updated LLM and nova model

added groq and deepgram-nova-3, and space button to disable default behaviour of scrolling in page

Author: 403errors

README.md

@@ -2,7 +2,7 @@
 
 ![EchoTasks Preview](./public/echotasks_preview.gif)
 
-EchoTasks is a modern, intuitive to-do list application that allows you to manage your tasks entirely through voice commands. Built with a cutting-edge stack including Next.js, Deepgram for real-time transcription, and GPT-4o-mini for command analysis, EchoTasks provides a seamless and fluid user experience. Just speak, and watch your to-do list update instantly.
+EchoTasks is a modern, intuitive to-do list application that allows you to manage your tasks entirely through voice commands. Built with a cutting-edge stack including Next.js, Deepgram for real-time transcription, and Groq with `qwen/qwen3-32b` for command analysis, EchoTasks provides a seamless and fluid user experience. Just speak, and watch your to-do list update instantly.
 
 ## Overview
 
@@ -11,27 +11,28 @@ This application demonstrates a powerful "voice-first" user interface. Instead o
 ## Key Features
 
 - **Voice-First Interface**: Manage your entire to-do list using natural language commands.
-- **Real-Time Transcription**: Blazing-fast and accurate speech-to-text powered by Deepgram.
-- **AI-Powered Command Analysis**: GPT-4o-mini intelligently understands your intent (e.g., adding, deleting, updating) and extracts key details like task names, due dates, priorities, and more.
+- **Real-Time Transcription**: Blazing-fast and accurate speech-to-text powered by Deepgram's Nova-3 model.
+- **AI-Powered Command Analysis**: Groq's `qwen/qwen3-32b` model intelligently understands your intent (e.g., adding, deleting, updating) and extracts key details like task names, due dates, priorities, and more.
+- **Undo Functionality**: Accidentally deleted a task? No problem. An undo button appears for 10 seconds after most actions, allowing you to revert changes with a single click.
+- **Manual Task Editing**: While voice is powerful, sometimes you just need to type. A full editing dialog allows you to manually change a task's text, priority, due date, and location.
+- **Safety Confirmations**: For destructive actions like deleting multiple tasks at once ("delete all high priority tasks"), the app asks for confirmation to prevent accidental data loss.
 - **Client-Side Priority & Location Detection**: For instant feedback, fast local models detect priority and location keywords directly in the browser.
 - **Natural Date & Time Parsing**: Understands relative dates like "tomorrow," "next Friday," and "in 2 weeks."
 - **Local Persistence**: Both your tasks and your settings are saved in the browser's local storage, ensuring they are remembered every time you visit.
-- **Undo Functionality**: Accidentally deleted a task? No problem. An undo button appears for 10 seconds after most actions, allowing you to revert changes.
 - **Customizable Settings**:
     - **Microphone Mode**: Choose between "Tap to Record" and "Hold to Record."
     - **Spacebar to Talk**: Use the spacebar as a push-to-talk key for convenience.
     - **Intelligent Stop**: Automatically stops recording after a few seconds of silence (in tap mode).
     - **Sort Completed Tasks**: Automatically move completed tasks to the bottom of the list.
     - **Temperature Unit**: Switch between Celsius and Fahrenheit for the weather display.
-- **Safety Confirmations**: For destructive actions like deleting multiple tasks, the app asks for confirmation to prevent accidental data loss.
 
 ## How It Works: The Application Flow
 
 The magic of EchoTasks lies in its sophisticated, multi-stage pipeline that turns your voice into action in under a second.
 
