[GH-ISSUE #22697] issue: New files added AFTER enabling BYPASS_EMBEDDING_AND_RETRIEVAL are NOT indexed for vector search, causing inconsistent search results #106787

New Issue

GiteaMirror · 2026-05-18T05:16:39-05:00

GiteaMirror commented

2026-05-18 05:16:39 -05:00

Originally created by @R-omk on GitHub (Mar 15, 2026).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/22697

Check Existing Issues

I have searched for any existing and/or related issues.
I have searched for any existing and/or related discussions.
I have also searched in the CLOSED issues AND CLOSED discussions and found no related items (your issue might already be addressed on the development branch!).
I am using the latest version of Open WebUI.

Installation Method

Docker

Open WebUI Version

v0.17.8

Ollama Version (if applicable)

No response

Operating System

Ubuntu 24.04

Browser (if applicable)

No response

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 provided every relevant configuration, setting, and environment variable used in my setup.
I have clearly listed every relevant configuration, custom setting, environment variable, and command-line option that influences my setup (such as Docker Compose overrides, .env values, browser settings, authentication configurations, etc).
I have documented step-by-step reproduction instructions that are precise, sequential, and leave nothing to interpretation. My steps:
Start with the initial platform/version/OS and dependencies used,
Specify exact install/launch/configure commands,
List URLs visited, user input (incl. example values/emails/passwords if needed),
Describe all options and toggles enabled or changed,
Include any files or environmental changes,
Identify the expected and actual result at each stage,
Ensure any reasonably skilled user can follow and hit the same issue.

Expected Behavior

When BYPASS_EMBEDDING_AND_RETRIEVAL is enabled, ALL files in a knowledge base should behave consistently:

Option A (preferred): All files (old and new) should work the same way

If bypass is enabled → Full text mode for ALL files
query_knowledge_files() should either work with full text or return a clear error

Option B: Clear user feedback

If bypass is enabled, warn users that new files won't be indexed for vector search
Or provide option to manually re-index files with the bypass setting

Option C: Intelligent fallback

Tools like query_knowledge_files() should detect when BYPASS_EMBEDDING_AND_RETRIEVAL is enabled
Fall back to full text search instead of vector search
Return consistent results from all files

Actual Behavior

When BYPASS_EMBEDDING_AND_RETRIEVAL is enabled AFTER files have already been indexed:

Old files (indexed BEFORE enabling the option):
- Vector search works (if indices exist)
- Files appear in query_knowledge_files() results
- Content is accessible via vector embeddings
New files (added AFTER enabling the option):
- NO vectors are created (bypass skips save_docs_to_vector_db)
- Files NEVER appear in query_knowledge_files() results
- Full text is available ONLY if file is attached to chat

This creates confusion because:

Users expect all files in the knowledge base to work the same way
There's no warning that new files won't be indexed
Tools like query_knowledge_files() don't know about the bypass setting
Search results are silently incomplete

Steps to Reproduce

Start Open WebUI with default configuration
Create a knowledge base (e.g., "Test KB")
Upload and index several files to "Test KB"
Verify files are indexed and searchable via query_knowledge_files()
Enable BYPASS_EMBEDDING_AND_RETRIEVAL = True in Settings > Documents
Upload a NEW file to the SAME "Test KB"
Use query_knowledge_files() to search for content from the new file
Observe: The new file never appears in search results, only old files work

Logs & Screenshots

N/A - Not applicable to this issue (behavioral/UX bug, not a crash or error).

Code evidence:

In backend/open_webui/routers/retrieval.py:1819-1827:

if request.app.state.config.BYPASS_EMBEDDING_AND_RETRIEVAL:
    Files.update_file_data_by_id(file.id, {"status": "completed"}, db=db)
    Files.update_file_hash_by_id(file.id, hash, db=db)
    return {
        "status": True,
        "collection_name": None,  # ← No vector collection created
        "filename": file.filename,
        "content": text_content,
    }

In backend/open_webui/tools/builtin.py:1902-1908:

if collection_names:
    query_results = await query_collection(
        collection_names=collection_names,
        queries=[query],
        embedding_function=embedding_function,  # ← Always attempts vector search
        k=count,
    )

Note: The query_knowledge_files() function does not check BYPASS_EMBEDDING_AND_RETRIEVAL and always attempts vector search, causing incomplete results when some files lack vector indices.

