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 #5290] [BUG]: CPU usage scales with default thread message count; persists in 1.11.2 #5001

New Issue
Closed
opened 2026-06-05 14:51:24 -04:00 by yindo · 3 comments
yindo commented 2026-06-05 14:51:24 -04:00
Owner
Copy Link
Copy Source

Originally created by @anthion on GitHub (Mar 27, 2026).
Original GitHub issue: https://github.com/Mintplex-Labs/anything-llm/issues/5290

How are you running AnythingLLM?

AnythingLLM desktop app

What happened?

Title: CPU usage scales with default thread message count; persists in 1.11.2
Environment:
∙ AnythingLLM version: 1.11.2 (desktop)
∙ OS: Windows 11 Pro 23H2
∙ Hardware: Intel i9-13900K, RTX 4090, 64GB RAM
∙ LLM backend: LM Studio (local)
Description:
CPU usage in the AnythingLLM desktop app scales proportionally with the number of messages in the active workspace thread. The default thread cannot be deleted, so it accumulates messages over time and the problem compounds.
Additional notes: This is reproducible on 1.11.2. The behavior is isolated to the active thread — an empty thread causes no elevated CPU usage. Minimizing the window drop CPU usage to near zero.

Are there known steps to reproduce?

Reproduction steps:
1. Open the desktop app with a thread containing many messages
2. Observe CPU usage in Task Manager
3. Delete messages from the thread one at a time
4. Observe CPU usage drop with each deletion
5. Minimize the app window
6. Observe CPU usage drop to near zero
Expected behavior: CPU usage should be low and stable when the app is idle and no inference is running.
Actual behavior: CPU usage is high and proportional to thread length while the window is visible, regardless of whether any inference or embedding is occurring. This starves the LLM backend of CPU resources mid-inference, causing degraded context and unreliable document retrieval.

Originally created by @anthion on GitHub (Mar 27, 2026). Original GitHub issue: https://github.com/Mintplex-Labs/anything-llm/issues/5290 ### How are you running AnythingLLM? AnythingLLM desktop app ### What happened? Title: CPU usage scales with default thread message count; persists in 1.11.2 Environment: ∙ AnythingLLM version: 1.11.2 (desktop) ∙ OS: Windows 11 Pro 23H2 ∙ Hardware: Intel i9-13900K, RTX 4090, 64GB RAM ∙ LLM backend: LM Studio (local) Description: CPU usage in the AnythingLLM desktop app scales proportionally with the number of messages in the active workspace thread. The default thread cannot be deleted, so it accumulates messages over time and the problem compounds. Additional notes: This is reproducible on 1.11.2. The behavior is isolated to the active thread — an empty thread causes no elevated CPU usage. Minimizing the window drop CPU usage to near zero. ### Are there known steps to reproduce? Reproduction steps: 1. Open the desktop app with a thread containing many messages 2. Observe CPU usage in Task Manager 3. Delete messages from the thread one at a time 4. Observe CPU usage drop with each deletion 5. Minimize the app window 6. Observe CPU usage drop to near zero Expected behavior: CPU usage should be low and stable when the app is idle and no inference is running. Actual behavior: CPU usage is high and proportional to thread length while the window is visible, regardless of whether any inference or embedding is occurring. This starves the LLM backend of CPU resources mid-inference, causing degraded context and unreliable document retrieval.
yindo added the possible bugDesktopinvestigating labels 2026-06-05 14:51:24 -04:00
yindo closed this issue 2026-06-05 14:51:24 -04:00
yindo commented 2026-06-05 14:51:25 -04:00
Author
Owner
Copy Link
Copy Source

@timothycarambat commented on GitHub (Mar 28, 2026):

Any estimate for how many is "many" in this example just so I can have an idea?

<!-- gh-comment-id:4149212388 --> @timothycarambat commented on GitHub (Mar 28, 2026): Any estimate for how many is "many" in this example just so I can have an idea?
yindo commented 2026-06-05 14:51:25 -04:00
Author
Owner
Copy Link
Copy Source

@anthion commented on GitHub (Mar 28, 2026):

Any estimate for how many is "many" in this example just so I can have an idea?
I will replicate the problem tomorrow (3/29) and update with hopefully detailed information.

<!-- gh-comment-id:4149223380 --> @anthion commented on GitHub (Mar 28, 2026): > Any estimate for how many is "many" in this example just so I can have an idea? I will replicate the problem tomorrow (3/29) and update with hopefully detailed information.
yindo commented 2026-06-05 14:51:25 -04:00
Author
Owner
Copy Link
Copy Source

@anthion commented on GitHub (Mar 29, 2026):

5% CPU

“I need the hole size for an M2 thread insert”
“I need the hole size for an M3 thread insert”
“I need the hole size for an M4 thread insert” (Asked twice)
“I need the hole size for an M5 thread insert”
“is there information for M6 thread insert hole size?”
“what about M7 thread inserts?”
“give me the hole size for M8 inserts”
“I need some info about adhesives in my documents”
“can you tell me about encoders?”

