Mintplex-Labs/anything-llm
1
0
Fork 0
mirror of https://github.com/Mintplex-Labs/anything-llm.git synced 2026-07-19 14:13:35 -04:00
Code Issues 328 Packages Projects Releases 10 Wiki Activity

[GH-ISSUE #2463] [BUG]: Agent do not use its web browsing feature (only one over ten) #1602

New Issue
Closed
opened 2026-02-22 18:25:36 -05:00 by yindo · 6 comments
yindo commented 2026-02-22 18:25:36 -05:00
Owner
Copy Link
Copy Source

Originally created by @flefevre on GitHub (Oct 13, 2024).
Original GitHub issue: https://github.com/Mintplex-Labs/anything-llm/issues/2463

How are you running AnythingLLM?

Docker (local)

What happened?

The agent make web browsing research only one time over ten, even if I ask directly "make a web browsing / research" in the prompt.

Are there known steps to reproduce?

when

  • i activate the web browsing feature, with google parameter for the agent
  • i call the agent with @agent in a discussion
  • i ask for instance "who is the prime minister in France to day, make a reasearch on internet"

The agent answered me
"I am sorry, I cannot perform an internet search because I am a language model and not a search engine. However, using the information I have up to now, the current Prime Minister of France is Jean Castex."

Nevertheless, i try several in time in the same discussion channel or open a new one, some time, the agent make the call to its web browsing feature and answered me fine.

Originally created by @flefevre on GitHub (Oct 13, 2024). Original GitHub issue: https://github.com/Mintplex-Labs/anything-llm/issues/2463 ### How are you running AnythingLLM? Docker (local) ### What happened? The agent make web browsing research only one time over ten, even if I ask directly "make a web browsing / research" in the prompt. ### Are there known steps to reproduce? when - i activate the web browsing feature, with google parameter for the agent - i call the agent with @agent in a discussion - i ask for instance "who is the prime minister in France to day, make a reasearch on internet" The agent answered me "I am sorry, I cannot perform an internet search because I am a language model and not a search engine. However, using the information I have up to now, the current Prime Minister of France is Jean Castex." Nevertheless, i try several in time in the same discussion channel or open a new one, some time, the agent make the call to its web browsing feature and answered me fine.
yindo added the possible bug label 2026-02-22 18:25:36 -05:00
yindo closed this issue 2026-02-22 18:25:37 -05:00
yindo commented 2026-02-22 18:25:37 -05:00
Author
Owner
Copy Link
Copy Source

@timothycarambat commented on GitHub (Oct 14, 2024):

This is related to your model - https://docs.anythingllm.com/agent-not-using-tools

Basically, you need a more reliable model. This is accomplished via moving to increasing one of two parameters:

  • the model's parameter (3B - > 8B) if possible
  • The quantization used. The default is typically 4-bit and its usually bad for tool calling. Q8 is best usually. This is of course a factor of the parameter size. A 70B Q4 usually is better than a Q8 3B model.
@timothycarambat commented on GitHub (Oct 14, 2024): This is related to your model - https://docs.anythingllm.com/agent-not-using-tools Basically, you need a more reliable model. This is accomplished via moving to increasing one of two parameters: - the model's parameter (3B - > 8B) if possible - The quantization used. The default is typically 4-bit and its usually bad for tool calling. Q8 is best usually. This is of course a factor of the parameter size. A 70B Q4 usually is better than a Q8 3B model.
yindo commented 2026-02-22 18:25:37 -05:00
Author
Owner
Copy Link
Copy Source

@Securiteru commented on GitHub (Jan 31, 2025):

@timothycarambat, is it normal for tool calls to not work with 4o?

I'm currently running the AppImage version of AnythingLLM (the latest) and tools doesn't seem to trigger at all.

Image

Image

Image

Sorry for hijacking this thread, I just down know if it's something that I'm doing or not.

