Skip to main content

Security: is it safe to let AI control your computer?

LapuAI works in your files, apps and browser. This page explains what it can reach, when it asks you first, what it sends to AI providers, and how to check what it did.

What LapuAI can reach on your computer

LapuAI is a desktop AI agent: an AI app that works in your apps, files and browser for you.

What LapuAI can do in each app depends on three things: the app, your operating system permissions and the permission mode you choose.

During a task, LapuAI can:

  • Read, search, create and edit files on your computer.
  • Run commands on your computer.
  • Work on websites in its own built-in browser. Its sign-ins are separate from your everyday browser.
  • Click and type in desktop apps on Windows and macOS. Some of these actions use your mouse and keyboard while they run.

For task examples, see AI agent use cases.

When LapuAI asks before it acts: Basic and Full access

Every Chat runs in one of two permission modes. You pick a default in Settings, and you can change the mode for a single Chat.

  • Basic mode asks before risky or outside actions. You can approve one action. You can also approve that action type on the same target for the remaining Chat. Deleting and other destructive actions ask again each time.
  • Full access skips the ordinary approvals. The checks marked “Asks” in both columns below still apply.

If you turn the built-in browser off, it stays off in both modes.

Which actions ask for your approval in Basic and in Full access
What the agent doesBasicFull access
Ordinary file, command and app actionsAsks for risky or outside actionsRuns without asking
Buy, delete, send, submit a form, upload a file, or change sharing or permissions on a websiteAsks every timeRuns without asking
Send again on a site after a send that may already have gone outAsksAsks
Open a website you have not allowed in the built-in browserAsksAsks
Install Python packages on your computer, or run a command LapuAI cannot fully checkAsksAsks
Get operating system permissions, such as Accessibility on a MacYour system asksYour system asks

LapuAI cannot yet recognize a send in a desktop app, such as an email app. In Basic mode, that send falls under the app's ordinary approval. In Full access, it does not ask.

Use Basic mode for new tasks, for important accounts and for anything that reads messages or pages from people you do not know.

What stays on your computer, and what is sent to AI providers

Stored on your computer

  • • Your Chats
  • • Your Workflows and their run history
  • • Your memories
  • • Built-in browser sign-ins, kept apart from your everyday browser

Sent when a step needs it

  • • Your instructions and the relevant Chat history
  • • Content from documents or spreadsheet cells
  • • Screenshots the agent takes
  • • Other task context, such as text from a website

These requests go through LapuAI's cloud to hosted AI providers. The AI uses them to plan the next step. LapuAI does not promise a fixed model or provider, because it can switch which provider handles a request.

LapuAI needs the internet to send model requests, so it is not an offline app. It is also not an air-gapped system, a computer kept off every network.

What diagnostics include on each plan

Diagnostics are reports that help LapuAI find bugs and improve the app. Full diagnostics are on by default on every plan.

The diagnostics setting changes diagnostics only. The AI still receives the task context it needs to do the work.

  • Free plan. Full diagnostics are required. LapuAI can collect and store Chat and tool content as part of them.
  • Paid plans. Full diagnostics stay on until you switch them to anonymized in Settings. LapuAI then keeps only anonymized product events and stops sending crash reports. Product diagnostics never stop completely.

How long LapuAI keeps data depends on the type of data. See the privacy policy. To ask us to erase your account data, write to privacy@lapu.ai.

How to check what the AI agent did

You check a task in the Chat where it ran.

LapuAI keeps logs of its steps, but it has no separate audit log and no setting for how long a log is kept. If your company needs a central log of every agent action, LapuAI does not provide one today.

In the Chat, three things help:

  • Work details. An answer shows the steps the agent took, including steps that failed.
  • Saved files. Each answer lists the files the agent saved. The Files button lists every file and browser download in the Chat.
  • Chats on your computer. Chats are stored locally, so you can reopen a past task and read it again.

Which parts run on your computer and which go to the cloud

LapuAI does the hands-on work on your computer. It reads and edits files, runs commands and works in your apps there. The thinking step runs on hosted AI models, reached through LapuAI's cloud, so the app needs the internet.

On the Enterprise plan, a company can use its own AI models in LapuAI, including self‑hosted ones. LapuAI's backend stays hosted, so that setup is not offline either. See LapuAI for teams.

If you need AI that works with no internet at all, LapuAI is not the right tool today. Apps such as Ollama and LM Studio run AI models you download on your own computer.

What no AI agent can fully prevent

The biggest security risk for an AI agent is prompt injection: hidden instructions inside content the agent reads. That content can be a web page, a PDF, an email or the output of a command. OWASP ranks prompt injection as the top risk for AI language model apps in 2025 (OWASP LLM Top 10, 2025). The AI cannot reliably tell those instructions apart from yours.

