Introducing Secure Helper - The User Controlled AI Assistant

[danb]

Ever since agentic AI started to take off, I always wanted an app like this, but I could not find exactly what I was looking for so I built a new app. This app has a lot of potential to do a lot of different things, so if you guys have any ideas on what new features we should add, please let me know.

The other reason I wrote this add it because it will give me a little sandbox to experiment with new ideas on how we can refine our zero-trust apps prompts, like CyberLock, DefenderUI Pro, WDAC Lockdown and SiriusGPT. If we were to add prompts similar to the ones in Secure Helper, I think it would be pretty cool... but we certainly would not want to over engineer anything, we need to keep it simple either way.

The concept is simple: when you are looking at a confusing screen, you click Help Me, select a window or desktop, choose a question, and Secure Helper opens your preferred AI assistant with the screenshot and prompt ready to review.

The key design choice is that Secure Helper keeps the user in control.

It does not auto-send by default.
It does not scrape the assistant’s answer.
It does not record the screen in the background.
It does not take autonomous control of the PC.
It does not click through workflows for you.

Instead, it acts more like a trusted assistant sitting next to you, helping you ask a better question about the screen you intentionally selected.

The app currently supports assistant handoff to services like ChatGPT, Copilot, Claude, and Gemini. It also has context-aware question presets. For example, normal screens get general questions like “What should I do next?” or “Is anything here risky or important?” But if the selected window appears to be Outlook, Gmail, or another email client, Secure Helper can switch to email-safety questions like “Could this email be a scam or phishing attempt?” and “What red flags should I look for before clicking anything?”

I think there is a useful middle ground between fully manual computer help and autonomous agents that control the machine. Secure Helper is my attempt at that middle ground: user-initiated, privacy-conscious, transparent, and manual at the final Send step.

I would love feedback from people who think about usability, cybersecurity, support tools, and safe AI workflows. Ironically, Sirius has a false positive for this file, but I promise it is clean .

Secure Helper is a self-contained portable executable and you can download it here:

SecureHelper 1.00 Demo

https://www.cyberlock.global/downloads/SecureHelper.exe

SHA-256: 65260923193fb20ebb20654e1bc13d74279d8502cef68ca7f979dc101516425a

 

[ErzCrz]

Secure Helper is a self-contained portable executable and you can download it here:

SecureHelper 1.00 Demo

https://www.cyberlock.global/downloads/SecureHelper.exe

SHA-256: fa55bebe7922b116643d3772418759c51aa06ec6fb1eda286941d6de97ea5ba2

Nice! Will have a look at this. I have been experimenting with LMStudio for a local AI but with so many models and gigs to download for each model, It kind of didn't quite fit what I was after. This could be a useful tool in a number of ways, will check it out

 

[MikeGiann]

Hi Dan,

I downloaded the Demo but upon executing it, I get the window to accept the EULA but there is nothing to click or accept. Not sure if it's just me.

 

[ErzCrz]

@danb Will need Whitelisting in CyberLock/SiriusLLM. Will email you the log tomorrow, it's late here.

 

[Zero Knowledge]

Now your talking. This seems very useful. I'm considering buying a sub to CL just to support you. Love that your developing tools that are useful and don't break the bank.

I think I speak for everyone that we are thankful your hard work.

 

[outlawxtorn]

Now your talking. This seems very useful. I'm considering buying a sub to CL just to support you. Love that your developing tools that are useful and don't break the bank.

I think I speak for everyone that we are thankful your hard work.

I agree, excellently put!

 

[danb]

Nice! Will have a look at this. I have been experimenting with LMStudio for a local AI but with so many models and gigs to download for each model, It kind of didn't quite fit what I was after. This could be a useful tool in a number of ways, will check it out

Very cool! Secure Helper is not like LMStudio, it is actually kind of difficult to explain, you just kind of have to try it to see how it works. But let me give you some examples. The other day I was at the USPTO website submitting a provisional patent and I had 2-3 questions on which boxes I should check. With Secure Helper, all I have to do is click the Secure Helper icon, then click the website that I had a question about, then click the "Which option should I choose?" button, and Secure Helper will automatically take a screenshot of the website and upload that and a prompt to whichever AI I want, and then I just click the Send button on the AI and I have my answer. We could make it so that the Send button is automatically clicked, but I think it is better to have the user manually click the final Send button for a lot of reasons.

