NextGPU-0.1.0-alpha

Release Notes

Highlights

Local AI chat experience

• Added a desktop chat experience with support for local model-based conversations.
• Added chat session management, including:
  • Creating new chats
  • Switching between chats
  • Deleting chats
  • Clearing recent chats
  • Preserving the selected model for a chat session
• Improved chat streaming behavior so responses appear progressively while they are generated.
• Added assistant “thinking” indicators during response generation.
• Improved reliability when switching sessions during active generation.
• Added safeguards to prevent duplicate or overlapping generation jobs in the same session.
• Improved final message saving and chat refresh behavior after generation completes.
• Added support for cancelling active generation.

Image generation support

• Added image-generation mode alongside text chat.
• Added a unified prompt input that supports:
  • Text prompts
  • Image prompts
  • Negative prompts
  • Size selection
  • Image model selection
• Added ComfyUI-backed image generation flow.
• Improved image generation robustness by:
  • Preventing concurrent image generations per session
  • Validating selected models
  • Showing clearer errors when generation fails
  • Saving generated image responses into the active chat
• Improved workflow construction for image models using model metadata instead of static templates.

Model management

• Added model discovery and management for locally available AI models.
• Added support for multiple model registries, including Ollama and ComfyUI-style models.
• Added model installation, deletion, pause, resume, and stop controls.
• Added progress reporting for model downloads.
• Added better handling for large downloads, including:
  • Partial download tracking
  • .part temporary files
  • Atomic moves after successful download
  • Resume support for interrupted downloads
  • Cleanup for failed or cancelled downloads
• Added UI states for downloading, paused, stopping, installed, and available models.
• Added confirmation dialogs before stopping downloads or deleting models.
• Improved sorting so active downloads appear prominently.
• Prevented in-progress models from appearing duplicated in installed and available lists.
• Improved progress parsing to avoid incorrect jumps to 100%.
• Added safer model deletion behavior for Ollama model name variants.
• Improved handling of incomplete Ollama model pulls.

ComfyUI model support

• Added support for ComfyUI model metadata, including:
  • Model files
  • Sampler
  • Scheduler
  • Registry type
  • Description fields
• Added ComfyUI model availability checks.
• Added ComfyUI model download and delete support.
• Added WSL-aware download handling for Windows environments.
• Added safer filesystem checks for model presence.
• Added registry-aware model keys so different model providers are handled correctly.

Windows and WSL setup

• Improved WSL startup handling by moving authentication and startup flow into the UI instead of blocking application startup.
• Added WSL authentication checks with clearer success/failure behavior.
• Added persistent WSL keep-alive handling.
• Added service startup handling inside WSL.
• Added automatic port proxy refresh for local services.
• Added manual WSL port proxy guidance for forwarding localhost ports.
• Improved WSL update logging and error handling.
• Added retry logic for Ollama installation inside WSL.
• Added verification that Ubuntu 24.04 is registered after installation.
• Lowered installation memory requirement checks to 12 GB.
• Improved installer preflight checks, including Windows 11 validation.
• Improved administrator privilege handling for the installer.
• Fixed desktop shortcut creation during installation.

Installer and packaging

• Switched Windows packaging to Inno Setup.
• Added installer configuration and uninstall scripts.
• Added cleanup behavior during uninstall.
• Updated application product naming, executable casing, icon, and installer metadata.
• Added build improvements for installer generation, including clearer errors when Inno Setup is missing.
• Added packaging output handling for generated installer artifacts.
• Bumped application version metadata for packaged builds.

Startup, shutdown, and reliability

• Improved application startup flow so WSL checks do not interrupt welcome/setup screens.
• Added protection against quitting while model downloads are in progress.
• Added confirmation behavior before exit when downloads are active.
• Improved shutdown analytics handling.
• Added better machine identity generation fallbacks.
• Added safer handling when no GPU is present or when only placeholder display devices are detected.
• Improved GPU detection behavior on systems without real NVIDIA hardware.
• Improved benchmark error handling to avoid null pointer failures and invalid results.
• Added clearer default error scores for failed storage benchmarks.

