[GH-ISSUE #13371] External Tools do not consistently work with custom models (based on downloaded models + custom prompts) #16895

New Issue

2026-04-19T22:43:20-05:00

GiteaMirror commented

2026-04-19 22:43:20 -05:00

Originally created by @St1ngAving on GitHub (Apr 30, 2025).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/13371

Check Existing Issues

I have searched the existing issues and discussions.
I am using the latest version of Open WebUI.

Installation Method

Docker

Open WebUI Version

v0.6.5

Ollama Version (if applicable)

v0.6.6

Operating System

Windows 11

Browser (if applicable)

Opera GX 118.0.5461.62

Confirmation

I have read and followed all instructions in README.md.
I am using the latest version of both Open WebUI and Ollama.
I have included the browser console logs.
I have included the Docker container logs.
I have listed steps to reproduce the bug in detail.

Expected Behavior

External tools (such as web browsing, code interpreter, etc.) should function reliably and consistently throughout the entire conversation when using custom models — just as they do with official models. Specifically:

External Tools should be triggerable multiple times in the same conversation.
External Tool responses should appear correctly without requiring manual resets or workarounds.
Custom prompts or modified models should not interfere with external tool functionality.

The behavior should be consistent, regardless of whether the model is official or custom.

Actual Behavior

When using external tools with custom models (created by overwriting downloaded models with custom prompts), external tool functionality becomes unreliable. Sometimes the external tool works only once at the beginning of the conversation. Other times, it doesn’t work at all, even on the first attempt. It's as if the model is simulating the request without actually invoking the external tool — no output is produced, and no external tool activity is triggered.

This issue does not occur with official models, which consistently trigger external tools correctly across multiple uses in the same conversation. No relevant errors appear in the browser console or Docker logs.

Steps to Reproduce

Download a base model.
Create a custom model by overwriting it with a custom prompt.
Start a new conversation using this custom model.
Ask the model to use an external tool — it may work the first time, or it may not work at all.
Try to use an external tool again later in the same conversation — it no longer works if it worked once before.
No errors appear in the browser console or Docker logs.

Logs & Screenshots

https://github.com/user-attachments/assets/5d276d92-31b2-4bd6-a727-702a825ea32d

Additional Information

The issue does not occur with regular/official models.
There are no visible errors or logs indicating what's wrong.

Originally created by @St1ngAving on GitHub (Apr 30, 2025). Original GitHub issue: https://github.com/open-webui/open-webui/issues/13371 ### Check Existing Issues - [x] I have searched the existing issues and discussions. - [x] I am using the latest version of Open WebUI. ### Installation Method Docker ### Open WebUI Version v0.6.5 ### Ollama Version (if applicable) v0.6.6 ### Operating System Windows 11 ### Browser (if applicable) Opera GX 118.0.5461.62 ### Confirmation - [x] I have read and followed all instructions in `README.md`. - [x] I am using the latest version of **both** Open WebUI and Ollama. - [x] I have included the browser console logs. - [x] I have included the Docker container logs. - [x] I have listed steps to reproduce the bug in detail. ### Expected Behavior External tools (such as web browsing, code interpreter, etc.) should function reliably and consistently throughout the entire conversation when using custom models — just as they do with official models. Specifically: - External Tools should be triggerable multiple times in the same conversation. - External Tool responses should appear correctly without requiring manual resets or workarounds. - Custom prompts or modified models should not interfere with external tool functionality. The behavior should be consistent, regardless of whether the model is official or custom. ### Actual Behavior When using external tools with custom models (created by overwriting downloaded models with custom prompts), external tool functionality becomes unreliable. Sometimes the external tool works only once at the beginning of the conversation. Other times, it doesn’t work at all, even on the first attempt. It's as if the model is simulating the request without actually invoking the external tool — no output is produced, and no external tool activity is triggered. This issue does not occur with official models, which consistently trigger external tools correctly across multiple uses in the same conversation. No relevant errors appear in the browser console or Docker logs. ### Steps to Reproduce 1. Download a base model. 2. Create a custom model by overwriting it with a custom prompt. 3. Start a new conversation using this custom model. 4. Ask the model to use an external tool — it may work the first time, or it may not work at all. 5. Try to use an external tool again later in the same conversation — it no longer works if it worked once before. 6. No errors appear in the browser console or Docker logs. ### Logs & Screenshots https://github.com/user-attachments/assets/5d276d92-31b2-4bd6-a727-702a825ea32d ### Additional Information - The issue does **not** occur with regular/official models. - There are no visible errors or logs indicating what's wrong.

GiteaMirror added the bug label 2026-04-19 22:43:20 -05:00

GiteaMirror closed this issue

2026-04-19 22:43:21 -05:00

GiteaMirror commented

2026-04-19 22:43:21 -05:00

@tjbck commented on GitHub (Apr 30, 2025):

Model dependent behaviour, try using gpt-4o.

@tjbck commented on GitHub (Apr 30, 2025): Model dependent behaviour, try using gpt-4o.

GiteaMirror commented

2026-04-19 22:43:22 -05:00

@St1ngAving commented on GitHub (Apr 30, 2025):

Model dependent behaviour, try using gpt-4o.

Thanks — confirmed, it works with gpt-4o!
So to clarify, is the issue essentially due to the 11B model lacking sufficient capacity to handle both the prompt and external tool management reliably? Just trying to understand the limitation better.

@St1ngAving commented on GitHub (Apr 30, 2025): > Model dependent behaviour, try using gpt-4o. Thanks — confirmed, it works with gpt-4o! So to clarify, is the issue essentially due to the 11B model lacking sufficient capacity to handle both the prompt and external tool management reliably? Just trying to understand the limitation better.

Sign in to join this conversation.

Branches Tags

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: github-starred/open-webui#16895