AI Assistant Changelog

Table of Contents

• Overview
• Version Comparison: v0.74 vs. v1.0
• What index.html Offers
• Framework and UI Changes
• Configuration Management
• AI Core Enhancements
• API and Networking
• Chat and History Management
• Text Copying Functionality
• Document Integration
• Model Management
• UI/UX Enhancements
• PDF Export
• Dependencies
• Miscellaneous
• Notes

Overview

The upgrade from Silentis AI Assistant v0.74 to v1.0 represents a transformative shift, moving from a Tkinter-based desktop application to a modern, web-driven architecture using Flask and PyQt5. This release introduces a responsive web interface, enhanced AI capabilities, robust API support, and user-focused features like text copying and document integration. The changelog below provides a detailed account of all changes, ensuring transparency and clarity for users and developers.

Version Comparison: v0.74 vs. v1.0

To highlight the evolution, the following table compares key aspects of v0.74 and v1.0:

Feature	v0.74	v1.0
Interface	Tkinter-based desktop GUI	Web-based UI with Flask, PyQt5, and Bootstrap
Backend	Local Python scripts	Flask server with REST API
Chat Interaction	Single-shot responses	Real-time streaming responses
Text Copying	Manual selection and copy	Dedicated copy buttons for messages and code blocks
Chat History	Basic in-memory storage	Persistent JSON storage with search and pagination
Document Support	Not available	Text and PDF uploads with context-aware responses
Model Management	Single model loading	Multiple models with UI selection and downloads
API Support	Not available	Comprehensive REST API with authentication
Theme	Fixed light theme	Light/dark theme toggle
Performance Monitoring	Not available	RAM and GPU usage indicators
Export Options	Basic text export	JSON and PDF exports with formatting

What index.html Offers

The new index.html serves as the cornerstone of Silentis v1.0’s web-based frontend, replacing the Tkinter GUI. Built with HTML, CSS, and JavaScript, it leverages Bootstrap for responsiveness, Highlight.js for code syntax highlighting, and Font Awesome for icons. Key features include:

• Responsive Design: Adapts seamlessly to desktop, tablet, and mobile devices, ensuring accessibility across platforms.
• Dynamic Chat Interface: Displays real-time streaming responses with a typing cursor effect, enhancing user engagement.
• Sidebar Navigation: Includes links to Chat, Chat History, Downloads, Settings, and About, mirroring the desktop app’s functionality in a web context.
• Interactive Elements:
  • Copy Buttons: Each message and code block has a clickable copy icon for quick text extraction.
  • Edit Buttons: User messages can be edited directly in the UI.
  • Theme Toggle: Switches between light and dark modes with a single click.
• Status Indicators: Shows model status, GPU usage, document status, and memory usage, updated dynamically via JavaScript.
• Modals: Displays progress for model loading and downloads, improving user feedback.
• Event-Driven Updates: Uses Server-Sent Events (SSE) to stream AI responses and update the UI without page reloads.
• Accessibility: Includes ARIA labels and keyboard shortcuts (e.g., Ctrl+Enter to send messages) for better usability.
• Asset Integration: Loads favicon, logo, and stylesheets from local or CDN sources, ensuring fast rendering.

The index.html file communicates with the Flask backend via AJAX and SSE, enabling a smooth, app-like experience while maintaining offline capability once models are loaded.

Framework and UI Changes

• Removed: Tkinter-based desktop interface, including modules like ttk, scrolledtext, messagebox, and filedialog, due to their limitations in scalability and cross-platform rendering.
• Added: Flask web framework for a robust backend, serving API routes such as /chat, /config, /models, and more, enabling a client-server architecture.
• Added: PyQt5 with QWebEngineView to render the web-based UI, providing a hybrid desktop-web experience with better performance than Tkinter.
• Added: Web frontend (index.html) with Bootstrap 5.3.2, Highlight.js for code highlighting, and Font Awesome for icons, creating a modern, responsive interface.
• Modified: Application initialization shifted from AIAssistantApp(tk.Tk) to AIAssistantApp(QApplication), with Flask server running in a separate thread for concurrent processing.

