b3a8fa6 fix: skip wrap for native reasoning_content round-trip
📊 Changes
1 file changed (+9 additions, -4 deletions)
View changed files
📝backend/open_webui/utils/misc.py (+9 -4)
📄 Description
Chat templates that don't strip (e.g. Gemma 4) leak the tags into the prompt, causing reasoning markup to bleed into the visible reply after a tool call. Only add the wrap for inline-tag-detected reasoning; native-field reasoning ships as reasoning_content alone.
Pull Request Checklist
Note to first-time contributors: Please open a discussion post in Discussions to discuss your idea/fix with the community before creating a pull request, and describe your changes before submitting a pull request.
This is to ensure large feature PRs are discussed with the community first, before starting work on it. If the community does not want this feature or it is not relevant for Open WebUI as a project, it can be identified in the discussion before working on the feature and submitting the PR.
Before submitting, make sure you've checked the following:
Target branch: Verify that the pull request targets the dev branch. PRs targeting main will be immediately closed.
Description: Provide a concise description of the changes made in this pull request down below.
Changelog: Ensure a changelog entry following the format of Keep a Changelog is added at the bottom of the PR description.
Documentation: Add docs in Open WebUI Docs Repository. Document user-facing behavior, environment variables, public APIs/interfaces, or deployment steps.
Dependencies: Are there any new or upgraded dependencies? If so, explain why, update the changelog/docs, and include any compatibility notes. Actually run the code/function that uses updated library to ensure it doesn't crash.
Testing: Perform manual tests to verify the implemented fix/feature works as intended AND does not break any other functionality. Include reproducible steps to demonstrate the issue before the fix. Test edge cases (URL encoding, HTML entities, types). Take this as an opportunity to make screenshots of the feature/fix and include them in the PR description.
Agentic AI Code: Confirm this Pull Request is not written by any AI Agent or has at least gone through additional human review AND manual testing. If any AI Agent is the co-author of this PR, it may lead to immediate closure of the PR.
Code review: Have you performed a self-review of your code, addressing any coding standard issues and ensuring adherence to the project's coding standards?
Design & Architecture: Prefer smart defaults over adding new settings; use local state for ephemeral UI logic. Open a Discussion for major architectural or UX changes.
Git Hygiene: Keep PRs atomic (one logical change). Clean up commits and rebase on dev to ensure no unrelated commits (e.g. from main) are included. Push updates to the existing PR branch instead of closing and reopening.
Title Prefix: To clearly categorize this pull request, prefix the pull request title using one of the following:
BREAKING CHANGE: Significant changes that may affect compatibility
build: Changes that affect the build system or external dependencies
ci: Changes to our continuous integration processes or workflows
chore: Refactor, cleanup, or other non-functional code changes
docs: Documentation update or addition
feat: Introduces a new feature or enhancement to the codebase
fix: Bug fix or error correction
i18n: Internationalization or localization changes
perf: Performance improvement
refactor: Code restructuring for better maintainability, readability, or scalability
style: Changes that do not affect the meaning of the code (white space, formatting, missing semi-colons, etc.)
test: Adding missing tests or correcting existing tests
WIP: Work in progress, a temporary label for incomplete or ongoing work
Changelog Entry
Description
convert_output_to_messages emits reasoning received via native provider fields (delta.reasoning_content/.reasoning/.thinking) only as the structured reasoning_content field on the assistant message, without the additional <think>…</think> wrap inside content. Reasoning detected from inline tags is unchanged. Fixes a regression introduced by PR #23742 where chat templates that don't strip <think> (Gemma 4's strip_thinking only removes <|channel>…<channel|>, same pattern for gpt-oss harmony and Qwen3-thinking) passed the foreign tags into the prompt after a tool round-trip, leaking literal thought / <channel|> markup into the visible reply.
Added
N/A
Changed
Reasoning items with attributes.type == 'reasoning_content' are no longer duplicated as a <think>…</think> wrap in content when reconstructing assistant history for the backend; only the structured reasoning_content field is sent. Inline-tag-detected reasoning keeps the existing behavior.
Deprecated
N/A
Removed
N/A
Fixed
Visible thought / <channel|> leak in replies after native tool calls on reasoning-capable models whose chat templates consume reasoning_content natively but do not strip <think> from content (Gemma 4, gpt-oss harmony, Qwen3-thinking). Regression of PR #23742.
Security
N/A
Breaking Changes
None.
