Custom AI processing prompts
in progress
C
Curtis Hibbs
Right now there are two built-in AI processing prompts: concise summaries and detailed summaries. We should be able to create additional, custom prompts that are clickable like the built-in ones.
Log In
P
Paul Richards
marked this post as
in progress
Hi all,
We're working on custom AI actions! We've started building this feature and wanted to get your input on how it should work.
Custom actions will be available in the slash menu in the notebook, so you can quickly access your saved prompts alongside other commands.
We're currently deciding how these actions should behave in chat, and we'd love your feedback:
Option A: In-chat: Trigger a custom AI action mid-conversation from options above the chat input, similar to how concise and detailed summaries show today. The response appears inline with your other chat messages, so you can ask follow-up questions and build on the result.
Option B: Standalone: Custom AI actions can only be triggered on new chats so you cannot trigger them inline with an existing conversation. The main pro of this option is to reduce clutter in the chat after you've already started a conversation.
Which would you prefer, and why? Or would you want both? Let us know in the comments.
J
Jame Healy
My most-used use case is Recall for YouTube summaries. I love the convenience of the extension interface discretely to the side, and it's very quick and easy to select one of the two standard Actions.
However, my ideal is to create my own extended quick Actions, or otherwise modify the standard Action prompts to include enhancements something like "If the YouTube video title or thumbnail is click-baity, emphasize a concise answer to the click-bait question in the very first paragraph of your summary" and/or "if the video title/thumbnail references a list of things, emphasize a summarized and enumerated list of those things in the very first paragraph of the summary".
Among other things, the bottom line is just to get to the bottom line simply by pressing the quick Action link (i.e. not necessarily just in the chat or notebook) ... then I can decide to what degree I want to invest in consuming/reading/watching the rest of the content.
I
Ignacio Poveda
I clearly prefer option A (in chat), because it allows an iterative flow of learning, especially important for generating questions, detecting errors and refining knowledge within the same conversation.
However, the main problem with option A is the clutter in the chat. The ideal solution would be to maintain activation within the chat, but with structured or collapsible results that do not saturate the conversation.
In addition, it would be very useful to be able to directly convert the results into notes or cards, since the actual use is iterative and based on continuous improvement.
s
schroell
Option B (Standalone) would be my go-to, since most use cases I'd have for Custom AI Actions or Slash Commands would start from a fresh chat.
That said, Option A makes sense too — and honestly, where I need it most is inside a card. Like, if I've added a YouTube video to Recall, the transcript is basically already loaded, and when I open that note and go to the note's chat, I want to be able to use Custom Actions with Slash Commands right there too.
P
Pete Bell
Option A: In-chat is non-negotiable. Option B: Standalone is a nice-to-have as a clean-start affordance, not a restriction. The "clutter" concern is better solved with collapsible action outputs or a visual separator, not by killing context continuity.
The bigger design question is what context an action operates on. Real workflows need more than "current message" or "full thread." A long research chat might span multiple topics, and I'd want to run "extract key arguments" against only the relevant cluster of messages, not everything. Even a lightweight "apply to: full chat / recent messages / selected" toggle at trigger time would make a real difference. Potential to turn a saved-prompt shortcut into something people build workflows around, and where Recall could pull ahead of competitors offering canned prompt templates. Just my 2c.
B
Banarescu Crina
Both x
U
Utkarsh Deep
Ideally, it should be a mix of option A and B that you suggested. We should be able to set any of the custom prompts as a default prompt, and also use the other non-default prompts in mid-chat.
If that's not possible and I'd have to choose just between A & B, then I'd prefer option A. We should be able to add more than 1 custom instruction overall. Ideally, the slash command mid-chat would work better.
P
Paul Richards
marked this post as
in progress
Hi all,
We're working on custom AI actions! We've started building this feature and wanted to get your input on how it should work.
Custom actions will be available in the slash menu in the notebook, so you can quickly access your saved prompts alongside other commands.
We're currently deciding how these actions should behave in chat, and we'd love your feedback:
Option A: In-chat: Trigger a custom AI action mid-conversation from options above the chat input, similar to how concise and detailed summaries show today. The response appears inline with your other chat messages, so you can ask follow-up questions and build on the result.
Option B: Standalone: Custom AI actions can only be triggered on new chats so you cannot trigger them inline with an existing conversation. The main pro of this option is to reduce clutter in the chat after you've already started a conversation.
Which would you prefer, and why? Or would you want both? Let us know in the comments.
t
tc
Paul Richards - I use an LLM front end called TypingMind and I've always liked the ability to call on different pieces of functionality from within the chat via a pop up menu (see screen shot).
J
Julia Beach
Snipd does this really well!
Sankari Nair
Merged in a post:
Universal Prompt
J
JC Grubbs
It would be great if there was a place to drop a universal prompt that would apply to all chat sessions. I use a similar feature on Claude and ChatGPT and it is super helpful for refining responses based on my preferences.
Sankari Nair
marked this post as
planned
Load More
→