[GH-ISSUE #23564] fix: search_notes returns empty results for all multi-word queries #58679

New Issue

GiteaMirror · 2026-05-05T23:41:48-05:00

GiteaMirror commented

2026-05-05 23:41:48 -05:00

Originally created by @jlgill on GitHub (Apr 10, 2026).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/23564

Expected Behavior

When a model calls search_notes with a multi-word query containing spaces and/or hyphens (e.g., "to-do list reminders"), it should return notes whose title or content contains all of those terms somewhere in the text—regardless of where each term appears.

Actual Behavior

search_notes returns [] for every multi-word query. Only single-word or single hyphenated-term queries produce results.

Steps to Reproduce

Install Open WebUI v0.8.12 via Docker with PostgreSQL backend.
Create a Note with a hyphenated term in the title (e.g., My To-Do List) and body content containing additional words like reminders, schedule, etc.
Create a custom model with Builtin Tools > Notes enabled, using Native function calling.
Start a chat and ask something that would trigger a search_notes call with multiple terms (e.g., "What does my to-do list say about reminders?").
Observe the search_notes tool call returns [] despite the note containing all search terms.
A single-term query like to-do works correctly and returns the note.

Root cause: In backend/open_webui/models/notes.py, search_notes() normalizes the query by stripping all hyphens and spaces from the entire string before performing a single ILIKE match:

normalized_query = query_key.replace('-', '').replace(' ', '')
# "to-do list reminders" -> "todolistreminders" -- never matches as a contiguous substring

The hyphen-stripping logic was intended to let todo match to-do (single-term normalization), but applying it to the whole multi-word query concatenates all terms into one string that never appears contiguously in any note content.

Fix: Split the query into individual terms, normalize each independently (strip hyphens), and require all terms to match somewhere in the title or content (AND semantics). Single-term behavior is unchanged.

Additionally, snippet extraction in tools/builtin.py attempts to find the entire multi-word query as one contiguous substring to center the preview—which also never matches for multi-word input. Fixed to locate the first matching individual term instead.

Logs & Screenshots

Tool call trace from chat history (PostgreSQL):

name="search_notes" arguments='{"query": "multi word query with hyphens"}' result="[]"
name="search_notes" arguments='{"query": "another multi term search"}' result="[]"

After fix, validated directly against PostgreSQL and inside the container:

Multi-word query with 4 terms: 1 result (was 0)
Two-term query: 2 results (was 0)
Single-term query: 3 results (unchanged -- no regression)

Additional Information

This also affects the scenario in #23362 -- when notes are attached to a folder and the model uses native function calling, notes are not auto-injected into context. The LLM falls back to search_notes, which hits this SQL bug and returns empty.
Fix branch: jlgill/gillson-open-webui:fix/search-notes-per-term-matching (rebased on dev)
Files changed: backend/open_webui/models/notes.py, backend/open_webui/tools/builtin.py

Installation Method: Docker
Open WebUI Version: v0.8.12
Operating System: Windows 11 / WSL2 + Docker Desktop
Browser: Chrome 136

Originally created by @jlgill on GitHub (Apr 10, 2026). Original GitHub issue: https://github.com/open-webui/open-webui/issues/23564 ### Expected Behavior When a model calls `search_notes` with a multi-word query containing spaces and/or hyphens (e.g., `"to-do list reminders"`), it should return notes whose title or content contains all of those terms somewhere in the text—regardless of where each term appears. ### Actual Behavior `search_notes` returns `[]` for **every** multi-word query. Only single-word or single hyphenated-term queries produce results. ### Steps to Reproduce 1. Install Open WebUI v0.8.12 via Docker with PostgreSQL backend. 2. Create a Note with a hyphenated term in the title (e.g., `My To-Do List`) and body content containing additional words like `reminders`, `schedule`, etc. 3. Create a custom model with Builtin Tools > Notes enabled, using Native function calling. 4. Start a chat and ask something that would trigger a `search_notes` call with multiple terms (e.g., "What does my to-do list say about reminders?"). 5. Observe the `search_notes` tool call returns `[]` despite the note containing all search terms. 6. A single-term query like `to-do` works correctly and returns the note. **Root cause:** In `backend/open_webui/models/notes.py`, `search_notes()` normalizes the query by stripping **all** hyphens and spaces from the entire string before performing a single `ILIKE` match: ```python normalized_query = query_key.replace('-', '').replace(' ', '') # "to-do list reminders" -> "todolistreminders" -- never matches as a contiguous substring ``` The hyphen-stripping logic was intended to let `todo` match `to-do` (single-term normalization), but applying it to the whole multi-word query concatenates all terms into one string that never appears contiguously in any note content. **Fix:** Split the query into individual terms, normalize each independently (strip hyphens), and require all terms to match somewhere in the title or content (AND semantics). Single-term behavior is unchanged. Additionally, snippet extraction in `tools/builtin.py` attempts to find the entire multi-word query as one contiguous substring to center the preview—which also never matches for multi-word input. Fixed to locate the first matching individual term instead. ### Logs & Screenshots Tool call trace from chat history (PostgreSQL): ``` name="search_notes" arguments='{"query": "multi word query with hyphens"}' result="[]" name="search_notes" arguments='{"query": "another multi term search"}' result="[]" ``` After fix, validated directly against PostgreSQL and inside the container: ``` Multi-word query with 4 terms: 1 result (was 0) Two-term query: 2 results (was 0) Single-term query: 3 results (unchanged -- no regression) ``` ### Additional Information - This also affects the scenario in #23362 -- when notes are attached to a folder and the model uses native function calling, notes are not auto-injected into context. The LLM falls back to `search_notes`, which hits this SQL bug and returns empty. - Fix branch: [jlgill/gillson-open-webui:fix/search-notes-per-term-matching](https://github.com/jlgill/gillson-open-webui/tree/fix/search-notes-per-term-matching) (rebased on `dev`) - Files changed: `backend/open_webui/models/notes.py`, `backend/open_webui/tools/builtin.py` **Installation Method:** Docker **Open WebUI Version:** v0.8.12 **Operating System:** Windows 11 / WSL2 + Docker Desktop **Browser:** Chrome 136

GiteaMirror closed this issue

2026-05-05 23:41:48 -05:00

GiteaMirror commented

2026-05-05 23:41:49 -05:00

@Classic298 commented on GitHub (Apr 10, 2026):

Stop spamming our discussions and issues, you commented this, and opened 2 issues and 5 discussions with the same exact text in 3 minutes or so. STOP. Final warning.

@Classic298 commented on GitHub (Apr 10, 2026): Stop spamming our discussions and issues, you commented this, and opened 2 issues and 5 discussions with the same exact text in 3 minutes or so. STOP. Final warning.

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#58679