Agent Prompt Optimizer by Kayba
Auto-optimize agent prompts from real failures.
21 followers
Auto-optimize agent prompts from real failures.
21 followers
Agent Prompt Optimizer turns your agents’ mistakes into better prompts autonomously. Instead of manual prompt engineering, it watches where agents fail, extracts reusable insights, and updates prompts so they stop repeating errors. Watch your agents get better with every run. Drop it into existing agents or frameworks (e.g. LangChain) with just a few lines of code. Fully open source - try it on your agent right now and tell us what it learns.
We built Agent Prompt Optimizer after talking to a lot of agent builders and seeing the same pattern everywhere: agents would make the same mistake over and over, and humans had to babysit them by tweaking prompts manually each time.
So we asked: what if the agent could learn from these failures itself?
Agent Prompt Optimizer watches agent runs, captures what worked vs what failed, and turns those insights into reusable prompt improvements. No extra labels, no fine-tuning. Just agents that stop repeating dumb mistakes and actually get better with use.
We’d love feedback on:
– Where in your agent stack repeated mistakes hurt you most
– How you’d like to visualize and approve prompt changes
– Any agents / frameworks you’d like to see integrated next
If you try it, please tell us what breaks and what (if anything) feels magical.