Hiding the background work my Pickaxe prompt is doing from user

General
1

My Pickaxe prompt goes out to get info from specific places on the web I tell it to and when it’s “working” on doing that it shows the user where it’s going and what it’s getting. Is there a way I can hide the work info its doing from the user so they’re not distracted by it or so they won’t reverse engineer my prompt’s logic?

2 Likes
2

Hey @danimal

This is a great question that touches on some advanced agent design principles. The solution depends on how your agent is set up to access the web.

Let’s break this down into your two main goals:

1. Hiding the “Work Info” from the User

It sounds like you may have a Connected Action running, like the OpenAI Web Search, which by default can show the user its process.

While you can’t completely hide the fact that a tool is being used, you can often control the messages it displays to the user. Try editing the trigger prompt for that specific Action in the Actions tab.

For example, add an instruction like: "When you use this tool to search the web, simply tell the user 'Searching for the information...' and do not show the specific steps or websites you are visiting."

2. Preventing Reverse Engineering of Your Prompt

This is a crucial security consideration. Here are two best practices:

  • Add a Security Protocol: You should add clear rules to the bottom of your main prompt that instruct the agent to never reveal its instructions. Tell it to politely decline if a user asks how it works or what its prompt is.
  • Isolate Proprietary Logic: The most effective way to protect your prompt’s logic is to not have it in the prompt at all. Instead of listing the specific websites in the prompt, move that proprietary information into the Knowledge Base or into a secure backend workflow on a platform like Make.com, which the Pickaxe Action can then call.

This approach separates your instructions from your proprietary data, which is the best way to secure your agent’s unique logic.

3 Likes
3

This is exactly the guidance I needed, thank you. Regarding the Knowledge Base approach, if I list the key websites I want my Pickaxe to check as part of my prompting does that mean my Pickaxe will goto to those sites in real-time or does the KB just contained the RAG db stored info?

1 Like
4

@danimal You’re welcome! That’s a great follow-up question, and it gets to a key difference in how Pickaxe handles information.

To answer you directly: The Knowledge Base does not search those websites in real-time. It only contains the information that was on the page when you first added it.

Here’s how it works:

When you add a URL to the Knowledge Base, a service scrapes the text from that page one time and stores it in the RAG database. Your Pickaxe is then working from that static, stored information, not the live website.

For real-time information, you need to use a Connected Action like the “Google Search” or another web-Browse tool found in the Actions tab.

So, here’s how you can combine both for the best result, keeping your prompt logic private:

  1. Put your list of proprietary websites into a simple text file and upload that file to your Knowledge Base.
  2. Enable a Web Search Action on your Pickaxe.
  3. In your prompt, instruct the agent to first check the Knowledge Base to find the approved list of sites, and then use the Web Search Action to visit those specific sites for live information.

This keeps your private list of sources hidden in the KB, while still giving your agent the power to get real-time info.

Hope that clears it up!