Welcome to Portkey Forum

Updated 10 months ago

Suggestions for improving portkey product

At a glance
Hi,
I had few suggestions for Portkey product which will make my team really productive

P0
  • Folders for Prompts (Currently, it become very difficult to organize the prompts)
  • Support for Hindi Unicode in displaying Prompt (Have already reported to Rohit)
P1
  • Personal Prompts and Team Prompts (So that navigating prompts does not become cluttered)
  • Keeping Production & Sandbox separate. (I am always worried if some changes in dev will break production. We use same prompt Id for Sandbox & Production. I would want to keep "text" of prompt same, but variables like "Model provider" or "Model Virtual" Key different.
  • Same for keeping Sandbox Keys and Production keys separate
  • Testing same prompt on multiple models and comparing the results.
P2
  • Evaluation/Test Framework. Maybe ability to rate prompt responses through another LLM Prompt.
S
V
17 comments
Another small suggestion
  • I really like this small note which gets generated on every update, but the message is too generic whenever I update parameters ("Prompt Parameters updates"). Could we make it specific like what parameter was updated?
Attachment
image.png
cc @ayush-portkey - he's the prompt engineer for this πŸ˜„
Thanks for the detailed suggestions @Siddharth Bulia! Really appreciate this. Sharing with the team, and will let you know here if/when we pick them up!

Folder for prompts is a hotly requested feature - we'll pick it up soon πŸ˜„
Thank you Vrushank!!
Another small suggestion around analytics -
Is the any to find cuts of cost around PromptId or PromptTag?
Attachment
image.png
@Siddharth Bulia what are you referring to as PromptTag? The specific prompt template version?
In the Model filter, you can find Saved Prompts at the bottom - all of your prompt templates that have requests withint the chosen timeframe will show up there. You can select one or more and see their analytics
Attachment
CleanShot_2024-05-15_at_15.34.112x.png
Ahh! Got it.
For some reason, Search Interface for Filter did not seem intuitive for me. Will use it. Thanks!
But the feature I am actually looking is more like "Group By" Function.
I want to compare which prompts are costing most money.
Not needed right away, but I could suggest for the roadmap.
Yep, we want to separate out prompt templates from the "Models" filter too. Soon πŸ˜„ cc @Sabbyasachi
Got it! Yep agreed this would be very useful. Noting it πŸ“
@Siddharth Bulia in the meanwhile, you can check out the Metadata tab on the Analytics page, where we show "Group By" - like summaries for any key:value pairs you send in the metadata. So you could achieve this for your prompt templates right now by passing a different custom metadata KV and then seeing the cost trends there

Docs for adding Metadata
Understood Vrushank. Will check out.

I will mostly just report every bug/feature which pop-up in my mind. πŸ˜…
You can always then prioritize in the order you find most suitable considering need of all other clients πŸ™‚
Haha and thanks for that!
Another feature request is ability to search within the prompt logs!!
The only currently I am able to pinpoint a prompt call is only through timestamp.
Another much requested feature πŸ˜„ This is on the roadmap
Add a reply
Sign up and join the conversation on Discord