LapuAI cannot fully prevent prompt injection. In Basic mode, the request before each purchase, deletion or send on a website is your chance to stop a wrong action. In Full access, those requests do not appear. Check important results yourself before you rely on them.

Certifications and security reviews

LapuAI does not hold a SOC 2 report today. This page claims no security certification.

Single sign-on (SSO) is available for teams. To set it up, book an intro call on LapuAI for teams.

To start a security review, or to report a security problem, write to security@lapu.ai. For plans and prices, see LapuAI pricing.

Questions to ask any AI agent vendor

A vendor that calls its AI agent “secure” has not given you evidence. Ask instead about controls you can see and test.

A useful approval request is specific. “Delete 12 files from Downloads/receipts?” names a target and a scope. “Allow computer access?” does not give you enough to decide.

Whether you review LapuAI or another product, ask:

  • Does each request show the exact file, command, website or app action before it runs?
  • Which actions ask every time, and which can run without asking?
  • Where are files, screenshots and instructions processed, and what do diagnostics keep?
  • How can you check afterwards what the agent did?

Try LapuAI in Basic mode

LapuAI runs on 64-bit Windows 10 or later and on Apple Silicon Macs with macOS 14 or later. Start with a small task in Basic mode and watch what it asks. Download LapuAI.

Frequently asked questions

Is it safe to let AI control my computer?
Limits and your checks reduce the risk, but they cannot remove software errors or hidden instructions in content. In Basic mode, LapuAI asks before risky or outside actions, such as a purchase or a send on a website. In Full access, it skips ordinary approvals, but some actions still ask. Review important results before you rely on them.
What can LapuAI access on my computer?
During a task, LapuAI can read, create and edit files and run commands. It can also work on websites in its own built-in browser. On Windows and macOS, it can click and type in desktop apps. The permission mode, your operating system permissions and the app decide what LapuAI does without asking.
How can I limit what the AI agent is allowed to do?
Use Basic mode. It asks before risky or outside actions. You can approve one action, or that action type on the same target for the rest of the Chat. Deleting and other destructive actions ask again each time. Set Basic mode as your default, and change the mode for a single Chat when you need to.
What data leaves my computer?
To plan each step, LapuAI sends task context through its cloud to hosted AI providers. That can include your instructions, Chat history, document or spreadsheet content, screenshots and website text. With full diagnostics on, LapuAI can also collect and store Chat and tool content as part of diagnostics. Your Chats, Workflows and memories are stored on your computer.
How do I check what an AI agent did?
Open the Chat where the task ran. An answer shows Work details: the steps the agent took, including steps that failed. The answer also lists the files the agent saved. The Files button lists every file and browser download in that Chat. Chats are stored on your computer, so you can reopen a past task.
Does LapuAI keep an audit log?
No. LapuAI keeps logs of its steps. It has no separate audit log and no setting for how long a log is kept. The record of a task is its Chat: the steps in Work details and the files the agent saved. If your company needs a central log of every agent action, LapuAI does not provide one today.
Which parts of LapuAI run on my computer?
LapuAI does the hands-on work on your computer: it edits files, runs commands and works in your apps there. Your Chats, Workflows and memories are stored on your computer. The thinking step runs on hosted AI models through LapuAI's cloud. LapuAI needs the internet and is not an offline app.
Can I use LapuAI offline or with my own AI model?
Not offline. LapuAI uses hosted AI models through LapuAI's cloud. On the Enterprise plan, a company can use its own AI models in LapuAI, including self‑hosted ones. LapuAI's backend stays hosted, so that setup is not offline either. Apps such as Ollama and LM Studio run AI models you download on your own computer.
What do LapuAI diagnostics include?
Full diagnostics are on by default, and they are required on the Free plan. With full diagnostics, LapuAI can collect and store Chat and tool content. On paid plans you can switch to anonymized diagnostics in Settings. LapuAI then keeps only anonymized product events and stops sending crash reports.
Is LapuAI SOC 2 certified?
No. LapuAI does not hold a SOC 2 report today, and this page claims no security certification. If you have a compliance requirement, write to security@lapu.ai to ask about the current status.
Can LapuAI take an action I did not approve?
Yes, in Full access. That mode skips ordinary approvals, but some checks still ask. LapuAI asks before it sends again on a website after a send that may have gone out. It asks before it opens a website you have not allowed. It asks before it runs Python or a package installer directly on your computer. Your operating system still asks for system permissions. In Basic mode, risky and outside actions ask first.

Try LapuAI on your next task

LapuAI is a desktop AI agent that works in your files, browser and apps, with you in control. Download the app, sign in and ask for the result you need.

  • Free plan
  • No credit card
  • Cancel anytime

No credit card needed. For 64-bit Windows 10 or 11 and Apple Silicon Macs on macOS 14 or later.