[GH-ISSUE #23948] feat: Workspace model access without base model access #35652

Closed
opened 2026-04-25 09:49:47 -05:00 by GiteaMirror · 14 comments
Owner

Originally created by @itsHenry35 on GitHub (Apr 21, 2026).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/23948

Check Existing Issues

  • I have searched for all existing open AND closed issues and discussions for similar requests. I have found none that is comparable to my request.

Verify Feature Scope

  • I have read through and understood the scope definition for feature requests in the Issues section. I believe my feature request meets the definition and belongs in the Issues section instead of the Discussions.

Problem Description

As described in https://github.com/open-webui/open-webui/issues/23946

Desired Solution you'd like

Add a env / config to allow global access of workspace model without base model access

Alternatives Considered

No response

Additional Context

No response

Originally created by @itsHenry35 on GitHub (Apr 21, 2026). Original GitHub issue: https://github.com/open-webui/open-webui/issues/23948 ### Check Existing Issues - [x] I have searched for all existing **open AND closed** issues and discussions for similar requests. I have found none that is comparable to my request. ### Verify Feature Scope - [x] I have read through and understood the scope definition for feature requests in the Issues section. I believe my feature request meets the definition and belongs in the Issues section instead of the Discussions. ### Problem Description As described in https://github.com/open-webui/open-webui/issues/23946 ### Desired Solution you'd like Add a env / config to allow global access of workspace model without base model access ### Alternatives Considered _No response_ ### Additional Context _No response_
Author
Owner

@dacai-academic commented on GitHub (Apr 22, 2026):

I have granted full permissions to both the user group and the individual user; however, the user is still unable to access the pre-configured models.

<!-- gh-comment-id:4293111468 --> @dacai-academic commented on GitHub (Apr 22, 2026): I have granted full permissions to both the user group and the individual user; however, the user is still unable to access the pre-configured models.
Author
Owner

@Classic298 commented on GitHub (Apr 22, 2026):

@dacai-academic user has to have access to the base model

<!-- gh-comment-id:4294091787 --> @Classic298 commented on GitHub (Apr 22, 2026): @dacai-academic user has to have access to the base model
Author
Owner

@itsHenry35 commented on GitHub (Apr 22, 2026):

I have granted full permissions to both the user group and the individual user; however, the user is still unable to access the pre-configured models.

Yeah, and that's why this issue exists. I have the save issue as yours and I think urgent fix is needed as this has broken our usual workflow.

<!-- gh-comment-id:4294704005 --> @itsHenry35 commented on GitHub (Apr 22, 2026): > I have granted full permissions to both the user group and the individual user; however, the user is still unable to access the pre-configured models. Yeah, and that's why this issue exists. I have the save issue as yours and I think urgent fix is needed as this has broken our usual workflow.
Author
Owner

@Classic298 commented on GitHub (Apr 22, 2026):

This is not to be "fixed"

This was a security vulnerability that was patched.

@itsHenry35 you should give your users access to the base model. Otherwise they are not eligible to access the base model.

If you want them to have access to the base model without SEEING the base model you can make use of the "Hide model" button in the admin panel > models where. This way they can query the workspace model because they have access to the base model but won't see the base model anywhere in the UI.

<!-- gh-comment-id:4295881160 --> @Classic298 commented on GitHub (Apr 22, 2026): This is not to be "fixed" This was a security vulnerability that was patched. @itsHenry35 you should give your users access to the base model. Otherwise they are not eligible to access the base model. If you want them to have access to the base model without SEEING the base model you can make use of the "Hide model" button in the admin panel > models where. This way they can query the workspace model because they have access to the base model but won't see the base model anywhere in the UI.
Author
Owner

@itsHenry35 commented on GitHub (Apr 22, 2026):

This is not to be "fixed"

This was a security vulnerability that was patched.

@itsHenry35 you should give your users access to the base model. Otherwise they are not eligible to access the base model.

If you want them to have access to the base model without SEEING the base model you can make use of the "Hide model" button in the admin panel > models where. This way they can query the workspace model because they have access to the base model but won't see the base model anywhere in the UI.

This workaround just has one big limitation that I can't do: i want some users to see it, while only hiding if for a group of users using that workspace model

<!-- gh-comment-id:4295890682 --> @itsHenry35 commented on GitHub (Apr 22, 2026): > This is not to be "fixed" > > This was a security vulnerability that was patched. > > [@itsHenry35](https://github.com/itsHenry35) you should give your users access to the base model. Otherwise they are not eligible to access the base model. > > If you want them to have access to the base model without SEEING the base model you can make use of the "Hide model" button in the admin panel > models where. This way they can query the workspace model because they have access to the base model but won't see the base model anywhere in the UI. This workaround just has one big limitation that I can't do: i want some users to see it, while only hiding if for a group of users using that workspace model
Author
Owner

@Classic298 commented on GitHub (Apr 22, 2026):

@itsHenry35 two workspace models, two base models?