Additional Information

No response

Originally created by @R-omk on GitHub (Mar 15, 2026). Original GitHub issue: https://github.com/open-webui/open-webui/issues/22697 ### Check Existing Issues - [x] I have searched for any existing and/or related issues. - [x] I have searched for any existing and/or related discussions. - [x] I have also searched in the CLOSED issues AND CLOSED discussions and found no related items (your issue might already be addressed on the development branch!). - [x] I am using the latest version of Open WebUI. ### Installation Method Docker ### Open WebUI Version v0.17.8 ### Ollama Version (if applicable) _No response_ ### Operating System Ubuntu 24.04 ### Browser (if applicable) _No response_ ### 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 **provided every relevant configuration, setting, and environment variable used in my setup.** - [x] I have clearly **listed every relevant configuration, custom setting, environment variable, and command-line option that influences my setup** (such as Docker Compose overrides, .env values, browser settings, authentication configurations, etc). - [x] I have documented **step-by-step reproduction instructions that are precise, sequential, and leave nothing to interpretation**. My steps: - Start with the initial platform/version/OS and dependencies used, - Specify exact install/launch/configure commands, - List URLs visited, user input (incl. example values/emails/passwords if needed), - Describe all options and toggles enabled or changed, - Include any files or environmental changes, - Identify the expected and actual result at each stage, - Ensure any reasonably skilled user can follow and hit the same issue. ### Expected Behavior When `BYPASS_EMBEDDING_AND_RETRIEVAL` is enabled, ALL files in a knowledge base should behave consistently: **Option A (preferred):** All files (old and new) should work the same way - If bypass is enabled → Full text mode for ALL files - `query_knowledge_files()` should either work with full text or return a clear error **Option B:** Clear user feedback - If bypass is enabled, warn users that new files won't be indexed for vector search - Or provide option to manually re-index files with the bypass setting **Option C:** Intelligent fallback - Tools like `query_knowledge_files()` should detect when `BYPASS_EMBEDDING_AND_RETRIEVAL` is enabled - Fall back to full text search instead of vector search - Return consistent results from all files ### Actual Behavior When `BYPASS_EMBEDDING_AND_RETRIEVAL` is enabled **AFTER** files have already been indexed: 1. **Old files (indexed BEFORE enabling the option):** - Vector search works (if indices exist) - Files appear in `query_knowledge_files()` results - Content is accessible via vector embeddings 2. **New files (added AFTER enabling the option):** - NO vectors are created (bypass skips `save_docs_to_vector_db`) - Files NEVER appear in `query_knowledge_files()` results - Full text is available ONLY if file is attached to chat This creates confusion because: - Users expect all files in the knowledge base to work the same way - There's no warning that new files won't be indexed - Tools like `query_knowledge_files()` don't know about the bypass setting - Search results are silently incomplete ### Steps to Reproduce 1. Start Open WebUI with default configuration 2. Create a knowledge base (e.g., "Test KB") 3. Upload and index several files to "Test KB" 4. Verify files are indexed and searchable via `query_knowledge_files()` 5. Enable `BYPASS_EMBEDDING_AND_RETRIEVAL = True` in Settings > Documents 6. Upload a NEW file to the SAME "Test KB" 7. Use `query_knowledge_files()` to search for content from the new file 8. **Observe:** The new file never appears in search results, only old files work ### Logs & Screenshots **N/A** - Not applicable to this issue (behavioral/UX bug, not a crash or error). **Code evidence:** In `backend/open_webui/routers/retrieval.py:1819-1827`: ```python if request.app.state.config.BYPASS_EMBEDDING_AND_RETRIEVAL: Files.update_file_data_by_id(file.id, {"status": "completed"}, db=db) Files.update_file_hash_by_id(file.id, hash, db=db) return { "status": True, "collection_name": None, # ← No vector collection created "filename": file.filename, "content": text_content, } ``` In `backend/open_webui/tools/builtin.py:1902-1908`: ```python if collection_names: query_results = await query_collection( collection_names=collection_names, queries=[query], embedding_function=embedding_function, # ← Always attempts vector search k=count, ) ``` **Note:** The `query_knowledge_files()` function does not check `BYPASS_EMBEDDING_AND_RETRIEVAL` and always attempts vector search, causing incomplete results when some files lack vector indices. ### Additional Information _No response_

GiteaMirror added the bug label 2026-05-18 05:16:39 -05:00

GiteaMirror closed this issue

2026-05-18 05:16:41 -05:00

GiteaMirror commented

2026-05-18 05:16:44 -05:00

@tjbck commented on GitHub (Mar 15, 2026):

Intended behaviour.

@tjbck commented on GitHub (Mar 15, 2026): Intended behaviour.

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#106787