Hardware detection and benchmarking

• Improved CPU, GPU, and storage benchmark error handling.
• Added defensive checks for missing CPU data, missing WSL configuration, and malformed benchmark output.
• Improved GPU benchmark response validation.
• Refreshed CPU, GPU, and storage seed data.
• Improved handling of fake or placeholder GPU devices.
• Added CPU fallback behavior for analytics machine hashing.

User interface improvements

• Added a unified setup flow combining welcome, installation, and hardware setup.
• Improved welcome screen illustration and layout.
• Improved launch/setup screen text layout and styling.
• Added hover cursors for interactive controls.
• Improved button, checkbox, dropdown, and menu styling.
• Added private mode toggle support.
• Improved top navigation layout and pinned-message controls.
• Added selected chat item visual state.
• Improved starred/pinned message state handling.
• Added rename dialog with keyboard support.
• Improved project dialog keyboard navigation.
• Added helpful tips auto-cycling progress indicator.
• Improved chat bubble layout, spacing, shapes, colors, and image clipping.
• Improved search behavior and search result navigation in the top bar.
• Added model filtering tabs for all, text generation, and image generation.
• Improved model list layout and extracted reusable model item components.
• Added clearer installed, installing, and available model sections.
• Improved delete/loading indicators for installed models.
• Added wrapping behavior for model task tags.

Prompt and session behavior

• Added prompt mode support for switching between text and image workflows.
• Centralized image model selection at the main screen level.
• Improved behavior when switching models mid-session.
• Added read-only behavior for model selection after a chat has started.
• Cleared stale prompt text when switching sessions.
• Fixed stale generating states when changing sessions.
• Improved auto-scroll behavior during streaming and session changes.
• Reset chat scroll state when switching sessions.

Analytics and event handling

• Improved analytics shutdown behavior so pending events are sent before exit.
• Improved analytics error handling.
• Added crash/failure reporting improvements.
• Improved event batching behavior.
• Added tracking for chat deletion events.

WebSocket reliability

• Added message tracking for WebSocket traffic.
• Added acknowledgement handling for sent messages.
• Added in-memory outbox support for pending WebSocket messages.
• Added retry behavior for unacknowledged messages.
• Added timeout and expiry handling for pending messages.
• Added disconnect cleanup behavior.
• Added failure notifications when acknowledgements cannot be validated.
• Added stronger message hash validation.
• Added tests around acknowledgement, retry, expiry, and disconnect flows.

Shared models and DTOs

• Added and expanded shared AI model metadata structures.
• Added registry support to AI model DTOs.
• Added ComfyUI model file metadata.
• Added prompt model persistence support for chat sessions.
• Improved default timestamp handling in shared report/session models.
• Refactored shared package names to the ai.nextgpu namespace.

Namespace and project structure

• Migrated package naming from the previous namespace to ai.nextgpu.
• Reorganized shared files under a common ai.nextgpu package structure.
• Updated imports, tests, resources, and build configuration to match the new namespace.
• Renamed UI package declarations and aligned application imports.

Bug Fixes

• Fixed elapsed time calculations in streaming behavior.
• Fixed chat session deletion when sessions are associated with a project.
• Fixed active chat handling when the selected chat is deleted.
• Fixed starred/pinned chat state issues.
• Fixed assistant thinking bubble visibility.
• Fixed inverted delete-state logic in installed model UI.
• Fixed stale “Generating…” state after session switches.
• Fixed model download UI race conditions after download completion.
• Fixed progress display spikes during downloads.
• Fixed installer desktop icon creation.
• Fixed test failures related to benchmark and AI service behavior.
• Fixed WSL startup timing so setup screens do not trigger credential prompts too early.

Testing and Quality

• Added and updated tests for:
  • Model downloads
  • Model deletion
  • ComfyUI download paths
  • Ollama streaming pulls
  • Benchmark success and failure cases
  • WebSocket acknowledgements and retry behavior
  • Chat deletion behavior
  • AI service expectations
• Improved mocked test configuration for service dependencies.
• Removed or adjusted obsolete tests after service refactors.
• Added broader defensive handling around system-level operations and external command output.