github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 00:22:43 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #2865] Privacy settings on models #27510

New Issue
Closed
opened 2026-04-22 04:54:09 -05:00 by GiteaMirror · 7 comments
GiteaMirror commented 2026-04-22 04:54:09 -05:00
Owner
Copy Link

Originally created by @trymeouteh on GitHub (Mar 1, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/2865

The ability to customise the privacy settings for each model. These privacy setting can restrict what the model can do with your device...

  • Access the internet. You can disable a model from accessing the internet, making sure it only runs offline
  • Reading files (Cannot read files on your system)
  • Writing files (Cannot create files on your system)
Originally created by @trymeouteh on GitHub (Mar 1, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/2865 The ability to customise the privacy settings for each model. These privacy setting can restrict what the model can do with your device... - Access the internet. You can disable a model from accessing the internet, making sure it only runs offline - Reading files (Cannot read files on your system) - Writing files (Cannot create files on your system)
GiteaMirror commented 2026-04-22 04:54:10 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (Mar 4, 2024):

Ollama consists of two parts which run locally on your own machine; an API server which serves the model, and a CLI which is used to talk to the local API server. A model running on the server absolutely cannot access the internet, read a file, or write a file. The CLI can read an image file (for instance w/ multi-modal models) which it will then pass the image data to the local server, but again, none of that data will ever leave your machine. You also would have to have specified the file name that you wanted to read and pass to the server (the server running the model will never access the file system).

That said, the server can access the internet, but only for two specific cases:

  • pulling a model; and
  • pushing a model

But it will never do that during inference. We would definitely have privacy settings if we were to ever add a feature like that. I'm going to go ahead and close the issue for now.

<!-- gh-comment-id:1975999567 --> @pdevine commented on GitHub (Mar 4, 2024): Ollama consists of two parts which run locally on your own machine; an API server which serves the model, and a CLI which is used to talk to the local API server. A model running on the server absolutely cannot access the internet, read a file, or write a file. The CLI _can_ read an image file (for instance w/ multi-modal models) which it will then pass the image data to the local server, but again, none of that data will ever leave your machine. You also would have to have specified the file name that you wanted to read and pass to the server (the server running the model will never access the file system). That said, the server _can_ access the internet, but only for two specific cases: * pulling a model; and * pushing a model But it will _never_ do that during inference. We would definitely have privacy settings if we were to ever add a feature like that. I'm going to go ahead and close the issue for now.
GiteaMirror commented 2026-04-22 04:54:11 -05:00
Author
Owner
Copy Link

@trymeouteh commented on GitHub (Mar 8, 2024):

I hope someday there are search engine models like Phind and YOU.com that you can run on your device that will make API calls to a search engine and get results back for you and for there to be models that can generate images for you.

If these models are available in Ollama, please do add privacy settings to perhaps limit what these models can do.

<!-- gh-comment-id:1985746376 --> @trymeouteh commented on GitHub (Mar 8, 2024): I hope someday there are search engine models like Phind and YOU.com that you can run on your device that will make API calls to a search engine and get results back for you and for there to be models that can generate images for you. If these models are available in Ollama, please do add privacy settings to perhaps limit what these models can do.
GiteaMirror commented 2026-04-22 04:54:12 -05:00
Author
Owner
Copy Link

@ti2ger92 commented on GitHub (Jan 28, 2025):

what about this? how do I turn it off?

https://ollama.com/blog/tool-support

<!-- gh-comment-id:2619742472 --> @ti2ger92 commented on GitHub (Jan 28, 2025): what about this? how do I turn it off? https://ollama.com/blog/tool-support
GiteaMirror commented 2026-04-22 04:54:13 -05:00
Author
Owner
Copy Link

@Gladwell5 commented on GitHub (Feb 2, 2025):

With Activity Monitor on a Mac, I see ollama has a small amount of data sent and received at inference time when the model is already downloaded. Any idea what that is?

