Pretty Prompt
Grammarly for prompting
Start new thread

Launching today: Instructions vs. Memory 🧠 aka Context v2

by
•

Not all context is equal.

So we stopped treating it that way.

Most AI tools store everything the same:

  • Writing rules

  • Personal background

  • Preferences

  • Temporary notes

But a rule like:

“Use bullet points.”

…shouldn’t behave the same as:

“I run a fintech startup and we've just raise $4m”

So we rebuilt our context system, to help Pretty Prompt give better results.

(Best part, the user owns that context, and can use it across any LLM)

🔒 Instructions

Explicit rules that always apply. Examples:

  • “No em-dashes”

  • “Keep answers under 512 characters”

  • “Use bullet points”

🧠 Memory

Background context that only appears when relevant. Examples:

  • “My audience is non-technical”

  • “I'm launching on Product Hunt and like to post from a builder's mindset”

  • “I’m writing a client proposal that should include...”

The best part:

You don’t organize any of this manually.

You just write naturally.

Pretty Prompt auto-classifies everything on the backend.

The goal is simple:

Make AI outputs be personal — without adding more work.

Still bootstrapped. Still shipping 🚀

PH community, give us feedback; we need it.

(PS - I vibed this thumbnail, don't judge me 🙈)

100 views

Add a comment

Replies

Best

I’ve honeslty struggled with AI tools mixing permanent preferences with random temporary context, so this makes a lot of sense to me.

I’d love to learn more about your experience with other tools in this space!
glad it resonated! Please give it a try I’d love some feedback
I’d love to learn more about your workflow! If you’re up for a chat my email is

My LLM isn't following the explicit rules no matter how many times I repeat them. Thoughts?

 Drop me an email to and I'll personally look into it and help :)

Easier there so I don't miss anything

Really like this distinction. Treating “instruction” and “memory” as the same thing is one of those small product choices that quietly creates a lot of mediocre AI output.

The auto-classification piece feels especially important. Most people will not maintain a context system manually, but they *will* write naturally if the product can infer what should persist, what should be temporary, and what should act like a rule.

One thing I’d be curious about: do you let users inspect or correct the classification after the fact? I’ve found that trust in AI memory often depends less on perfect recall and more on whether people feel they can see and shape what the system thinks it knows.

they can always edit the context snippet :) It’s a work in progress still. Open to your feedback after you’ve tried it!