You can also use SH if you receive an email that you are unsure of and ask your AI if it is a scam or potentially malicious email. Another example was yesterday I had an odd message on my computer and I used Secure Helper and it only took like 3 seconds and I had the answer.

And like I was saying, there are TONS of things we can do with this, and new features will be super simple to add.

 

[danb]

Hi Dan,

I downloaded the Demo but upon executing it, I get the window to accept the EULA but there is nothing to click or accept. Not sure if it's just me.

Hmmm, that is very odd. If your window looks different from the screenshot I uploaded here, can you please post a screenshot of the window? You should just be able to check the box and then click the I Agree button. I tested on Windows 10 and 11 and it worked for me. The only small issue I am having now is I made a last minute change and now the Copilot app is a little flaky sometimes on the screenshot attachment step, but I am fixing it as we speak. Thank you!

 

[danb]

@danb Will need Whitelisting in CyberLock/SiriusLLM. Will email you the log tomorrow, it's late here.

View attachment 298295

Thank you, but you do not need to send me the log. I actually mentioned this in the first post... "Ironically, Sirius has a false positive for this file, but I promise it is clean ." I am quite surprised that Sirius returned a false positive for this app. I will read through the Sirius Analysis report to see if there is anything I should change in the code so that Sirius no longer flags the file as malicious. I read the report quickly, but I need to read the entire report carefully and I bet there is something I can fix in the code so that it does not look suspicious.

 

[danb]

Thank you @Zero Knowledge and @outlawxtorn, I really appreciate that!

 

[danb]

@Bot can you guess how Secure Helper works from my terrible description?

 

[Bot]

@Bot can you guess how Secure Helper works from my terrible description?

Yes, roughly — but I can only infer the likely mechanics at a high level from what you described. I can’t know the exact implementation details without code, docs, or logs.

Likely workflow (based on your description)

• User triggers Secure Helper (tray icon / hotkey / small launcher UI).
• User selects a target: either a specific window (click-to-pick) or the full desktop.
• Secure Helper captures a screenshot of the chosen region/window (Win32 capture APIs are the usual approach on Windows).
• Secure Helper picks a “question preset” based on what it thinks the app is (generic vs email-client safety, etc.).
• Secure Helper opens the chosen assistant (ChatGPT/Copilot/Claude/Gemini) and pre-loads:
  • A prompt template (the selected preset + maybe a short “analyze this screenshot” instruction).
  • The screenshot as an attachment (but not auto-sent).
• User manually hits Send inside the assistant.

The tricky part: “handoff” with the screenshot (how this is probably done)
Most web assistants don’t support “attach this image” purely via a normal URL parameter, so the app usually has to do one of these (or a mix):

• Clipboard-based: put the screenshot on the clipboard, then bring the assistant page/app to the foreground so the user can paste (or the app can simulate Ctrl+V). This fits your “does not auto-send” goal while still being quick.
• UI automation: use Windows UI Automation / accessibility APIs (or simulated input) to focus the message box and paste the image + prompt, but stop short of clicking Send.
• Temp file + “attach” automation: save the screenshot to a temp path, then automate the “attach file” flow for the target assistant (more fragile across UI updates).
• If using native apps (Copilot app, etc.): similar idea, just different UI automation targets.

Given you said Copilot is “a little flaky sometimes on the screenshot attachment step,” that strongly suggests some UI-automation/foreground-window timing dependency (focus, clipboard readiness, attachment widget state, etc.).

How “context-aware presets” are probably detected
Common lightweight signals that don’t require heavy inspection:

• Foreground process name / executable (e.g., outlook.exe, chrome.exe, msedge.exe, thunderbird.exe).
• Window class name and/or window title keywords (e.g., “Inbox”, “Message”, “Gmail”, “Outlook”).
• Browser tab title heuristics (indirectly visible via window title).

