Sidekitt

• Your websearch talking buddy with memory

Sidekitt help

Commands

• Chat normally - The assistant will decide if it needs to search

• /exit or quit - Exit the application

• /clear - Clear conversation memory

• /undo -

• /export-

• /import-

• /stats-

• /research -

• /listen - 2-way talking option (experimental)

How It Works

Answers:

• Math and coding questions
• Historical facts and timeless knowledge
• Questions about the conversation itself

Searches:

• Questions about current events, news, or real-time info
• Information that could be outdated (games, products, recent activities)
• Uncertain about accuracy

Quick Start

source .venv/bin/activate
python main.py

---

Credits

• Andrew Kidoo - https://github.com/kustomzone
• Colin Urban - https://github.com/colinurbs

Features

• 🔄 Streaming Responses - See tokens generated in real-time
• 🧠 LLM-Driven Search - Qwen2.5-3B decides when to search the web
• 🌐 Real Web Scraping - Fetches actual page content with BeautifulSoup
• 💾 Persistent Memory - SQLite database remembers conversations
• 📚 Source Citations - Shows URLs when search was used
• 🎨 Rich CLI - Beautiful terminal UI with markdown rendering
• ⚡ Local Execution - No API keys, runs via llama-cpp-python

Dependencies

See requirements.txt:

• llama-cpp-python (local model execution)
• huggingface_hub (model downloads)
• ddgs (web search)
• rich (terminal UI)
• beautifulsoup4 (HTML parsing)
• requests (HTTP requests)

Technical Stack

• Model: Qwen2.5-3B-Instruct (GGUF Q4_K_M, ~2GB)
• Context: 4096 tokens
• Search: DuckDuckGo + requests + BeautifulSoup
• Memory: SQLite (last 10 messages)
• UI: Rich library with live updates

Suggested Future Improvements

1. Configuration file - Customize model, search count, temperature
2. Conversation export - Save/load chat histories to JSON
3. Context summarization - Compress old messages when limit reached
4. Undo command - /undo to remove last exchange
5. Token counter - Display context window usage
6. Multi-query search - Try alternative queries if results are poor
7. Better error handling - Graceful network/model failure recovery
8. Timestamps - Track message timing in database
9. Search history - Remember what searches worked well
10. Configurable streaming - Toggle streaming on/off

---