10% CPU

what documents are available to you at the moment?
tell me about the Teensy
how many analog inputs does it have?
how many SPI devices can the teensy accommodate?
do you have any cost information about it?
how do you know that the market pricing varies?
what can you tell me about the power input and output on the teensy 4?
is there a limit to the number of I2C devices it can accommodate?
do you know what a QPD is?
tell me about the quadrant photodiode device in my documents
do you have cost info for it?
is there any info about transimpedance amplifiers?
do you have the main comment block for the main.cpp file available?
do you have dimensional info about the QPD?
do you have dimensional info about any things in the documents?
based on the circular area of the photosensitive device can you approximate the diameter?
what is the gap size between quadrants?
is there a signal rise time specified for the QPD?

35 to 50% CPU

what documents are available to you at the moment?
tell me about the Teensy
how many analog inputs does it have?
how many SPI devices can the teensy accommodate?
do you have any cost information about it?
how do you know that the market pricing varies?
what can you tell me about the power input and output on the teensy 4?
is there a limit to the number of I2C devices it can accommodate?
do you know what a QPD is?
tell me about the quadrant photodiode device in my documents
do you have cost info for it?
is there any info about transimpedance amplifiers?
do you have the main comment block for the main.cpp file available?
do you have dimensional info about the QPD?
do you have dimensional info about any things in the documents?
based on the circular area of the photosensitive device can you approximate the diameter?
what is the gap size between quadrants?
is there a signal rise time specified for the QPD?
tell me about stepper motor drivers
can you tell me how to configure the microsteps?
can you tell me how many pins the drivers have? what connector I should use on my PCB?
is there any mass or weight info for any devices in the documents?
can you explain why the quadrant photodiode needs a bias voltage?
what is the adhesive used for?
why do you include the context numbers in your responses?
if I have a long text file such as my main.cpp file it would require it to be split up into a chain of smaller chunks is that correct?
so is the answer yes or no? since the main.cpp was only partially indexed I’m trying to figure out a way to prepare the documents better

100% CPU

I got to question #16 and it completely s#it the bed and stopped responding for some unknown reason. I think it got stuck in a thinking loop. I deleted the question attempts beyond #16 and it came back online responding to "hello" but failed again.
Apparently it fails without warning with too many threads?
So at the moment I don't have a 100% example thread but I have a couple times in the past. I can post screen grabs if you like.
Currently ALLM is useless for my small project.

<!-- gh-comment-id:4150619557 --> @anthion commented on GitHub (Mar 29, 2026): 5% CPU “I need the hole size for an M2 thread insert” “I need the hole size for an M3 thread insert” “I need the hole size for an M4 thread insert” (Asked twice) “I need the hole size for an M5 thread insert” “is there information for M6 thread insert hole size?” “what about M7 thread inserts?” “give me the hole size for M8 inserts” “I need some info about adhesives in my documents” “can you tell me about encoders?” 10% CPU what documents are available to you at the moment? tell me about the Teensy how many analog inputs does it have? how many SPI devices can the teensy accommodate? do you have any cost information about it? how do you know that the market pricing varies? what can you tell me about the power input and output on the teensy 4? is there a limit to the number of I2C devices it can accommodate? do you know what a QPD is? tell me about the quadrant photodiode device in my documents do you have cost info for it? is there any info about transimpedance amplifiers? do you have the main comment block for the main.cpp file available? do you have dimensional info about the QPD? do you have dimensional info about any things in the documents? based on the circular area of the photosensitive device can you approximate the diameter? what is the gap size between quadrants? is there a signal rise time specified for the QPD? 35 to 50% CPU what documents are available to you at the moment? tell me about the Teensy how many analog inputs does it have? how many SPI devices can the teensy accommodate? do you have any cost information about it? how do you know that the market pricing varies? what can you tell me about the power input and output on the teensy 4? is there a limit to the number of I2C devices it can accommodate? do you know what a QPD is? tell me about the quadrant photodiode device in my documents do you have cost info for it? is there any info about transimpedance amplifiers? do you have the main comment block for the main.cpp file available? do you have dimensional info about the QPD? do you have dimensional info about any things in the documents? based on the circular area of the photosensitive device can you approximate the diameter? what is the gap size between quadrants? is there a signal rise time specified for the QPD? tell me about stepper motor drivers can you tell me how to configure the microsteps? can you tell me how many pins the drivers have? what connector I should use on my PCB? is there any mass or weight info for any devices in the documents? can you explain why the quadrant photodiode needs a bias voltage? what is the adhesive used for? why do you include the context numbers in your responses? if I have a long text file such as my main.cpp file it would require it to be split up into a chain of smaller chunks is that correct? so is the answer yes or no? since the main.cpp was only partially indexed I’m trying to figure out a way to prepare the documents better 100% CPU I got to question #16 and it completely s#it the bed and stopped responding for some unknown reason. I think it got stuck in a thinking loop. I deleted the question attempts beyond #16 and it came back online responding to "hello" but failed again. Apparently it fails without warning with too many threads? So at the moment I don't have a 100% example thread but I have a couple times in the past. I can post screen grabs if you like. Currently ALLM is useless for my small project.
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 Desktop investigating 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#5001
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?