MIT License · 100% local · 100% free
TalhaGPT wraps Qwen3:8B in a full assistant layer: an agent loop with tool calling, persistent memory, ChromaDB-powered RAG, web access, vision and image generation — all running locally through Ollama.
// features
Qwen3:8B is the model. TalhaGPT is everything around it that makes it useful every day.
Runs entirely on your machine via Ollama + Qwen3:8B. No API keys, no subscriptions, no data leaving your computer.
Conversations and notes survive restarts. TalhaGPT remembers what matters across sessions.
ChromaDB + Sentence Transformers index your documents and retrieve them by semantic similarity.
An agent loop lets the model pick tools, read results and keep reasoning toward a final answer.
Searches the internet and fetches full web pages when local knowledge isn't enough.
Captures your screen and understands what's on it — ask questions about anything visible.
Generate images directly from the chat without leaving your terminal.
Optional text-to-speech so the assistant can answer out loud.
Automated tests and GitHub Actions keep every commit honest.
// architecture
Instead of forwarding every prompt to the model, TalhaGPT lets the model decide when a tool is useful, read the result, and keep reasoning until it reaches a final answer.
Long-term information lives in a local vector store and is retrieved through semantic similarity — not just whatever fits in the current context window.
Index any document with add_document_to_memory and query it later with natural language via search_memory.
// tools
Each tool is registered with the agent. The model picks the right one at the right time — you just ask.
| tool | what it does |
|---|---|
| get_weather | Weather information |
| get_system_status | CPU, RAM and system resources |
| save_note | Persistent notes |
| launch_app | Launch allowed applications |
| add_document_to_memory | Index documents into RAG |
| search_memory | Semantic memory search |
| generate_image | Image generation |
| web_search | Internet search |
| fetch_web_page | Retrieve web page contents |
| capture_screen | Screen capture |
// stack
// install
Windows, macOS and Linux are all supported. Install Python 3.12+ and Ollama, then:
# 0. Install Python 3.12+ and Ollama first
$ git clone https://github.com/dinguk0624/TalhaGPT.git # clone the repo
$ cd TalhaGPT
$ py -m pip install -r requirements.txt # dependencies
$ ollama pull qwen3:8b # download the model
$ py main.py # run it
Requirements: Python 3.12+, Ollama, ~8 GB free RAM for Qwen3:8B
// faq
Yes. TalhaGPT is MIT-licensed open source. The model (Qwen3:8B) runs locally through Ollama, so there are no API costs at all.
Qwen3:8B needs roughly 6–8 GB of free RAM (more is better). A GPU helps speed things up, but it also runs on CPU.
No — Qwen3:8B is the model. TalhaGPT is the assistant layer around it: the agent loop, memory, RAG and tools that turn a raw LLM into a useful daily assistant.
Your conversations and memory stay local. The only outbound requests happen when you explicitly use web search or page fetching tools.
// open source
TalhaGPT is MIT-licensed. Star the repo, open an issue, or send a pull request.
★ Star on GitHub