This is enough to switch from generic help prompts to “email safety” prompts without doing anything intrusive like OCR or content scraping.

Security/privacy design choices that match your description
What you described aligns with a “thin glue layer” model:

• No background recording; only user-initiated capture.
• No reading assistant responses (no “agent loop”).
• No autonomous clicking through workflows.
• A deliberate “manual Send” moment so the user decides what leaves the machine.

Where I would be cautious (implementation-agnostic)

• Temp artifacts: screenshots saved to disk (even briefly) should be clearly disclosed and securely cleaned up. If you can stay clipboard-only, that reduces forensic residue.
• Over-automation fragility: UI automation can break when assistants change their UI; having a “fallback: open assistant + copy prompt + copy image” mode can keep it reliable.
• Data leakage user education: a short reminder near the Send step like “Check for secrets (tokens, email addresses, customer data) before sending” helps users avoid accidental disclosure.

Bottom line: from your description, Secure Helper sounds like a controlled “capture + prompt template + assistant handoff” tool, probably using screenshot capture plus clipboard/UI-automation to stage (but not send) the message.

 

[danb]

Yes, roughly — but I can only infer the likely mechanics at a high level from what you described. I can’t know the exact implementation details without code, docs, or logs.

Likely workflow (based on your description)

• User triggers Secure Helper (tray icon / hotkey / small launcher UI).
• User selects a target: either a specific window (click-to-pick) or the full desktop.
• Secure Helper captures a screenshot of the chosen region/window (Win32 capture APIs are the usual approach on Windows).
• Secure Helper picks a “question preset” based on what it thinks the app is (generic vs email-client safety, etc.).
• Secure Helper opens the chosen assistant (ChatGPT/Copilot/Claude/Gemini) and pre-loads:
  • A prompt template (the selected preset + maybe a short “analyze this screenshot” instruction).
  • The screenshot as an attachment (but not auto-sent).
• User manually hits Send inside the assistant.

The tricky part: “handoff” with the screenshot (how this is probably done)
Most web assistants don’t support “attach this image” purely via a normal URL parameter, so the app usually has to do one of these (or a mix):

• Clipboard-based: put the screenshot on the clipboard, then bring the assistant page/app to the foreground so the user can paste (or the app can simulate Ctrl+V). This fits your “does not auto-send” goal while still being quick.
• UI automation: use Windows UI Automation / accessibility APIs (or simulated input) to focus the message box and paste the image + prompt, but stop short of clicking Send.
• Temp file + “attach” automation: save the screenshot to a temp path, then automate the “attach file” flow for the target assistant (more fragile across UI updates).
• If using native apps (Copilot app, etc.): similar idea, just different UI automation targets.

Given you said Copilot is “a little flaky sometimes on the screenshot attachment step,” that strongly suggests some UI-automation/foreground-window timing dependency (focus, clipboard readiness, attachment widget state, etc.).

How “context-aware presets” are probably detected
Common lightweight signals that don’t require heavy inspection:

• Foreground process name / executable (e.g., outlook.exe, chrome.exe, msedge.exe, thunderbird.exe).
• Window class name and/or window title keywords (e.g., “Inbox”, “Message”, “Gmail”, “Outlook”).
• Browser tab title heuristics (indirectly visible via window title).

This is enough to switch from generic help prompts to “email safety” prompts without doing anything intrusive like OCR or content scraping.

Security/privacy design choices that match your description
What you described aligns with a “thin glue layer” model:

• No background recording; only user-initiated capture.
• No reading assistant responses (no “agent loop”).
• No autonomous clicking through workflows.
• A deliberate “manual Send” moment so the user decides what leaves the machine.

Where I would be cautious (implementation-agnostic)

• Temp artifacts: screenshots saved to disk (even briefly) should be clearly disclosed and securely cleaned up. If you can stay clipboard-only, that reduces forensic residue.
• Over-automation fragility: UI automation can break when assistants change their UI; having a “fallback: open assistant + copy prompt + copy image” mode can keep it reliable.
• Data leakage user education: a short reminder near the Send step like “Check for secrets (tokens, email addresses, customer data) before sending” helps users avoid accidental disclosure.

