🤖 Gayatri AI

A powerful, privacy-first desktop AI assistant

---

🎓 Built Under the DBert Internship Program

This project was developed as part of the DBert Internship Program — a hands-on initiative designed to equip students and developers with real-world AI and software development skills.

💡 Want to build real-world AI projects like this one? Join the DBert Internship Program →

---

✨ Features

• 🤖 Autonomous Agent Mode — Break down complex, high-level goals into multi-step action plans, dispatch specialized tools (web_search, read_context, memory_lookup, reason), track progress live, and persist task states in SQLite
• 🔬 Autonomous Deep Research Mode — Multi-hop inquiry with iterative query decomposition, evidence gap analysis, source verification, and auto-export to structured Markdown dossiers
• 🧬 Persistent Long-Term Memory (Second Brain) — Background extraction of user preferences, constraints, and facts with confidence scoring and an interactive Memory Vault UI modal
• 🧠 Local LLM Integration — 100% offline, privacy-first AI conversations with dynamic model discovery and streaming
• 🌐 Web Search Grounding — Live search with hybrid relevance scoring (Semantic Embeddings via all-MiniLM-L6-v2 + TF-IDF)
• 🔗 URL Crawling & Deep Scrape — Concurrently crawls and extracts content from pasted URLs and internal links
• 📄 Markdown-Driven RAG & Context Files — Workspace-based context system (workspace/) with guardrail documents (system_rules.md, project_context.md, brand_voice.md)
• 💬 Multi-Session Chat History — Persistent chat sessions stored in a local SQLite database with renaming, search, and deletion
• 📱 Responsive Desktop Layout — Collapsible sidebar, stats dock, and adaptive padding for any screen resolution
• ⚡ Live Streaming Responses — Real-time token streaming with multi-tier fallback resilience and step-by-step execution indicators
• 🔍 6 Intelligent Modes — Auto, Agent Mode 🤖, Deep Research 🔬, Web Search, URL Only, or Search Off
• 📊 Session Statistics & Diagnostics — Tracks query keywords, searches performed, and sources discovered; includes built-in connectivity diagnostic suite

---

🛠️ Tech Stack

Layer	Technology	Description
Agent Engine	agent.py	Multi-step task planner, tool dispatcher & SQLite execution tracker
Deep Research	deep_research.py	Multi-hop search planner, gap analyzer & dossier synthesizer
Second Brain	memory_manager.py	Background memory extractor & semantic retrieval engine
Context / RAG	context_manager.py	Sentence-Transformers (all-MiniLM-L6-v2) + TF-IDF chunk ranking
Web Search & Scraping	search_engine.py	DuckDuckGo search + multi-threaded concurrent web crawler
Database Layer	db.py	Local SQLite database for sessions, messages, memories, and agent tasks
Language	Python 3.10+	Clean, modular asynchronous architecture

---

🚀 Getting Started

Prerequisites

-## Installation

# 1. Clone the repository
git clone https://github.com/Gayatri-Education/Gayatri-AI.git
cd Gayatri-AI

# 2. Install dependencies
pip install -r requirements.txt

# 3. Run the app
python main.py

💡 Windows Quick Start: You can also launch Gayatri AI by double-clicking run_gayatri_ai.bat.

📁 Project Structure

Gayatri-AI/
├── main.py              # App entrypoint — UI orchestration and event routing
├── agent.py             # Autonomous Agent Mode (goal decomposition & tool dispatching)
├── config.py            # App-wide configuration, search parameters, and theme tokens
├── ui_components.py     # Reusable Flet widgets, Memory Vault modal, and theme styling
├── llm_client.py        # LM Studio API client with streaming and fallback support
├── search_engine.py     # Web search, URL crawler, and hybrid relevance ranker
├── deep_research.py     # Multi-hop inquiry, gap analysis, and dossier synthesizer
├── memory_manager.py    # Long-term memory extraction & semantic retrieval engine
├── context_manager.py   # Workspace RAG file manager with embedding + ranking
├── db.py                # SQLite database layer for sessions, memories, and agent tasks
├── diagnostic_test.py   # Connectivity and search diagnostic suite
├── run_gayatri_ai.bat   # Windows batch launcher script
├── requirements.txt     # Python package dependencies
├── workspace/           # Markdown context files & active system guardrails
├── exports/             # Exported research dossiers and chat transcripts
└── gayatri.db           # Local SQLite database (auto-created on first run)

---

⚙️ Configuration

Key settings live in config.py and can also be adjusted via the in-app Settings dialog:

Setting	Default	Description
DEFAULT_MODEL	Auto-detected	Default LLM model ID (discovered from /v1/models)
AGENT_MAX_STEPS	6	Maximum steps in an autonomous agent execution plan
AGENT_MAX_TOKENS	2500	Max output token budget for agent result synthesis
MAX_RESEARCH_HOPS	2	Max multi-hop research exploration loops
MAX_SUB_QUERIES	3	Max sub-questions decomposed per research hop
ENABLE_LONG_TERM_MEMORY	True	Persistent memory & Second Brain active
MAX_MEMORY_ITEMS_IN_PROMPT	6	Maximum relevant memories injected per conversation turn
SEARCH_RESULTS_PER_QUERY	6	Number of web results per search
MAX_MD_TOKENS	1200	Max context token budget for workspace RAG files
SIDEBAR_WIDTH	280	Collapsible sidebar width in pixels

---

🔍 Modes & Execution Strategies

Mode	Badge	Description
Auto	Auto	AI automatically decides whether web search or context retrieval is needed
Agent Mode	Agent 🤖	Full autonomous execution: plans multi-step strategy, runs tools (web_search, read_context, memory_lookup, reason), and streams comprehensive solutions
Deep Research	Deep Research 🔬	Multi-hop autonomous inquiry: recursively analyzes evidence gaps, gathers cross-source data, and synthesizes structured research dossiers
Web Search	Web Search	Forces search and source synthesis on every query
URL Only	URL Only	Directly crawls and summarizes specific URLs provided in your prompt
Search Off	Search Off	Pure offline local LLM generation with zero internet access

---

🤝 Contributing

Contributions are welcome! This project is actively developed under the DBert Internship Program. If you're a DBert intern or alumni:

1. Fork the repository
2. Create a new branch (git checkout -b feature/your-feature)
3. Commit your changes (git commit -m 'Add your feature')
4. Push and open a Pull Request

---

📄 License

This project is open-source and developed for educational purposes under the DBert Internship Program.

---

🔗 Links

🌐 DBert Internship Program	https://dbert.online/
📦 Repository	https://github.com/Gayatri-Education/Gayatri-AI

|

Developed with ❤️ under the DBert Internship Program