<!-- gh-comment-id:2629380047 --> @Gladwell5 commented on GitHub (Feb 2, 2025): With Activity Monitor on a Mac, I see ollama has a small amount of data sent and received at inference time when the model is already downloaded. Any idea what that is?
GiteaMirror commented 2026-04-22 04:54:16 -05:00
Author
Owner
Copy Link

@kravivar commented on GitHub (Apr 8, 2026):

I just wanted to report this incident here. Attached screenshot.

Short Summary: I asked the model to count number of words and it initially gave me wrong answer and I challenged it saying you are wrong can re-evaluate and give me correct number. It used an online service.

@pdevine: FYI.

Image Image Image Image
<!-- gh-comment-id:4205642431 --> @kravivar commented on GitHub (Apr 8, 2026): I just wanted to report this incident here. Attached screenshot. Short Summary: I asked the model to count number of words and it initially gave me wrong answer and I challenged it saying you are wrong can re-evaluate and give me correct number. It used an online service. @pdevine: FYI. <img width="746" height="258" alt="Image" src="https://github.com/user-attachments/assets/68643551-f49a-495f-99c8-f1e24b594935" /> <img width="768" height="669" alt="Image" src="https://github.com/user-attachments/assets/bb364910-92b9-473d-98b5-417e57dad057" /> <img width="692" height="220" alt="Image" src="https://github.com/user-attachments/assets/5b40bc41-334a-4689-911f-356af25d2e76" /> <img width="283" height="164" alt="Image" src="https://github.com/user-attachments/assets/aaa40f8d-5b26-4f14-8d1b-e8401e05e3a1" />
GiteaMirror commented 2026-04-22 04:54:18 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (Apr 8, 2026):

@kravivar The model basically just lied to you (i.e. it was a hallucination). You can check the log to see if there was a web search, but if you have everything turned off it's not possible for it to do that.

<!-- gh-comment-id:4208366138 --> @pdevine commented on GitHub (Apr 8, 2026): @kravivar The model basically just lied to you (i.e. it was a hallucination). You can check the log to see if there was a web search, but if you have everything turned off it's not possible for it to do that.
GiteaMirror commented 2026-04-22 04:54:21 -05:00
Author
Owner
Copy Link

@kravivar commented on GitHub (Apr 9, 2026):

You are correct @pdevine it did lie to me. I was too naive to post this prematurely, apologies. I was able to validate on server logs and system log messages.

~ cat ~/.ollama/logs/server.log | grep -i 'wordcounter.net'
~
~ sudo log show --predicate 'eventMessage CONTAINS[c] "wordcounter.net"' --last 24h --debug --info
Filtering the log data using "composedMessage CONTAINS[c] "wordcounter.net""
Timestamp                       Thread     Type        Activity             PID    TTL  
--------------------------------------------------------------------------------------------------------------------
Log      - Default:          0, Info:                0, Debug:             0, Error:          0, Fault:          0
Activity - Create:           0, Transition:          0, Actions:           0
Image
<!-- gh-comment-id:4210964306 --> @kravivar commented on GitHub (Apr 9, 2026): You are correct @pdevine it did lie to me. I was too naive to post this prematurely, apologies. I was able to validate on server logs and system log messages. ``` ~ cat ~/.ollama/logs/server.log | grep -i 'wordcounter.net' ~ ~ sudo log show --predicate 'eventMessage CONTAINS[c] "wordcounter.net"' --last 24h --debug --info Filtering the log data using "composedMessage CONTAINS[c] "wordcounter.net"" Timestamp Thread Type Activity PID TTL -------------------------------------------------------------------------------------------------------------------- Log - Default: 0, Info: 0, Debug: 0, Error: 0, Fault: 0 Activity - Create: 0, Transition: 0, Actions: 0 ``` <img width="744" height="551" alt="Image" src="https://github.com/user-attachments/assets/11831270-d309-47ea-8873-dae66477e6cf" />
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#27510
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?