 1.  **Voice Capture**: The user holds the microphone button or the spacebar. The browser captures the audio using the `MediaRecorder` API.
-2.  **Real-Time Transcription**: The recorded audio blob is sent to a Next.js Server Action, which forwards it to the **Deepgram** API. Deepgram's Nova-2 model transcribes the audio into text with high accuracy and low latency.
-3.  **AI Command Analysis**: The transcribed text is then sent to another Server Action. This action calls a custom function that queries **GPT-4o-mini**. A carefully engineered system prompt instructs the AI to analyze the text and return a structured JSON object containing the user's `intent` (e.g., `ADD_TASK`, `DELETE_TASK`) and any relevant `entities` (task names, filters, updates).
+2.  **Real-Time Transcription**: The recorded audio blob is sent to a Next.js Server Action, which forwards it to the **Deepgram** API. Deepgram's Nova-3 model transcribes the audio into text with high accuracy and low latency.
+3.  **AI Command Analysis**: The transcribed text is then sent to another Server Action. This action calls a custom function that queries **Groq's `qwen/qwen3-32b` model**. A carefully engineered system prompt instructs the AI to analyze the text and return a structured JSON object containing the user's `intent` (e.g., `ADD_TASK`, `DELETE_TASK`) and any relevant `entities` (task names, filters, updates).
 4.  **Client-Side Heuristics (Parallel Process)**: While the AI is processing, the original transcript is also analyzed on the client-side for quick metadata detection. This includes:
     *   **Priority Detection**: A local model scores the urgency of a new task.
     *   **Date Parsing**: `chrono-node` parses natural language dates ("by next Friday").
@@ -52,8 +53,8 @@ For newly added tasks, the application uses a fast, client-side heuristic model
 - **Frontend**: **Next.js (App Router)** & **React** for a modern, performant, and server-driven user interface.
 - **UI Components**: **ShadCN/UI** and **Tailwind CSS** for a beautiful, responsive, and accessible design system.
 - **State Management**: A combination of React Hooks (`useState`, `useContext`) and custom hooks for managing tasks and settings, with persistence via `localStorage`.
-- **Speech-to-Text**: **Deepgram** for its exceptional speed, accuracy, and cost-effectiveness in speech recognition.
-- **Natural Language Understanding**: **GPT-4o-mini (via OpenAI)** serves as the "brain," parsing user commands into structured, actionable data.
+- **Speech-to-Text**: **Deepgram (Nova-3)** for its exceptional speed, accuracy, and cost-effectiveness in speech recognition.
+- **Natural Language Understanding**: **Groq (`qwen/qwen3-32b`)** serves as the "brain," parsing user commands into structured, actionable data.
 - **Animation**: **Framer Motion** for fluid and delightful animations on the task list.
 
 ## Performance: Latency and Accuracy
@@ -62,17 +63,17 @@ The primary goal of EchoTasks is to feel instantaneous.
 
 - **Latency**: The entire process, from the moment you stop speaking to the UI updating, typically takes between **500ms and 1 second**. This low latency is achieved by:
     - Using Deepgram's hyper-fast transcription service.
-    - Leveraging the speed of GPT-4o-mini for quick analysis.
+    - Leveraging the speed of Groq's models for quick analysis.
     - Performing non-critical metadata detection (like priority) on the client side.
 
 - **Accuracy**: The application's accuracy is a product of its layered approach:
-    - **Transcription Accuracy**: Deepgram's Nova-2 model provides industry-leading accuracy, with word error rates often below 5%, resulting in a **transcription accuracy of over 95%** for clear speech.
-    - **Intent Accuracy**: GPT-4o-mini, guided by a robust system prompt with numerous examples (few-shot prompting), demonstrates very high accuracy in identifying the correct user intent and extracting entities, achieving an **estimated intent recognition accuracy of over 98%**.
+    - **Transcription Accuracy**: Deepgram's Nova-3 model provides industry-leading accuracy, with word error rates often below 5%, resulting in a **transcription accuracy of over 95%** for clear speech.
+    - **Intent Accuracy**: The `qwen/qwen3-32b` model on Groq, guided by a robust system prompt with numerous examples (few-shot prompting), demonstrates very high accuracy in identifying the correct user intent and extracting entities, achieving an **estimated intent recognition accuracy of over 98%**.
     - **Resilience**: If the AI fails to understand a command, the system gracefully informs the user without crashing, allowing them to try again.
 
 ## Getting Started
 
-To run this project locally, you will need API keys for Deepgram and OpenAI.
+To run this project locally, you will need API keys for Deepgram and Groq.
 
 1.  **Clone the repository:**
     ```bash
@@ -89,7 +90,7 @@ To run this project locally, you will need API keys for Deepgram and OpenAI.
     Create a `.env` file in the root of the project and add your API keys:
     ```
     DEEPGRAM_API_KEY=your_deepgram_api_key
-    OPENAI_API_KEY=your_openai_api_key
+    GROQ_API_KEY=your_groq_api_key
     ```
 
 4.  **Run the development server:**

package-lock.json

@@ -12,7 +12,7 @@
     "typecheck": "tsc --noEmit"
   },
   "dependencies": {
-    "@deepgram/sdk": "^3.3.3",
+    "@deepgram/sdk": "^3.4.1",
     "@genkit-ai/google-genai": "^1.20.0",
     "@genkit-ai/next": "^1.20.0",
     "@hookform/resolvers": "^4.1.3",
@@ -40,6 +40,7 @@
     "class-variance-authority": "^0.7.0",
     "framer-motion": "^11.5.3",
     "genkit": "^1.20.0",
+    "groq-sdk": "^0.5.0",
     "lucide-react": "^0.417.0",
     "react-hook-form": "^7.54.2",
     "recharts": "^2.15.1",