[BUG]: Cannot find Bangla document uploaded to the Workspace

[mayeenulislam]

How are you running AnythingLLM?

AnythingLLM desktop app

What happened?

I uploaded a markdown file to the Workspace:

Using Ollama and gemma2:latest (max tokens: 4096) I asked:

@agent can you scrape the animals.md and summarize in bangla?

✅ Based on my CPU and RAM it took time, but responded with some text after summarization

I then thought to speed up the processing using Groq API. So I changed the settings of the Workspace to:

LLM Provider: Groq Chat model: gemma2-9b-it Chat mode: Chat

Whereas the vector database shows a vector count 57

Document similarity threshold: High Agent LLM Provider: Groq Agent Model: gemma2-9b-it

But with every query after setting Groq, it ❌ failed:

@agent can you scrape the animals.md and summarize in bangla? @agent Summarize the “animals.md” file in Bangla

With a significant improvement in speed, it says that it doesn't have any access to the file.

I tried changing the Agent Model to Ollama:gemma2:2b, but it stays the same.

Are there known steps to reproduce?

1. Make a Workspace
2. Upload a markdown file to the workspace
3. Use Ollama and gemma2:2b model
4. Ask to summarize the file
5. It will do it successfully
6. Add Groq API for both chat provider and agent provider
7. Ask to summarize the file again
8. It will fail finding the file

[timothycarambat]

https://docs.anythingllm.com/agent-not-using-tools

Not a bug, just the LLM being bad at creating valid tool calls. In fact, in that screenshot I dont even see the "thoughts" of the agent calling tools - did they never show up?

[mayeenulislam]

The thoughts were there. But couldn't find them later.