suggestions waiting for approval.
Spend. Month to date, by job and by person, against the cap.
SpendMonth to date, against the company cap. Figures shown as a share of the cap.
Company cap used46%
Days left in month14
On track to finish at≈ 88%
Connections. Each system, its status, and what StackProbe is allowed to do there.
Reports. The monthly summary, and export of everything.
Illustrations with invented data. The console is part of the managed service; you never have to run anything yourself.
Two ways to run
Assisted
The AI reads and drafts. A person approves anything that goes out.
Autonomous
Once a job has proven itself and its scope is written down, the AI acts on its own within limits you set. You can move jobs between plans one at a time.
Both: a fixed monthly fee plus usage, with caps you set.
We take it on
A free stack check finds what you already rely on and how fragile it is. We move automations over one at a time, running each beside the original until it's proven. Your data stays yours: export files, memory, skills and history at any time.
Questions
Is my data safe?
Each business gets its own sealed workspace, kept apart from every other client. Passwords and keys are encrypted and unlocked only for the moment a job needs them; they never appear in chat, in files or on a laptop. Everything is backed up off-site every few hours.
Do I need to install anything?
No. Your staff use StackProbe from tools they already have: Microsoft Teams, email, or their Claude app, signing in with their work account.
What if I already use Zapier or n8n?
Keep using them. StackProbe isn't a tool for building automations, and it isn't a chatbot. Build wherever you like. StackProbe gives what you've built an always-on, governed home and takes responsibility for it.
Does the AI act on its own?
Only if you choose that. Every client starts in Assisted mode, where a person approves anything that goes out. Once a job has proven itself and its scope is written down, you can move it to Autonomous mode, where it acts within limits you set.
What does it cost?
A fixed monthly fee plus usage, with caps you set. Book a free stack check and we'll quote based on what you actually run.
Who fixes it when something breaks?
We do. Health checks run every five minutes. Failures alert our team, and we respond and fix them, usually before you notice.
Who owns the automations once they're on StackProbe?
You do. The skills, scripts, memory and history in your workspace are yours. Export everything at any time.
Can my staff ask it questions, or does it only run scheduled jobs?
Both. Anyone with a work account can ask from Teams, email or their Claude app. Quick questions use a light lane, so they're fast and cheap.
What happens to my existing scripts?
They keep running as code, on schedule, inside your workspace, with the same monitoring and record as everything else. Nothing has to be rebuilt.
Can the AI spend money?
Only within caps you set, per company and per person. In Assisted mode, anything that costs money or leaves the business needs a person's approval first.
Find out what's fragile, free
Answer a few questions about what runs automatically in your business. About five minutes.
Start the free stack check