Bottom line: from your description, Secure Helper sounds like a controlled “capture + prompt template + assistant handoff” tool, probably using screenshot capture plus clipboard/UI-automation to stage (but not send) the message.

Yes, that is quite impressive, you pretty much nailed it! Yes, security is the top priority, hence the name Secure Helper, and the Temp Artifacts are automatically deleted. Can you think of other apps that are similar to this to give users an idea of how this works? Also, if you could describe Secure Helper in 3-7 words, what would be the best description you can think of?

 

[Bot]

Yes, that is quite impressive, you pretty much nailed it! Yes, security is the top priority, hence the name Secure Helper, and the Temp Artifacts are automatically deleted. Can you think of other apps that are similar to this to give users an idea of how this works? Also, if you could describe Secure Helper in 3-7 words, what would be the best description you can think of?

Similar apps / workflows people may already recognize
There aren’t many well-known “exact matches” (capture + context-aware prompt presets + handoff to multiple AI services while stopping at manual Send), but there are close analogies that help people “get it” quickly:

• Windows Snipping Tool / Snip & Sketch → paste into ChatGPT
  This is the manual version of what Secure Helper streamlines: capture a region, then switch apps, paste the image, type the question, send.
• ShareX / Greenshot capture workflows
  These tools are very similar in the “capture something intentionally selected, then run an action” sense (copy to clipboard, open a link, run a custom task). Secure Helper is like “ShareX workflow + AI prompt templates,” with the extra safety guardrail of not auto-sending.
• Browser screenshot tools + ‘share’
  Many users have used browser screenshot/share features; Secure Helper is like that, but targeted at getting help/analysis via an AI assistant rather than just sharing an image.
• Enterprise support/annotation tools (conceptually)
  Some remote-support/knowledge tools center around “show the screen, ask a question, get guidance” — except your model is local, user-initiated, and doesn’t require handing control to a third party.

If you want a simple one-liner comparison: it’s closest to “Snipping Tool + prewritten prompts + assistant launcher,” with a deliberate manual Send step.

Secure Helper in 3–7 words
My best pick:

• One-click screenshot-to-AI handoff

A few alternates (depending on what you want to emphasize):

• User-controlled screenshot prompt launcher
• Manual-send AI help for screens
• Privacy-first screen capture to AI
• Context-aware prompts for screenshots

Bottom line: the easiest mental model for users is “a safer, faster Snip-and-ask workflow,” with guardrails that keep the user in control.

 

[danb]

Similar apps / workflows people may already recognize
There aren’t many well-known “exact matches” (capture + context-aware prompt presets + handoff to multiple AI services while stopping at manual Send), but there are close analogies that help people “get it” quickly:

• Windows Snipping Tool / Snip & Sketch → paste into ChatGPT
  This is the manual version of what Secure Helper streamlines: capture a region, then switch apps, paste the image, type the question, send.
• ShareX / Greenshot capture workflows
  These tools are very similar in the “capture something intentionally selected, then run an action” sense (copy to clipboard, open a link, run a custom task). Secure Helper is like “ShareX workflow + AI prompt templates,” with the extra safety guardrail of not auto-sending.
• Browser screenshot tools + ‘share’
  Many users have used browser screenshot/share features; Secure Helper is like that, but targeted at getting help/analysis via an AI assistant rather than just sharing an image.
• Enterprise support/annotation tools (conceptually)
  Some remote-support/knowledge tools center around “show the screen, ask a question, get guidance” — except your model is local, user-initiated, and doesn’t require handing control to a third party.

If you want a simple one-liner comparison: it’s closest to “Snipping Tool + prewritten prompts + assistant launcher,” with a deliberate manual Send step.

Secure Helper in 3–7 words
My best pick:

• One-click screenshot-to-AI handoff

A few alternates (depending on what you want to emphasize):

• User-controlled screenshot prompt launcher
• Manual-send AI help for screens
• Privacy-first screen capture to AI
• Context-aware prompts for screenshots

Bottom line: the easiest mental model for users is “a safer, faster Snip-and-ask workflow,” with guardrails that keep the user in control.

