Long assistant messages can contain Open WebUI structural <details> blocks for reasoning and tool calls. When a structural tool block follows assistant prose without a block boundary, Marked can parse the raw tag and escaped attributes as ordinary markdown, causing broken rendering and excessive DOM text in large chats.
What changed
Normalize only Open WebUI structural details blocks before markdown lexing.
Preserve consecutive structural details so grouped reasoning/tool rendering remains intact.
Add focused unit coverage for boundary insertion, consecutive blocks, and unmatched/ordinary details.
Testing
npm run test:frontend -- --run
Production image build completed successfully.
Manual logged-in browser checks on a large chat with reasoning/tool details: desktop, mobile viewport, and Chromium with JavaScript JIT disabled. No raw structural <details> tags or escaped > / " entities were visible before or after expanding grouped details.
🔄 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/24966
**Author:** [@noufalkv](https://github.com/noufalkv)
**Created:** 5/20/2026
**Status:** ❌ Closed
**Base:** `dev` ← **Head:** `perf/tor-chat-rendering`
---
### 📝 Commits (10+)
- [`9d87e3f`](https://github.com/open-webui/open-webui/commit/9d87e3f2e5a32b6c5dd11d8c0e1a405b5ea66b78) perf(codeblock): avoid eager work for streaming code blocks
- [`a94b0e6`](https://github.com/open-webui/open-webui/commit/a94b0e67a8727da2143925a129188d0a2d979ccc) perf(tool-calls): lazy-decode collapsed tool results
- [`27767e4`](https://github.com/open-webui/open-webui/commit/27767e41b1146e077169152e6272bbe0ed3a0ecf) perf(markdown): keep tool-call details out of markdown recursion
- [`ed4bd3a`](https://github.com/open-webui/open-webui/commit/ed4bd3aecb592e8339bc7a4ed180da730c809ff3) fix(chat): prevent stale high-latency chat loads
- [`830fbfc`](https://github.com/open-webui/open-webui/commit/830fbfc6cea09cad13934047dffa8534694e5fd0) perf(streaming): remove noisy completion proxy logs
- [`46be871`](https://github.com/open-webui/open-webui/commit/46be871d946e407b8f2bbc6129a1175d0d0966ba) fix(chat): show navigation feedback before autosave
- [`bcf336f`](https://github.com/open-webui/open-webui/commit/bcf336fcaad020c49cfa25b4a8887de0f5dd165d) perf(sidebar): parallelize initial chat metadata loads
- [`2c50e56`](https://github.com/open-webui/open-webui/commit/2c50e56ab3246a88820fbf01066953f75a720225) fix(markdown): keep reasoning details out of markdown recursion
- [`c34221b`](https://github.com/open-webui/open-webui/commit/c34221b7c6d5e16c656b8aae5fd621fe13ba6299) build(frontend): make sourcemap generation configurable
- [`61b2192`](https://github.com/open-webui/open-webui/commit/61b21920a341c37183d361db70d4ec0c8027f2a5) fix(markdown): separate structural details from prose
### 📊 Changes
**11 files changed** (+500 additions, -149 deletions)
<details>
<summary>View changed files</summary>
📝 `Dockerfile` (+4 -0)
📝 `src/lib/components/chat/Chat.svelte` (+64 -30)
📝 `src/lib/components/chat/Messages/CodeBlock.svelte` (+86 -10)
📝 `src/lib/components/chat/Messages/Markdown.svelte` (+4 -1)
📝 `src/lib/components/chat/Messages/Markdown/MarkdownTokens.svelte` (+101 -68)
📝 `src/lib/components/common/ToolCallDisplay.svelte` (+76 -19)
📝 `src/lib/components/layout/Sidebar.svelte` (+11 -11)
➕ `src/lib/utils/marked/details-normalization.test.ts` (+92 -0)
➕ `src/lib/utils/marked/details-normalization.ts` (+59 -0)
📝 `src/routes/+layout.svelte` (+2 -9)
📝 `vite.config.ts` (+1 -1)
</details>
### 📄 Description
## Why
Long assistant messages can contain Open WebUI structural `<details>` blocks for reasoning and tool calls. When a structural tool block follows assistant prose without a block boundary, Marked can parse the raw tag and escaped attributes as ordinary markdown, causing broken rendering and excessive DOM text in large chats.
## What changed
- Normalize only Open WebUI structural details blocks before markdown lexing.
- Preserve consecutive structural details so grouped reasoning/tool rendering remains intact.
- Add focused unit coverage for boundary insertion, consecutive blocks, and unmatched/ordinary details.
## Testing
- `npm run test:frontend -- --run`
- Production image build completed successfully.
- Manual logged-in browser checks on a large chat with reasoning/tool details: desktop, mobile viewport, and Chromium with JavaScript JIT disabled. No raw structural `<details>` tags or escaped `>` / `"` entities were visible before or after expanding grouped details.
---
<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/24966
Author: @noufalkv
Created: 5/20/2026
Status: ❌ Closed
Base:
dev← Head:perf/tor-chat-rendering📝 Commits (10+)
9d87e3fperf(codeblock): avoid eager work for streaming code blocksa94b0e6perf(tool-calls): lazy-decode collapsed tool results27767e4perf(markdown): keep tool-call details out of markdown recursioned4bd3afix(chat): prevent stale high-latency chat loads830fbfcperf(streaming): remove noisy completion proxy logs46be871fix(chat): show navigation feedback before autosavebcf336fperf(sidebar): parallelize initial chat metadata loads2c50e56fix(markdown): keep reasoning details out of markdown recursionc34221bbuild(frontend): make sourcemap generation configurable61b2192fix(markdown): separate structural details from prose📊 Changes
11 files changed (+500 additions, -149 deletions)
View changed files
📝
Dockerfile(+4 -0)📝
src/lib/components/chat/Chat.svelte(+64 -30)📝
src/lib/components/chat/Messages/CodeBlock.svelte(+86 -10)📝
src/lib/components/chat/Messages/Markdown.svelte(+4 -1)📝
src/lib/components/chat/Messages/Markdown/MarkdownTokens.svelte(+101 -68)📝
src/lib/components/common/ToolCallDisplay.svelte(+76 -19)📝
src/lib/components/layout/Sidebar.svelte(+11 -11)➕
src/lib/utils/marked/details-normalization.test.ts(+92 -0)➕
src/lib/utils/marked/details-normalization.ts(+59 -0)📝
src/routes/+layout.svelte(+2 -9)📝
vite.config.ts(+1 -1)📄 Description
Why
Long assistant messages can contain Open WebUI structural
<details>blocks for reasoning and tool calls. When a structural tool block follows assistant prose without a block boundary, Marked can parse the raw tag and escaped attributes as ordinary markdown, causing broken rendering and excessive DOM text in large chats.What changed
Testing
npm run test:frontend -- --run<details>tags or escaped>/"entities were visible before or after expanding grouped details.🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.