I`m creating some agents in the agent builder but facing some high token usage. There is any place to share and download shared skills by the community to help agents and get better results?
High token usage from creating new skills or executing agents or both?
Have you seen this?
These are skills more from the command line than in kibana but perhaps this will help
High token usage for executing agents.
Using Elastic Inference LLMs does not give us a way to limit token usage, so our biggest challenge here is how to enable the AI Agent to more users.
We already had cases where 1 user asked the agent for an observability analysis without specifying context and it consumed 82 millions input tokens.
I already tried the Elastic Agent skills provided in the Elastic Github and now i`m trying to create some custom skills to give better instructions to the agent to use less tokens. I also tried Elastic Caveman skill but it showed to only reduce the output token usage, where my problem is with the input tokens.