@Securiteru commented on GitHub (Jan 31, 2025): @timothycarambat, is it normal for tool calls to not work with 4o? I'm currently running the AppImage version of AnythingLLM (the latest) and tools doesn't seem to trigger at all. ![Image](https://github.com/user-attachments/assets/d94d260b-4d8b-4d06-baa0-49955eacc9ac) ![Image](https://github.com/user-attachments/assets/255d4c37-3834-42d4-b2e6-74ce15549d07) ![Image](https://github.com/user-attachments/assets/88887102-0a8a-4f2a-8ab0-0af5b20f2a80) Sorry for hijacking this thread, I just down know if it's something that I'm doing or not.
yindo commented 2026-02-22 18:25:37 -05:00
Author
Owner
Copy Link
Copy Source

@timothycarambat commented on GitHub (Jan 31, 2025):

@Securiteru

I'm currently running the AppImage version of AnythingLLM (the latest) and tools doesn't seem to trigger at all.

The agent flow actually never and ran at all. It exited instantly before it could even run. Can you check the logs or run the AppImage in debug? Basically, something is crashing the agent before it can even execute so nothing happens.

Using gpt-4o for agent. Ignore the workspace name, this is not deepseek. This is what agent calls look like when they work.

Image
@timothycarambat commented on GitHub (Jan 31, 2025): @Securiteru > I'm currently running the AppImage version of AnythingLLM (the latest) and tools doesn't seem to trigger at all. The agent flow actually never and ran at all. It exited instantly before it could even run. Can you check the logs or run the AppImage in [debug](https://docs.anythingllm.com/installation-desktop/debug)? Basically, something is crashing the agent before it can even execute so nothing happens. Using gpt-4o for agent. Ignore the workspace name, this is not deepseek. This is what agent calls look like when they work. <img width="1904" alt="Image" src="https://github.com/user-attachments/assets/12534827-2c42-4f58-8b4c-6504731b6ea2" />
yindo commented 2026-02-22 18:25:37 -05:00
Author
Owner
Copy Link
Copy Source

@boomboom77365 commented on GitHub (Mar 16, 2025):

My agent refuses to use web-browsing skill to look for anything at all. I'm running DeepSeek-R1-Distill-Llama-8B-Q6_K. My answers are always very strange and ignorant.

@boomboom77365 commented on GitHub (Mar 16, 2025): My agent refuses to use web-browsing skill to look for anything at all. I'm running DeepSeek-R1-Distill-Llama-8B-Q6_K. My answers are always very strange and ignorant.
yindo commented 2026-02-22 18:25:37 -05:00
Author
Owner
Copy Link
Copy Source

@o0LINNY0o commented on GitHub (Jul 13, 2025):

This is driving me insane ! , every YT video makes it look so easy and im getting zero help . ive created tickets in Discord channel and in Reddit on /LocalLLaMA .

Ive ran as Admin , uninstalled and re-installed , ensured firewalls are not on ,

its not pulling any of the tools either , ive used Qwen314 4_k_M a& Agentic 7B Q8 ... nothing changes

@o0LINNY0o commented on GitHub (Jul 13, 2025): This is driving me insane ! , every YT video makes it look so easy and im getting zero help . ive created tickets in Discord channel and in Reddit on /LocalLLaMA . Ive ran as Admin , uninstalled and re-installed , ensured firewalls are not on , its not pulling any of the tools either , ive used Qwen314 4_k_M a& Agentic 7B Q8 ... nothing changes
yindo commented 2026-02-22 18:25:37 -05:00
Author
Owner
Copy Link
Copy Source

@one-dumb-dude commented on GitHub (Nov 2, 2025):

@o0LINNY0o Did you ever figure this out?

@one-dumb-dude commented on GitHub (Nov 2, 2025): @o0LINNY0o Did you ever figure this out?
yindo changed title from [BUG]: Agent do not use its web browsing feature (only one over ten) to [GH-ISSUE #2463] [BUG]: Agent do not use its web browsing feature (only one over ten) 2026-06-05 14:41:40 -04:00
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
Desktop
Docker
Integration Request
Integration Request
OS: Linux
OS: Mobile
OS: Windows
UI/UX
blocked
bug
bug
core-team-only
documentation
duplicate
embed-widget
enhancement
feature request
github_actions
good first issue
investigating
needs info / can't replicate
possible bug
pull-request
Mirrored from GitHub Pull Request
 question
stage: specifications
wontfix
No Label possible bug
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
yindo
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: Mintplex-Labs/anything-llm#1602
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?