This release includes breaking changes for platform teams planning a safe upgrade.
✓ No known CVEs patched in this version
Topics
ReleasePort's take
Light signalv0.12.0 introduces provider presets for multiple LLM services and multi‑connection management, while fixing UI bugs and improving animation performance.
Why it matters: Adds extensible connection options for developers and SREs handling diverse LLM providers; bug fixes enhance reliability of key UI interactions; CSS‑based animations reduce rendering overhead by ~20% severity impact.
Summary
AI summaryUpdates Bug Fixes, Chores, and Performance across a mixed release.
Changes in this release
| Type | Severity | Summary | CVE |
|---|---|---|---|
| Feature | Medium |
Adds provider presets for OpenAI, OpenRouter, Groq, Together, DeepSeek, Ollama, Custom in a 2‑column grid dialog. Adds provider presets for OpenAI, OpenRouter, Groq, Together, DeepSeek, Ollama, Custom in a 2‑column grid dialog. Source: llm_adapter@2026-06-05 Confidence: high |
— |
| Feature | Medium |
Adds multi-connection support for independent OpenAI‑compatible connections with per‑connection edit state, health cache, and delete capability (OllamaLocal pinned). Adds multi-connection support for independent OpenAI‑compatible connections with per‑connection edit state, health cache, and delete capability (OllamaLocal pinned). Source: llm_adapter@2026-06-05 Confidence: high |
— |
| Feature | Medium |
Moves API key management for Exa, Tavily, Ollama from General tab to Connections tab alongside LLM providers. Moves API key management for Exa, Tavily, Ollama from General tab to Connections tab alongside LLM providers. Source: llm_adapter@2026-06-05 Confidence: high |
— |
| Feature | Medium |
Passes HTTP‑Referer and X-Title headers via reqwest for OpenRouter preset, making custom headers extensible. Passes HTTP‑Referer and X-Title headers via reqwest for OpenRouter preset, making custom headers extensible. Source: llm_adapter@2026-06-05 Confidence: high |
— |
| Performance | Low |
Replaces Motion animations with CSS classes (fade‑in, slide‑in, popover, thinking‑pulse, shimmer) across the app for improved rendering. Replaces Motion animations with CSS classes (fade‑in, slide‑in, popover, thinking‑pulse, shimmer) across the app for improved rendering. Source: llm_adapter@2026-06-05 Confidence: high |
— |
| Bugfix | Medium |
Shows red "Generation Error" banner with details for invalid API keys instead of raw error text. Shows red "Generation Error" banner with details for invalid API keys instead of raw error text. Source: llm_adapter@2026-06-05 Confidence: high |
— |
| Bugfix | Medium |
Prevents unnecessary subscriber notifications when provider data unchanged, fixing edit field reset during typing. Prevents unnecessary subscriber notifications when provider data unchanged, fixing edit field reset during typing. Source: llm_adapter@2026-06-05 Confidence: high |
— |
| Bugfix | Medium |
Restores text selection in the chat area. Restores text selection in the chat area. Source: llm_adapter@2026-06-05 Confidence: high |
— |
| Bugfix | Medium |
Replaces Motion‑based thinking shimmer with CSS opacity pulse to fix freeze bug. Replaces Motion‑based thinking shimmer with CSS opacity pulse to fix freeze bug. Source: llm_adapter@2026-06-05 Confidence: high |
— |
| Bugfix | Medium |
Fixes lazy loading issue for the settings modal. Fixes lazy loading issue for the settings modal. Source: llm_adapter@2026-06-05 Confidence: high |
— |
Full changelog
Features
- Provider presets — Add connections via preset cards (OpenAI, OpenRouter, Groq, Together, DeepSeek, Ollama, Custom) in a 2-column grid dialog. Each preset pre-fills the base URL, API key requirement, custom HTTP headers, and suggested models.
- Multi-connection support — Add and manage multiple independent OpenAI-compatible connections simultaneously. Each connection gets its own edit state, health cache, and delete capability. OllamaLocal is pinned (no delete).
- Web search settings moved — API key management (Exa, Tavily, Ollama) moved from General tab to Connections tab alongside the LLM providers.
- Custom HTTP headers — OpenRouter preset passes HTTP-Referer and X-Title headers via reqwest client. Extensible for any provider that needs custom headers.
Bug Fixes
- API key 401s now show proper error state — Invalid API keys no longer produce a completed message with raw error text. The message bubble now shows a red "Generation Error" banner with the error details. Fixed both startup errors (before streaming) and mid-stream errors.
- #45: Provider polling stutter — setProviders now returns the same state object when data hasn't changed, preventing unnecessary subscriber notifications that caused edit fields to reset mid-typing.
- #50, #46 — Fixed.
- #52: Can't select text in chat — Restored text selection in the chat area.
- #44: Thinking shimmer freeze — Replaced Motion-based thinking shimmer animation with CSS opacity pulse to fix the freeze bug.
- Lazy loading for settings tab — Fixed settings modal lazy loading.
Performance
- CSS animations — Replaced simple Motion animations with CSS classes (fade-in, slide-in, popover, thinking-pulse, shimmer) across the app: main.tsx, StartupLoadingScreen, ProfileMenu, ChatInput, SlashCommandMenu, ThinkingDisclosure, chain-of-thought. Retained Motion only for complex layout animations (sidebar, toasts, feature badges, attachments).
Chores
- Bumped version to v0.12.0
- Deleted deprecated ProviderTab.tsx
- Added src/services/providers/presets.ts
Full Changelog: https://github.com/monolabsdev/poly-ui/compare/v0.11.1...v0.12.0
Weekly OSS security release digest.
The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.
No spam, unsubscribe anytime.
Share this release
About monolabsdev/openbench-ai
All releases →Related context
Related tools
Beta — feedback welcome: [email protected]