Configuration Management

• Modified: ConfigManager class enhanced with new options:
  • model_path: Path to the active GGUF model file.
  • selected_model: ID of the currently selected model.
  • download_folder: Directory for model downloads.
  • api_enabled: Toggle for REST API activation.
  • api_port: Port for API access (default 5000).
  • api_key: Optional key for API authentication.
  • document_content: Stores uploaded document text.
  • current_chat_id: Tracks the active chat session.
  • draft_message: Saves unsent message drafts.
  • theme: Stores light/dark theme preference.
• Added: Persistent chat history storage in chat_history.json, with methods load_chat_history and save_chat_history for robust management.
• Added: supported_models dictionary, detailing models like:
  • Reasoner v1 (8GB RAM, Apache 2.0).
  • Llama 3 8B Instruct (8GB RAM, LLaMA License).
  • DeepSeek-R1-Distill-Qwen-7B (8GB RAM, MIT).
  • Phi-3 Mini Instruct (4GB RAM, MIT).
• Added: Automatic backup of config.json before updates to prevent data loss.

AI Core Enhancements

• Modified: AICore class now supports multiple models, each with tailored system prompts (e.g., reasoning-focused for Reasoner v1, concise for Phi-3 Mini).
• Added: detect_model method to identify models by filename patterns (e.g., “llama”, “phi-3”).
• Added: check_system_requirements using psutil to ensure sufficient RAM for selected models.
• Added: update_system_prompt to dynamically adjust prompts based on model type and document context, improving response relevance.
• Added: generate_response_stream for real-time response streaming via Server-Sent Events, reducing latency.
• Added: is_document_relevant to check if uploaded documents apply to user queries, enhancing context-aware responses.
• Modified: load_model now supports Windows DLL loading in PyInstaller bundles and sets a testing context size (n_ctx=512).
• Added: unload_model method to free memory when switching models, improving resource efficiency.
• Added: Temperature and max token settings adjustable via the UI, allowing users to control response creativity and length.

API and Networking

• Added: Flask-based REST API with endpoints:
  • POST /chat: Streams AI responses for queries.
  • GET/POST/DELETE /chat_history: Manages chat history.
  • POST /edit_message: Edits specific messages.
  • POST /export_chat, POST /import_chat: Handles JSON chat exports/imports.
  • POST /load_model: Loads a specified model.
  • GET/POST /config: Retrieves or updates configuration.
  • GET /models: Lists supported models.
  • POST /download_model: Initiates model downloads with progress updates.
  • POST /upload_document: Uploads text/PDF files.
  • POST /export_pdf: Generates PDF chat exports.
• Added: require_api_key decorator for endpoint security, requiring a valid API key when enabled.
• Added: Configurable API settings in the UI (api_enabled, api_port, api_key), with validation for port ranges (1024–65535).
• Added: Error handling for network issues, with JSON error responses (e.g., 401 for invalid API keys).

Chat and History Management

• Added: Persistent chat history in chat_history.json, storing chat IDs, titles, messages, and timestamps.
• Added: Chat history UI with:
  • Search bar for filtering by title or content.
  • Pagination for navigating large histories.
  • Options to load, edit, or delete chats.
• Added: Message editing via /edit_message endpoint and UI buttons, preserving conversation context.
• Modified: Chat interactions now stream responses in real-time, replacing v0.74’s single-shot generation for a more interactive experience.
• Added: Draft message support, saved as draft_message in config.json, auto-restored on chat reload.
• Added: Auto-titling of chats based on the first user message, editable via the UI.

Text Copying Functionality

A highly requested feature, text copying in v1.0 is seamless and user-friendly, addressing v0.74’s reliance on manual selection.

• Added: Copy Message Button:
  • Each chat message (user and AI) includes a copy icon (Font Awesome clipboard).
  • Clicking copies the entire message text to the clipboard using the navigator.clipboard API.
  • Visual feedback (e.g., icon change or tooltip) confirms the copy action.
• Added: Copy Code Block Button:
  • Code blocks in AI responses (highlighted by Highlight.js) have a dedicated copy button.
  • Copies only the code, excluding markdown formatting (e.g., ```).
  • Supports multiple code blocks per message, each with its own button.
• Added: Accessibility considerations:
  • Copy buttons include ARIA labels (e.g., aria-label="Copy message").
  • Keyboard support allows tabbing to buttons and activating with Enter.
• Removed: Dependency on manual text selection and Ctrl+C, which was error-prone for long messages or code.
• Added: Error handling for clipboard access, displaying a fallback message if permissions are denied.

