v0.0.5
Highlights
New Desktop Overlay Experience
- Added a dedicated always-on-top Overlay window for quick AI access outside the main app.
- Added compact and expanded overlay modes with improved sizing, positioning, and visual polish.
- Added global shortcut support for opening the Overlay directly at the cursor.
- Added Overlay settings for enablement, startup behavior, sizing, anchoring, hotkey configuration, and prompt auto-hide.
- Added a lightweight Prompt Popup flow that can submit prompts directly into the Overlay.
Provider & Model Improvements
- Added Fireworks AI as a first-class provider across chat, model selection, settings, usage metrics, and title generation.
- Added Fireworks model discovery support.
- Improved OpenRouter custom model setup with automatic model metadata autofill.
- Added provider dashboard shortcut buttons in Provider Hub.
- Improved Alibaba model handling, including automatic deep-thinking capability inference.
- Improved OpenRouter image-generation support in chat streaming, including persistence and inline rendering of generated images.
Research & Web Search
- Rebuilt the built-in web search system into a modular main-process service.
- Added Tavily and DuckDuckGo fallback support.
- Added configurable Tavily search depth / search speed preferences.
- Added batched web search support for parallel research queries.
- Improved research-loop recovery when providers return blank, repeated, or tool-only responses.
- Improved web search logging, fallback behavior, query trimming, and validation.
- Added optional image inclusion preference for web research results.
Chat & Streaming
- Unified provider streaming through a shared provider runtime and normalized streaming pipeline.
- Send and regenerate now use the same streaming orchestration path.
- Improved tool-call visibility during active responses.
- Persisted tool-only assistant responses correctly.
- Improved reasoning/search/tool timeline rendering.
- Improved final answer synthesis after research/tool rounds.
- Improved token usage reporting so stats reflect the final visible answer round.
UI & UX Polish
- Refactored the message renderer into smaller, focused components.
- Improved response info popovers, usage stats, and assistant message actions.
- Improved model selector layout, keyboard navigation, provider rail, and capability badges.
- Flattened settings cards, dropdowns, popovers, and menu surfaces for a cleaner visual style.
- Improved sidebar behavior, titlebar controls, shell navigation, and mouse-button navigation.
- Added scoped Select All behavior for chat content.
- Improved copy/paste reliability with clipboard utility fallbacks.
Security & Stability
- Removed the deprecated terminal bridge.
- Enabled sandboxing for Electron windows.
- Tightened preload and IPC boundaries.
- Added safer external URL handling.
- Added atomic writes for chat and secure storage to reduce corruption risk.
- Required encrypted secure storage for secrets where OS-backed encryption is available.
- Added clipboard read IPC through a narrow trusted bridge for context-menu paste fallback.
MCP Improvements
- Split MCP manager logic into smaller helper modules.
- Improved MCP runtime metadata persistence and reduced redundant disk writes.
- Improved MCP tool exposure policy handling.
- Improved approval handling and renderer payload behavior.
- Added tests around MCP storage, manager behavior, transports, and approval flows.
Build & Tooling
- Switched project scripts and CI workflows toward Bun.
- Added reusable Bun setup for GitHub Actions.
- Updated release workflow to use Bun-based commands.
- Removed Playwright dependency and unused Settings barrel exports.
- Added more tests across provider runtime, streaming, MCP, model selector, settings, overlay, and web search behavior.
Windows Installer
Windows installer artifact:
ZuraAI-Setup-0.0.5.exe
SHA-256:
8EBEA455375CE7B2B99F41E0ECC0F0FCDD21F9F4DF4035E7E204A48EF6401E3C