Base model 1 hidden for everyone
Base model 2 shown to everyone
Workspace model 1 based on base model 1
Workspace model 2 based on base model 2

<!-- gh-comment-id:4295899132 --> @Classic298 commented on GitHub (Apr 22, 2026): @itsHenry35 two workspace models, two base models? Base model 1 hidden for everyone Base model 2 shown to everyone Workspace model 1 based on base model 1 Workspace model 2 based on base model 2
Author
Owner

@itsHenry35 commented on GitHub (Apr 22, 2026):

@itsHenry35 two workspace models, two base models?

Base model 1 hidden for everyone Base model 2 shown to everyone Workspace model 1 based on base model 1 Workspace model 2 based on base model 2

Nah it's just I want to let some users use that base model directly while hiding it for the other group of users using the workspace model

<!-- gh-comment-id:4295911511 --> @itsHenry35 commented on GitHub (Apr 22, 2026): > [@itsHenry35](https://github.com/itsHenry35) two workspace models, two base models? > > Base model 1 hidden for everyone Base model 2 shown to everyone Workspace model 1 based on base model 1 Workspace model 2 based on base model 2 Nah it's just I want to let some users use that base model directly while hiding it for the other group of users using the workspace model
Author
Owner

@Classic298 commented on GitHub (Apr 22, 2026):

@itsHenry35 ok got it

Two choices

  1. what i described before with duplication
  2. use a custom pipe. All the pipe does is route to that base model - set the pipe as the workspace base model.. and then set the access control of the ACTUAL base model so that that one group has access to it and the other doesn't.

Would that help?

<!-- gh-comment-id:4295928023 --> @Classic298 commented on GitHub (Apr 22, 2026): @itsHenry35 ok got it Two choices 1) what i described before with duplication 2) use a custom pipe. All the pipe does is route to that base model - set the pipe as the workspace base model.. and then set the access control of the ACTUAL base model so that that one group has access to it and the other doesn't. Would that help?
Author
Owner

@itsHenry35 commented on GitHub (Apr 22, 2026):

@itsHenry35 ok got it

Two choices

  1. what i described before with duplication
  2. use a custom pipe. All the pipe does is route to that base model - set the pipe as the workspace base model.. and then set the access control of the ACTUAL base model so that that one group has access to it and the other doesn't.

Thanks! That duplication method does work but that leads to actual confusion. The following is what we do:
Group A have the permission to create custom models so that they fine tune for different circumstances for group C
Group A and B are normal users that use the base models
Group C are a restricted people to use the workspace only model.
So if I'm using that duplication method or custom pipeline, it means that people in group A will see two duplicate models which is not too good and it leads to confusion.

<!-- gh-comment-id:4295981477 --> @itsHenry35 commented on GitHub (Apr 22, 2026): > [@itsHenry35](https://github.com/itsHenry35) ok got it > > Two choices > > 1. what i described before with duplication > 2. use a custom pipe. All the pipe does is route to that base model - set the pipe as the workspace base model.. and then set the access control of the ACTUAL base model so that that one group has access to it and the other doesn't. Thanks! That duplication method does work but that leads to actual confusion. The following is what we do: Group A have the permission to create custom models so that they fine tune for different circumstances for group C Group A and B are normal users that use the base models Group C are a restricted people to use the workspace only model. So if I'm using that duplication method or custom pipeline, it means that people in group A will see two duplicate models which is not too good and it leads to confusion.
Author
Owner

@Classic298 commented on GitHub (Apr 22, 2026):

Hang on hold up hang on hang on wait wait wIt, let me make sure I understand the historical state.

Before the patch — how were users creating workspace models on base models they didn't have access to? The UI never showed inaccessible models in the base model dropdown. Were your authors building these via API calls, bypassing the UI's access controls?

Because if so, the "workflow that broke" is admins scripting around a UI-enforced access restriction. That's the exploit the patch closed, not a feature that got removed.

On the pipe approach — Group A are your model authors. They're technical enough to understand "this is the routing pipe for the restricted workspace model, don't use it directly." You can also hide the pipe from everyone (including Group A) at the admin level and they can still use it to build models — they only need to reference it at workspace-model creation time, and once the workspace model is built, the pipe can stay hidden. No confusion, no duplicate entries visible to end users.

