How can I make my pickaxe bot respond faster?

General
,
1

I have build a prompt where my Pickaxe works with a user to determine an amount of money to ask for a certain job. All of the logic works and it is quite reliable. However, it takes the bot ages to respond to an answer. How can I speed up the process? It works a lot faster when I run the same prompt in ChatGPT or Gemini.

2

Appreciate you calling this out, Arne. I filed it internally as PRD-513 so we can dig into the latency spike youre seeing versus ChatGPT/Gemini. Ill update the thread once we have either a fix or guidance on speeding that flow back up.

1 Like
3

Thank you. It is my intention to be selling this solution to my clients, but if the speed does not increase, I need to find another solution to do this.

5

Hi @arnehulstein,

Good to hear the logic is working well. When a Pickaxe feels slower than running the same prompt directly in ChatGPT or Gemini, it’s usually related to things like the prompt structure, how many Actions are connected, how the tool calls those actions, or how much context is being passed in the conversation. Since Pickaxes sit on top of an LLM and can also trigger external actions or pull knowledge, those extra steps can sometimes add a bit of delay. In most cases though, the difference shouldn’t be very noticeable.

In many cases the speed can be improved with a few small adjustments. Things like tightening the prompt, reviewing how actions are triggered, or simplifying the response flow often make a noticeable difference.

That said, the best advice really comes from seeing the actual setup. If you’re open to it, please send us a quick email at info@pickaxe.co with:
• the Pickaxe link
• the email associated with your account
• a short note about which tool this is

Happy to take a look and suggest specific tweaks that should help speed things up.

1 Like
6

im also finding we can get away with a faster/cheaper model on tasks that arent too demanding. I a imagine you have a pricing scheme defined in the knowledge base, so the LLM is really only interpreting the needs of the user and then matching it to the correct products, etc.
That might be a simpler model imho

7

Thanks Abhi. I will send you an email.

@dougfresh This actually is almost like an interview with many different options that need to be considered. So there is quite a bit of intelligent choosing that the LLM needs to do.

1 Like
8

Thanks for sending that through, Arne — I grabbed the deployment link + account details from your email and attached them to our latency ticket so engineering can dig in with the exact setup you are selling. I will circle back here once we have guidance or a fix.