search

Mintplex-Labs / anything-llm

The all-in-one Desktop & Docker AI application with built-in RAG, AI agents, and more.
https://anythingllm.com
MIT License
27.47k stars 2.77k forks source link

[FEAT]: Ability to turn off default `@agent` skills to prevent tool call conflicts #2615

Open kirkjaa opened 1 week ago

kirkjaa commented 1 week ago

What would you like to see?

Hello Timothy and the team, First off AnythingLLM is awesome we've been using this in the production environment for various cases from reporting, to teaching coding and even CFA. However as we integrated custom agents into the workspace like GPT reseacher which yield result but the default agent would over take and oversimplify the report instead of the results from the GPT Reseacher. If we can turn the default agent in the workspace I believe it would solve the conflict with our custom agent

timothycarambat commented 1 week ago

The workspace agent is only invoked if @agent is called. When you say you added custom agents, how do you mean - through custom code?

kirkjaa commented 1 week ago

Hello!

We created a custom agent via upload to custom skills as your documentation at: https://docs.anythingllm.com/agent/custom/introduction

Everything works fine with GPT Researcher, however, the issue that we're experiencing is that the default @agent would summarize the content that GPT Researcher generates because of the default setting of the agent within the workspace ( I think RAG and long-term memory) . I've uploaded the demo video via this unlisted YouTube https://youtu.be/aszG--1Q7Qs the first 30 seconds should explain the issue we have here on the default skill agent oversimplified the content report of our GPT Researcher custom skill agents.

timothycarambat commented 1 week ago

Then you dont need the default agent disabled, you need the default skills disabled. Are you running a modified version of anythingllm locally? If so, simply comment out these lines and those tools will be disabled

https://github.com/Mintplex-Labs/anything-llm/blob/cf3b085a3af710f631bbcc08e21b54cbb953c90f/server/utils/agents/defaults.js#L21-L23

kirkjaa commented 1 week ago

Thank you for pointing us in the right direction! Yes, this definitely would work from the look of the code. We're running a docker version of anything-llm, much appreciated.

kirkjaa commented 1 week ago

Then you dont need the default agent disabled, you need the default skills disabled. Are you running a modified version of anythingllm locally? If so, simply comment out these lines and those tools will be disabled

https://github.com/Mintplex-Labs/anything-llm/blob/cf3b085a3af710f631bbcc08e21b54cbb953c90f/server/utils/agents/defaults.js#L21-L23

Hello, that didn't work for some reason after we comment out the default skill, however we were able to get the report out put with this prompt in the wokspace chat @agent @gpt_researcher do research and "return the exact output from the agent in the chat in plaintext/markdown"