Additional Information
Problem. The partial fix in 3dd825581 introduced pending_reasoning so assistant tool-call messages include a reasoning_content field (needed by Kimi K2.5). It still also emits a <think>…</think> wrap in content. Gemma 4's strip_thinking macro only strips <|channel>…<channel|>, so the <think> tags pass into the prompt. The model's own channel markers then either get rendered as empty special tokens or bleed into delta.content, producing visible thought\n<channel|>… leakage.
Fix. Branch on the existing provenance marker attributes.type == 'reasoning_content' (set at middleware.py:3774 only for reasoning from native provider fields, and already used at middleware.py:3852 by the inside_tag_block guard). For those items, emit only the structured field; skip the <think> wrap. Inline-tag-detected items keep the wrap since those models emit and expect inline tags.
Scope. One function, one branch. No changes to DB schema, frontend rendering, export, or other call sites. No new config.
Manual verification. Direct API curl: before, delta.content starts with literal thought / \n / <channel|>; after, reasoning streams in delta.reasoning_content and content stays clean. End-to-end via OpenWebUI + Gemma 4 + a Python tool: reply renders a clean "Thought for N seconds" collapsible.
Screenshots or Videos
Before:
After:
Contributor License Agreement
By submitting this pull request, I confirm that I have read and fully agree to the Contributor License Agreement (CLA), and I am providing my contributions under its terms.
Note
Deleting the CLA section will lead to immediate closure of your PR and it will not be merged in.
🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.
## 📋 Pull Request Information
**Original PR:** https://github.com/open-webui/open-webui/pull/23843
**Author:** [@nulltea](https://github.com/nulltea)
**Created:** 4/17/2026
**Status:** ❌ Closed
**Base:** `dev` ← **Head:** `fix/reasoning-content-roundtrip-native-path`
---
### 📝 Commits (1)
- [`b3a8fa6`](https://github.com/open-webui/open-webui/commit/b3a8fa6a983301603514eb866d39d00f33c2d1c1) fix: skip <think> wrap for native reasoning_content round-trip
### 📊 Changes
**1 file changed** (+9 additions, -4 deletions)
<details>
<summary>View changed files</summary>
📝 `backend/open_webui/utils/misc.py` (+9 -4)
</details>
### 📄 Description
Chat templates that don't strip <think> (e.g. Gemma 4) leak the tags into the prompt, causing reasoning markup to bleed into the visible reply after a tool call. Only add the wrap for inline-tag-detected reasoning; native-field reasoning ships as reasoning_content alone.
<!--
⚠️ CRITICAL CHECKS FOR CONTRIBUTORS (READ, DON'T DELETE) ⚠️
1. Target the `dev` branch. PRs targeting `main` will be automatically closed.
2. Do NOT delete the CLA section at the bottom. It is required for the bot to accept your PR.
-->
# Pull Request Checklist
### Note to first-time contributors: Please open a discussion post in [Discussions](https://github.com/open-webui/open-webui/discussions) to discuss your idea/fix with the community before creating a pull request, and describe your changes before submitting a pull request.
This is to ensure large feature PRs are discussed with the community first, before starting work on it. If the community does not want this feature or it is not relevant for Open WebUI as a project, it can be identified in the discussion before working on the feature and submitting the PR.
**Before submitting, make sure you've checked the following:**
- [x] **Target branch:** Verify that the pull request targets the `dev` branch. **PRs targeting `main` will be immediately closed.**
- [x] **Description:** Provide a concise description of the changes made in this pull request down below.
- [x] **Changelog:** Ensure a changelog entry following the format of [Keep a Changelog](https://keepachangelog.com/) is added at the bottom of the PR description.
- [x] **Documentation:** Add docs in [Open WebUI Docs Repository](https://github.com/open-webui/docs). Document user-facing behavior, environment variables, public APIs/interfaces, or deployment steps.
- [x] **Dependencies:** Are there any new or upgraded dependencies? If so, explain why, update the changelog/docs, and include any compatibility notes. Actually run the code/function that uses updated library to ensure it doesn't crash.
- [x] **Testing:** Perform manual tests to **verify the implemented fix/feature works as intended AND does not break any other functionality**. Include reproducible steps to demonstrate the issue before the fix. Test edge cases (URL encoding, HTML entities, types). Take this as an opportunity to **make screenshots of the feature/fix and include them in the PR description**.
- [x] **Agentic AI Code:** Confirm this Pull Request is **not written by any AI Agent** or has at least **gone through additional human review AND manual testing**. If any AI Agent is the co-author of this PR, it may lead to immediate closure of the PR.