The duplication objection only holds up if you assume the admin-built routing layer needs to be visible to the people who built it. It doesn't.

Try the pipe approach with the pipe hidden. I'm fairly confident it solves your case without any architectural change needed.

So the setup is:

  • Group A → sees the pipe (clearly named, so they know it's for building the restricted workspace model)
  • Group B → sees the base model directly (their normal workflow)
  • Group C → sees only the workspace model

Access control handles the rest — Group C has no direct base model access, the workspace model routes through the pipe, the pipe hits the base model internally. No duplication visible to any user, no security compromise, and Group A doesn't get confused because they configured this and the pipe is named for its purpose.

I don't really see an issue here

<!-- gh-comment-id:4296067033 --> @Classic298 commented on GitHub (Apr 22, 2026): Hang on hold up hang on hang on wait wait wIt, let me make sure I understand the historical state. Before the patch — how were users creating workspace models on base models they didn't have access to? The UI never showed inaccessible models in the base model dropdown. Were your authors building these via API calls, bypassing the UI's access controls? Because if so, the "workflow that broke" is admins scripting around a UI-enforced access restriction. That's the exploit the patch closed, not a feature that got removed. On the pipe approach — Group A are your model authors. They're technical enough to understand "this is the routing pipe for the restricted workspace model, don't use it directly." You can also hide the pipe from everyone (including Group A) at the admin level and they can still use it to build models — they only need to reference it at workspace-model creation time, and once the workspace model is built, the pipe can stay hidden. No confusion, no duplicate entries visible to end users. The duplication objection only holds up if you assume the admin-built routing layer needs to be visible to the people who built it. It doesn't. Try the pipe approach with the pipe hidden. I'm fairly confident it solves your case without any architectural change needed. So the setup is: - Group A → sees the pipe (clearly named, so they know it's for building the restricted workspace model) - Group B → sees the base model directly (their normal workflow) - Group C → sees only the workspace model Access control handles the rest — Group C has no direct base model access, the workspace model routes through the pipe, the pipe hits the base model internally. No duplication visible to any user, no security compromise, and Group A doesn't get confused because they configured this and the pipe is named for its purpose. I don't really see an issue here
Author
Owner

@itsHenry35 commented on GitHub (Apr 22, 2026):

Hang on hold up hang on hang on wait wait wIt, let me make sure I understand the historical state.

Before the patch — how were users creating workspace models on base models they didn't have access to? The UI never showed inaccessible models in the base model dropdown. Were your authors building these via API calls, bypassing the UI's access controls?

Because if so, the "workflow that broke" is admins scripting around a UI-enforced access restriction. That's the exploit the patch closed, not a feature that got removed.

On the pipe approach — Group A are your model authors. They're technical enough to understand "this is the routing pipe for the restricted workspace model, don't use it directly." You can also hide the pipe from everyone (including Group A) at the admin level and they can still use it to build models — they only need to reference it at workspace-model creation time, and once the workspace model is built, the pipe can stay hidden. No confusion, no duplicate entries visible to end users.

The duplication objection only holds up if you assume the admin-built routing layer needs to be visible to the people who built it. It doesn't.

Try the pipe approach with the pipe hidden. I'm fairly confident it solves your case without any architectural change needed.

So the setup is:

  • Group A → sees the pipe (clearly named, so they know it's for building the restricted workspace model)
  • Group B → sees the base model directly (their normal workflow)
  • Group C → sees only the workspace model

Access control handles the rest — Group C has no direct base model access, the workspace model routes through the pipe, the pipe hits the base model internally. No duplication visible to any user, no security compromise, and Group A doesn't get confused because they configured this and the pipe is named for its purpose.

I don't really see an issue here

Thanks! That may just seem fine and I'll try. Just fyi, in the past, the authors had that base model that they could see them in the webui.

