Short answer
An AI agent that uses software like a person logs into your tools on its own secure computer, reads the screen and clicks, types and downloads just as an employee would. That lets it work in systems that have no API, such as old portals and vendor sites. Oraculo does this, with approvals for important actions.
The software that has no door
Modern tools expose an API: a documented way for other software to read and write data. Plenty of the software that small businesses depend on does not. The state licensing portal. The supplier's ordering site. A practice-management system from 2012. The carrier's quote page. The county permit website. Your bank's business dashboard. These are built for humans with browsers.
Traditional automation needs an API or a fragile recorded script. If there is no door, the work stays manual: someone logs in, clicks through five screens, downloads a PDF and types a number into a spreadsheet. That is exactly the kind of chore people postpone.
How it works
Oraculo's employee has its own secure computer, separate from yours. It can open a browser or an application there, sign in with access you give it, look at the screen the way a person does, and operate it: click a button, fill a form, read a table, download a statement, upload a document.
- You tell it what you need in plain language: 'Pull last month's remittance reports from the supplier portal and match them to our invoices.'
- It logs in on its secure computer using the access you provided.
- It navigates the screens, reads what it finds and does the task step by step.
- It saves the result where you want it (email, a sheet, your accounting tool) and reports what it did.
- It remembers the route, so next month's version is faster.
Where a proper API exists, Oraculo can use that instead, which is more reliable. The screen-based approach is the fallback that makes the rest possible: if it has an API or a login, Oraculo can usually work with it.
Real chores this covers
| Chore | Why it is stuck manual | What the agent does |
|---|---|---|
| Supplier reorders | Portal only, no export | Logs in, checks stock levels you flagged, builds the cart, asks for your approval before ordering |
| Monthly statements | Download behind a login | Fetches PDFs, reads them, files them and records totals |
| Insurance or carrier quotes | Re-typing client data into forms | Fills each carrier form from one record and compiles results |
| Permit or license status | Checked by hand weekly | Looks up each record and reports changes |
| Old industry software | Built before APIs were common | Enters and retrieves data on screen, keeping the CRM in sync |
| Marketplace or platform dashboards | Limited export options | Reads orders and metrics into your weekly report |
Safety is the real question
Handing an agent a login rightly raises eyebrows. The honest framing: it is a decision about access, like giving a new hire a login, and the same habits apply. Use dedicated accounts with limited permissions where the software allows it. Give access only to what the task needs. Decide which actions need your approval.
- Approvals. Important actions can require your OK first: placing an order, sending money, submitting a form with legal weight.
- Separate computer. Its work happens on its own secure computer, not on your laptop.
- Data handling. Data is encrypted in transit and at rest, access is role-based and logged, it is never sold or used for ads, and you can export everything and leave.
- Two-step logins. Many portals send a code to a phone or email. Plan how verification codes will reach the agent, and where you want a person in the loop.
A deeper walk through the questions to ask is in can an AI employee log into my software safely and the AI employee security checklist.
A delegation example
The scope in that message matters: a read, a summary, and one low-risk email. Placing or canceling is excluded. Starting with read-only tasks and widening as trust builds is the sensible path, and it mirrors how you would train a new coordinator.
Limits, stated plainly
- Screens change. A website redesign can alter the path. The agent adapts to many changes, but you should expect the occasional need to re-check a flow.
- Slower than an API. Screen work takes minutes, not milliseconds. For a weekly or daily chore that is fine.
- Terms of service. Some sites prohibit automated access. Check the rules of the portals you point it at, and use legitimate access.
- CAPTCHAs and bot checks. Sites that deliberately block automation may require a person to step in.
- Judgment. It follows your instructions. It should not decide on its own what to buy, pay or submit.
Pricing and next step
Billing uses a published task-time chart: you pay for work done. Starter is $12 per hour with a 6 hour weekly minimum, $312 a month. Business is $8 per hour at 43 hours a week, $1,500 a month. Setup is $499 one-time with four 1-hour Zoom sessions with an engineer who connects your tools with you. A 30-day money-back guarantee applies.
This same capability sits underneath other jobs such as AI invoicing and supply reordering. List the three portals you dread opening, then book a 20-minute demo.
Key takeaway
If the only way into your software is a login screen, an agent that can use that screen turns manual chores into delegated ones. Start read-only, add approvals and widen from there.
Frequently asked questions
What does 'uses software like a person' mean?
It means the agent operates the screen: it logs in, reads what is displayed, clicks buttons and fills in forms, rather than connecting through a technical interface. That is why it can work in tools that have no API.
Does it run on my computer?
No. It works on its own secure computer, separate from yours. You give it access to the specific tools it needs and can require approval for important actions.
What about two-factor authentication?
Many sites require verification codes. Decide with the engineer during setup how those will be handled, for example a dedicated account or a person approving each login. Do not disable security to make it convenient.
Can it break my software?
Any user can make mistakes. Reduce risk by starting with read-only tasks, using limited-permission accounts and requiring approval before actions like orders, payments or deletions.
Is this allowed by the websites I use?
It depends on each service's terms. Some prohibit automated access. Review them for the portals you plan to use and use accounts you are entitled to access.