- [x] **Code review:** Have you performed a self-review of your code, addressing any coding standard issues and ensuring adherence to the project's coding standards?
- [x] **Design & Architecture:** Prefer smart defaults over adding new settings; use local state for ephemeral UI logic. Open a Discussion for major architectural or UX changes.
- [x] **Git Hygiene:** Keep PRs atomic (one logical change). Clean up commits and rebase on `dev` to ensure no unrelated commits (e.g. from `main`) are included. Push updates to the existing PR branch instead of closing and reopening.
- [x] **Title Prefix:** To clearly categorize this pull request, prefix the pull request title using one of the following:
- **BREAKING CHANGE**: Significant changes that may affect compatibility
- **build**: Changes that affect the build system or external dependencies
- **ci**: Changes to our continuous integration processes or workflows
- **chore**: Refactor, cleanup, or other non-functional code changes
- **docs**: Documentation update or addition
- **feat**: Introduces a new feature or enhancement to the codebase
- **fix**: Bug fix or error correction
- **i18n**: Internationalization or localization changes
- **perf**: Performance improvement
- **refactor**: Code restructuring for better maintainability, readability, or scalability
- **style**: Changes that do not affect the meaning of the code (white space, formatting, missing semi-colons, etc.)
- **test**: Adding missing tests or correcting existing tests
- **WIP**: Work in progress, a temporary label for incomplete or ongoing work
# Changelog Entry
### Description
- `convert_output_to_messages` emits reasoning received via native provider fields (`delta.reasoning_content`/`.reasoning`/`.thinking`) only as the structured `reasoning_content` field on the assistant message, without the additional `<think>…</think>` wrap inside `content`. Reasoning detected from inline tags is unchanged. Fixes a regression introduced by PR #23742 where chat templates that don't strip `<think>` (Gemma 4's `strip_thinking` only removes `<|channel>…<channel|>`, same pattern for gpt-oss harmony and Qwen3-thinking) passed the foreign tags into the prompt after a tool round-trip, leaking literal `thought` / `<channel|>` markup into the visible reply.
### Added
- N/A
### Changed
- Reasoning items with `attributes.type == 'reasoning_content'` are no longer duplicated as a `<think>…</think>` wrap in `content` when reconstructing assistant history for the backend; only the structured `reasoning_content` field is sent. Inline-tag-detected reasoning keeps the existing behavior.
### Deprecated
- N/A
### Removed
- N/A
### Fixed
- Visible `thought` / `<channel|>` leak in replies after native tool calls on reasoning-capable models whose chat templates consume `reasoning_content` natively but do not strip `<think>` from `content` (Gemma 4, gpt-oss harmony, Qwen3-thinking). Regression of PR #23742.
### Security
- N/A
### Breaking Changes
- None.
---
### Additional Information
**Problem.** The partial fix in 3dd825581 introduced `pending_reasoning` so assistant tool-call messages include a `reasoning_content` field (needed by Kimi K2.5). It still also emits a `<think>…</think>` wrap in `content`. Gemma 4's `strip_thinking` macro only strips `<|channel>…<channel|>`, so the `<think>` tags pass into the prompt. The model's own channel markers then either get rendered as empty special tokens or bleed into `delta.content`, producing visible `thought\n<channel|>…` leakage.
**Fix.** Branch on the existing provenance marker `attributes.type == 'reasoning_content'` (set at `middleware.py:3774` only for reasoning from native provider fields, and already used at `middleware.py:3852` by the `inside_tag_block` guard). For those items, emit only the structured field; skip the `<think>` wrap. Inline-tag-detected items keep the wrap since those models emit and expect inline tags.
**Scope.** One function, one branch. No changes to DB schema, frontend rendering, export, or other call sites. No new config.
**Manual verification.** Direct API curl: before, `delta.content` starts with literal `thought` / `\n` / `<channel|>`; after, reasoning streams in `delta.reasoning_content` and content stays clean. End-to-end via OpenWebUI + Gemma 4 + a Python tool: reply renders a clean "Thought for N seconds" collapsible.
### Screenshots or Videos
Before:
<img width="1076" height="957" alt="Zen 2026-04-17 15 03 27" src="https://github.com/user-attachments/assets/0e84eab1-2db2-4eb7-bd9e-8288a0257105" />
After:
<img width="1094" height="408" alt="Zen 2026-04-17 15 03 40" src="https://github.com/user-attachments/assets/63b1dd9b-f137-4c55-9294-2872db23eb88" />
### Contributor License Agreement
<!--
🚨 DO NOT DELETE THE TEXT BELOW 🚨
Keep the "Contributor License Agreement" confirmation text intact.
