SDK Refresh & Startup Stability

What's New

• Latest AI SDKs — Updated Claude Agent SDK, Codex SDK, OpenCode SDK, @ai-sdk/react, and ai to current versions
• More Reliable OpenCode Loading — Added health checks and request timeouts around OpenCode provider loading to avoid hanging requests
• Clearer OpenCode Fallback State — OpenCode model selection now shows an explicit unavailable state instead of feeling like an endless loader

Fixed

• Startup Loader Crash — Fixed a renderer crash caused by unexpected provider payload shapes, which could leave the app stuck on the logo loading screen in dev
• OpenCode Provider Fetch Failure Handling — Improved behavior when the local OpenCode server is unavailable or unhealthy

Improvements

• Better resilience at the renderer boundary by normalizing provider-list responses before UI consumption
• Improved release metadata consistency with the version bump to v0.0.15

Claude Opus 4.6 & GPT-5.3-Codex support

Added

• Claude Opus 4.6 model support with enhanced coding capabilities
• GPT-5.3-Codex model support for improved code generation

File Indexing & Notifications

This release introduces project file indexing for fast search, desktop notification preferences, and improved file search with fuzzy matching.

What's New

• Project File Indexing — SQLite-based file indexing with FTS5 for fast, full-text search across project files
• Desktop Notifications — New notification preferences with notifications integrated directly into the chat flow
• Smarter File Search — Path-segment fuzzy matching for more accurate and flexible file search results

Improvements

• Enhanced multi-repo support with refactored diff management
• Noisy files are now excluded from diffs for cleaner output
• General code cleanup and improved readability
• Removed unnecessary console logs from file watcher and diff management hooks

Message Queue

This release adds a message queue system for improved streaming message handling with visual feedback.

What's New

• Message Queue System — Streaming messages are now queued with a dedicated UI component showing queued message status
• Integrated Status Display — Message queue display is now part of the sub-chat status card for a cleaner, unified experience

Changes

• Removed standalone message queue component in favor of the integrated status card display

Real-time Git Status & Multi-Provider Plan Mode

This release brings significant improvements to git workflow responsiveness and expands plan mode support to additional AI providers.

Highlights

Real-time Git Status — Git status now uses a smart caching layer with file watcher integration, giving you instant feedback as files change without hammering the filesystem.

Live Diff Stats — Watch diff statistics update in real-time as Claude streams changes to your files.

Plan Mode for Codex & OpenCode — Both Codex and OpenCode providers now support plan mode with proper function call handling, giving you more options for the "read first, act later" workflow.

Batch Discard — Select multiple changed files and discard them all at once from the changes panel.

What's New

Added

• Live updates for diff stats during streaming with regular refresh intervals
• Status cache for git status results with file watcher integration for real-time updates
• Plan mode support for Codex and OpenCode providers with function call item handling
• Batch discard changes for multiple files with enhanced selection handling in changes file list
• Output buffer for terminal sessions with enhanced log retrieval

Changed

• Enhanced worktree cleanup during project deletion
• Improved chat deletion confirmation dialog in UI
• Updated screenshot for improved visual representation

Fixed

• Background color for empty state indicators in sidebar components
• Formatting in Electron application diagram

Ralph completion small fix

A small release that removes legacy completion signaling and simplifies Ralph's prompt instructions.

Changed

• Removed legacy <promise>COMPLETE</promise> completion signal handling in the Ralph orchestrator.
• Simplified Ralph prompt instructions for final story completion.

Terminal transparency polish & refreshed media

A small release focused on terminal polish and documentation refreshes.

Added

• Terminal transparency support for scrollable elements.

Changed

• Refined terminal UI components for cleaner layout and interactions.
• Updated README with refreshed screenshots and media assets.

Providers, previews, and polish

What's New

• Provider management updates with improved terminal transparency handling
• Local preview viewport presets with better error handling

Improvements

• Chat view mode integration and clearer sidebar visibility behavior

Maintenance

• Removed unused workflow scripts from the package setup
• Removed the legacy fizzy-prancing-pinwheel plan document

Worktrees, Auto-Start & Dev Preview

Smarter worktree handling, Ralph auto-start for hands-free story continuation, and a built-in local dev server preview — plus sidebar UX improvements and faster git status updates.

What's New

• Local dev server preview — Configure a runCommand in project settings to launch and preview your dev server directly inside the app
• Ralph auto-start — Ralph can now automatically continue building stories without manual intervention, with fallback handling for seamless continuation
• Sidebar collapse states — Pinned, recent, and drafts sections in the sidebar are now independently collapsible
• Project settings in header — Quick access to project settings dialog from the ProjectSelectorHeader

Improvements

• Branch-based worktree paths — Worktree directories now use the branch name instead of chatId, making them easier to identify in your filesystem
• Faster git status — Git status watcher now uses a max-wait timer to batch rapid file system events and reduce redundant updates
• Transparent UI elements — SiriOrb and NewChatForm now use transparent backgrounds for better theme integration

Housekeeping

• Enabled format-on-save with quickfix and import organization
• Removed obsolete DIFF.md and OPENCODE.mdx files

Keybindings, Compaction & Per-Chat Models

This release introduces a customizable keybindings system, session compaction for long conversations, and the ability to use different AI models per sub-chat — plus major internal improvements to the chat architecture.

What's New

Customizable Keybindings

A centralized keybindings system that lets you view and override keyboard shortcuts to match your workflow.

Session Compaction (/compact)

Long conversations can now be compacted to reduce context size while preserving key information. Use the /compact command to summarize and trim your session history.

Per-Subchat Model Selection

Each sub-chat tab can now use a different AI model. Switch between Opus, Sonnet, and Haiku within the same chat session — useful for using a faster model on simple tasks and a more capable one for complex reasoning.

Ralph Auto-Continue

Ralph mode now automatically continues execution across user stories, reducing manual intervention when working through PRDs.

Improvements

• Refactored active-chat into focused, composable hooks for better maintainability
• Added Windows and Linux build jobs to the release CI workflow

Full Changelog

See CHANGELOG.md for the complete list of changes.