<!-- gh-comment-id:4296107074 --> @itsHenry35 commented on GitHub (Apr 22, 2026): > Hang on hold up hang on hang on wait wait wIt, let me make sure I understand the historical state. > > Before the patch — how were users creating workspace models on base models they didn't have access to? The UI never showed inaccessible models in the base model dropdown. Were your authors building these via API calls, bypassing the UI's access controls? > > Because if so, the "workflow that broke" is admins scripting around a UI-enforced access restriction. That's the exploit the patch closed, not a feature that got removed. > > On the pipe approach — Group A are your model authors. They're technical enough to understand "this is the routing pipe for the restricted workspace model, don't use it directly." You can also hide the pipe from everyone (including Group A) at the admin level and they can still use it to build models — they only need to reference it at workspace-model creation time, and once the workspace model is built, the pipe can stay hidden. No confusion, no duplicate entries visible to end users. > > The duplication objection only holds up if you assume the admin-built routing layer needs to be visible to the people who built it. It doesn't. > > Try the pipe approach with the pipe hidden. I'm fairly confident it solves your case without any architectural change needed. > > So the setup is: > > * Group A → sees the pipe (clearly named, so they know it's for building the restricted workspace model) > * Group B → sees the base model directly (their normal workflow) > * Group C → sees only the workspace model > > Access control handles the rest — Group C has no direct base model access, the workspace model routes through the pipe, the pipe hits the base model internally. No duplication visible to any user, no security compromise, and Group A doesn't get confused because they configured this and the pipe is named for its purpose. > > I don't really see an issue here Thanks! That may just seem fine and I'll try. Just fyi, in the past, the authors had that base model that they could see them in the webui.
Author
Owner

@Classic298 commented on GitHub (Apr 22, 2026):

Ah gotcha — i understand.

Let me know how the pipe setup goes. If it works we can close this one out, and if you hit something that doesn't fit, let me know.

<!-- gh-comment-id:4296126111 --> @Classic298 commented on GitHub (Apr 22, 2026): Ah gotcha — i understand. Let me know how the pipe setup goes. If it works we can close this one out, and if you hit something that doesn't fit, let me know.
Author
Owner

@itsHenry35 commented on GitHub (Apr 23, 2026):

Ah gotcha — i understand.

Let me know how the pipe setup goes. If it works we can close this one out, and if you hit something that doesn't fit, let me know.

@Classic298 Just tried and don't know why the model created by custom pipeline just hasn't shown up.
UPDATE:
I think there might have been something wrong with that caching model option because as I turned that off, the model just showed up... Maybe forgot to clear the cache in the code?

<!-- gh-comment-id:4303503604 --> @itsHenry35 commented on GitHub (Apr 23, 2026): > Ah gotcha — i understand. > > Let me know how the pipe setup goes. If it works we can close this one out, and if you hit something that doesn't fit, let me know. @Classic298 Just tried and don't know why the model created by custom pipeline just hasn't shown up. UPDATE: I think there might have been something wrong with that caching model option because as I turned that off, the model just showed up... Maybe forgot to clear the cache in the code?
Author
Owner

@itsHenry35 commented on GitHub (Apr 23, 2026):

I think it just worked. Code is copied from the docs. However there seems to be some issues.

from pydantic import BaseModel, Field
from fastapi import Request

from open_webui.models.users import Users
from open_webui.utils.chat import generate_chat_completion

class Pipe:
def init(self):
pass

async def pipe(
    self,
    body: dict,
    __user__: dict,
    __request__: Request,
) -> str:
    user = await Users.get_user_by_id(
        "5d494790-fcc8-4de6-a9f4-0eacbcc7217d"
    )  # admin user
    body["model"] = "deepseek-chat"
    return await generate_chat_completion(__request__, body, user)

The model isn't a thinking model but I don't know why it's showing thinking process in the workspace model which is super strange.
Base model:
Image
Pipeline model:
Image
Workspace model:
Image

<!-- gh-comment-id:4303545350 --> @itsHenry35 commented on GitHub (Apr 23, 2026): I think it just worked. Code is copied from the docs. However there seems to be some issues. from pydantic import BaseModel, Field from fastapi import Request from open_webui.models.users import Users from open_webui.utils.chat import generate_chat_completion class Pipe: def __init__(self): pass async def pipe( self, body: dict, __user__: dict, __request__: Request, ) -> str: user = await Users.get_user_by_id( "5d494790-fcc8-4de6-a9f4-0eacbcc7217d" ) # admin user body["model"] = "deepseek-chat" return await generate_chat_completion(__request__, body, user) The model isn't a thinking model but I don't know why it's showing thinking process in the workspace model which is super strange. Base model: <img width="1596" height="569" alt="Image" src="https://github.com/user-attachments/assets/76dd7530-2cae-498c-964a-da5883e2aee3" /> Pipeline model: <img width="1581" height="678" alt="Image" src="https://github.com/user-attachments/assets/0f1be2f3-f526-47b7-9c7b-5523b13530ab" /> Workspace model: <img width="1467" height="585" alt="Image" src="https://github.com/user-attachments/assets/42aa2255-17cc-4550-a0c6-3d3b749a9bae" />
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/open-webui#35652