Deleting it will trigger the CLA-Bot to INVALIDATE your PR.
Your PR will NOT be reviewed or merged until you check the box below confirming that you have read and agree to the terms of the CLA.
-->
- [x] By submitting this pull request, I confirm that I have read and fully agree to the [Contributor License Agreement (CLA)](https://github.com/open-webui/open-webui/blob/main/CONTRIBUTOR_LICENSE_AGREEMENT), and I am providing my contributions under its terms.
> [!NOTE]
> Deleting the CLA section will lead to immediate closure of your PR and it will not be merged in.
---
<sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
📋 Pull Request Information
Original PR: https://github.com/open-webui/open-webui/pull/23843
Author: @nulltea
Created: 4/17/2026
Status: ❌ Closed
Base:
dev← Head:fix/reasoning-content-roundtrip-native-path📝 Commits (1)
b3a8fa6fix: skip wrap for native reasoning_content round-trip📊 Changes
1 file changed (+9 additions, -4 deletions)
View changed files
📝
backend/open_webui/utils/misc.py(+9 -4)📄 Description
Chat templates that don't strip (e.g. Gemma 4) leak the tags into the prompt, causing reasoning markup to bleed into the visible reply after a tool call. Only add the wrap for inline-tag-detected reasoning; native-field reasoning ships as reasoning_content alone.
Pull Request Checklist
Note to first-time contributors: Please open a discussion post in Discussions to discuss your idea/fix with the community before creating a pull request, and describe your changes before submitting a pull request.
This is to ensure large feature PRs are discussed with the community first, before starting work on it. If the community does not want this feature or it is not relevant for Open WebUI as a project, it can be identified in the discussion before working on the feature and submitting the PR.
Before submitting, make sure you've checked the following:
devbranch. PRs targetingmainwill be immediately closed.devto ensure no unrelated commits (e.g. frommain) are included. Push updates to the existing PR branch instead of closing and reopening.Changelog Entry
Description
convert_output_to_messagesemits reasoning received via native provider fields (delta.reasoning_content/.reasoning/.thinking) only as the structuredreasoning_contentfield on the assistant message, without the additional<think>…</think>wrap insidecontent. Reasoning detected from inline tags is unchanged. Fixes a regression introduced by PR #23742 where chat templates that don't strip<think>(Gemma 4'sstrip_thinkingonly removes<|channel>…<channel|>, same pattern for gpt-oss harmony and Qwen3-thinking) passed the foreign tags into the prompt after a tool round-trip, leaking literalthought/<channel|>markup into the visible reply.Added
Changed
attributes.type == 'reasoning_content'are no longer duplicated as a<think>…</think>wrap incontentwhen reconstructing assistant history for the backend; only the structuredreasoning_contentfield is sent. Inline-tag-detected reasoning keeps the existing behavior.Deprecated
Removed
Fixed
thought/<channel|>leak in replies after native tool calls on reasoning-capable models whose chat templates consumereasoning_contentnatively but do not strip<think>fromcontent(Gemma 4, gpt-oss harmony, Qwen3-thinking). Regression of PR #23742.Security
Breaking Changes
Additional Information
Problem. The partial fix in
3dd825581introducedpending_reasoningso assistant tool-call messages include areasoning_contentfield (needed by Kimi K2.5). It still also emits a<think>…</think>wrap incontent. Gemma 4'sstrip_thinkingmacro only strips<|channel>…<channel|>, so the<think>tags pass into the prompt. The model's own channel markers then either get rendered as empty special tokens or bleed intodelta.content, producing visiblethought\n<channel|>…leakage.Fix. Branch on the existing provenance marker
attributes.type == 'reasoning_content'(set atmiddleware.py:3774only for reasoning from native provider fields, and already used atmiddleware.py:3852by theinside_tag_blockguard). For those items, emit only the structured field; skip the<think>wrap. Inline-tag-detected items keep the wrap since those models emit and expect inline tags.Scope. One function, one branch. No changes to DB schema, frontend rendering, export, or other call sites. No new config.
Manual verification. Direct API curl: before,
delta.contentstarts with literalthought/\n/<channel|>; after, reasoning streams indelta.reasoning_contentand content stays clean. End-to-end via OpenWebUI + Gemma 4 + a Python tool: reply renders a clean "Thought for N seconds" collapsible.Screenshots or Videos
Before:
After:
Contributor License Agreement
🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.