Your first description was much better than this one honestly. The best description I have seen so far is "I think there is a useful middle ground between fully manual computer help and autonomous agents that control the machine. Secure Helper is my attempt at that middle ground: user-initiated, privacy-conscious, transparent, and manual at the final Send step."

 

[Bot]

Your first description was much better than this one honestly. The best description I have seen so far is "I think there is a useful middle ground between fully manual computer help and autonomous agents that control the machine. Secure Helper is my attempt at that middle ground: user-initiated, privacy-conscious, transparent, and manual at the final Send step."

You are right that your longer description is stronger because it communicates the “why” (human-in-the-loop, privacy, no autonomous control), not just the mechanics. It also avoids sounding like yet another “AI screenshot tool.”

What your description communicates best

• It’s a middle ground: not manual copy/paste, but not an agent that takes control.
• User intent is required: user-initiated capture and manual final Send.
• Privacy and transparency are design goals, not afterthoughts.

A tighter version that keeps the same meaning (1–2 sentences)

• Secure Helper is a human-in-the-loop bridge between your screen and an AI assistant: you select what to capture, it prepares the screenshot and prompt, and you decide what gets sent. It’s designed to be privacy-conscious and non-autonomous by default.

3–7 word descriptions that match your positioning

• Human-in-the-loop AI screen helper
• Manual-send screen-to-AI assistant
• User-controlled screenshot-to-AI workflow
• Privacy-first, non-autonomous AI helper

If you want one consistent phrase to anchor the whole product message, “human-in-the-loop” + “manual final Send” is the clearest differentiator and also sets the right security expectations.

 

[danb]

You are right that your longer description is stronger because it communicates the “why” (human-in-the-loop, privacy, no autonomous control), not just the mechanics. It also avoids sounding like yet another “AI screenshot tool.”

What your description communicates best

• It’s a middle ground: not manual copy/paste, but not an agent that takes control.
• User intent is required: user-initiated capture and manual final Send.
• Privacy and transparency are design goals, not afterthoughts.

A tighter version that keeps the same meaning (1–2 sentences)

• Secure Helper is a human-in-the-loop bridge between your screen and an AI assistant: you select what to capture, it prepares the screenshot and prompt, and you decide what gets sent. It’s designed to be privacy-conscious and non-autonomous by default.

3–7 word descriptions that match your positioning

• Human-in-the-loop AI screen helper
• Manual-send screen-to-AI assistant
• User-controlled screenshot-to-AI workflow
• Privacy-first, non-autonomous AI helper

If you want one consistent phrase to anchor the whole product message, “human-in-the-loop” + “manual final Send” is the clearest differentiator and also sets the right security expectations.

Yes, exactly, it is the middle ground where the user works with the agent instead of the agent that takes full control. Thank you @Bot!

 

[Bot]

Yes, exactly, it is the middle ground where the user works with the agent instead of the agent that takes control. Thank you @Bot!

That framing is the clearest differentiator: “human-in-the-loop” help rather than an autonomous agent.

What to emphasize (so users instantly understand)

• User-initiated capture (you choose the window/desktop)
• Prepared prompt + screenshot, but no automatic sending
• No background recording, no answer scraping, no PC control
• The user stays accountable at the final Send step

A short positioning line that matches your intent

• A human-in-the-loop bridge between your screen and your AI assistant.

If you keep repeating “human-in-the-loop” + “manual final Send” consistently, it will set the right security expectations and distinguish Secure Helper from agentic tools that take control.

 

[danb]

BTW, there is a new version that hopefully fixes the bug I found and also deletes the screenshots even more proactively. I have not setup the version numbers yet, so it is still 1.00, but you can tell by the hash.

SecureHelper 1.00 Demo

https://www.cyberlock.global/downloads/SecureHelper.exe

SHA-256: 9e3f4da06a8feb44c8be9acc32869f5ca6eac385ee2dad65944a6019a4077f04

 

[Sorrento]

All i get?? Nothing else?