Document Integration

• Added: Support for uploading .txt and .pdf files using PyPDF2, with content stored in document_content.
• Added: Context-aware responses incorporating document content via update_system_prompt and is_document_relevant.
• Added: UI indicators showing “Document Loaded” in the sidebar when a document is active.
• Added: Document size limit of 10,000 characters to optimize performance, with truncation warnings in the UI.
• Added: Clear document option to remove uploaded content and reset context.

Model Management

• Added: UI-based model selection, listing models with RAM requirements, licenses, and descriptions.
• Added: Model download system using requests, with streaming downloads and progress bars in the UI.
• Modified: Model path configuration moved to a dynamic UI setting, replacing hardcoded paths.
• Added: Model validation to check GGUF file integrity before loading, reducing crashes.
• Added: Support for model-specific parameters (e.g., temperature, max tokens) saved per model in config.json.

UI/UX Enhancements

• Added: Light/dark theme toggle, stored in theme config, with CSS transitions for smooth switching.
• Added: Real-time chat UI with typing cursor animation and Highlight.js for code block rendering.
• Added: Sidebar status indicators for:
  • Model name and status (e.g., “Loaded: Llama 3”).
  • GPU usage (e.g., “Active” or “Inactive”).
  • Document status (e.g., “Loaded” or “None”).
  • Memory usage via psutil (e.g., “6.2/8 GB”).
• Added: Progress modals for model loading and downloads, with percentage updates.
• Added: Keyboard shortcuts:
  • Ctrl+Enter: Send message.
  • Ctrl+N: New chat.
  • Ctrl+L: Clear chat.
• Added: Character counter for chat input, capping at 4096 characters with a live display.
• Removed: Tkinter-specific elements like rounded input canvas, custom sidebar buttons, and static logo display.
• Added: Error container in the UI for displaying issues (e.g., “Model failed to load”), with dismissible alerts.

PDF Export

• Modified: Switched from fpdf to reportlab for PDF generation, supporting richer formatting and styles.
• Added: PDF exports now include message roles (User/AI), content, and timestamps, accessible via /export_pdf.
• Added: Customizable PDF metadata (e.g., title, author) set via the UI.

Dependencies

• Added: New dependencies:
  • flask: Powers the web server and API.
  • psutil: Monitors system resources.
  • requests: Handles model downloads.
  • PyQt5: Renders the web view.
  • PyPDF2: Processes PDF uploads.
  • reportlab: Generates PDFs.
• Removed: Tkinter-related dependencies:
  • tkinter: Native GUI framework.
  • PIL: Image processing for Tkinter.
  • fpdf: Old PDF generation library.
• Added: CDN dependencies for Bootstrap, Highlight.js, and Font Awesome in index.html.

Miscellaneous

• Added: PyQtBridge class for native file/folder dialogs and external link handling in PyQt5.
• Added: Enhanced error handling, with UI alerts and detailed logs saved to error.log.
• Added: Memory usage monitoring via psutil, displayed in the sidebar for transparency.
• Modified: Version incremented from 0.74 to 1.0, reflecting the major rewrite.
• Modified: resource_path function now supports web assets (e.g., favicon, logo) in PyInstaller bundles.
• Added: Startup checks for configuration directory existence, creating it if missing.
• Added: Logging for API requests, stored in api.log for debugging.

Notes

- Version 1.0 emphasizes modularity, scalability, and user experience, with a web-based interface that supports future extensions like mobile apps or cloud integration. - Tkinter-specific features (e.g., custom window icons, precise widget styling) were replaced with web equivalents, prioritizing cross-platform consistency. - The API-driven design opens possibilities for remote access or third-party integrations, pending user demand. - Text copying and other UX enhancements were prioritized based on user feedback from v0.74. - All changes maintain Silentis’s commitment to offline, privacy-first AI processing.

Last Updated: April 12, 2025
© 2025 Silentis. All rights reserved.