DemoGPT is an open-source project by Melih Ünsal hosted at github.com/melih-unsal/DemoGPT. It is structured around two core capabilities: DemoGPT AgentHub, a modular framework for constructing AI agents, and DemoGPT Core, an automatic Streamlit app generator that converts plain-English instructions into runnable Python applications. DemoGPT AgentHub provides two agent types: ToolCallingAgent, which makes a single tool call per query for fast and predictable responses, and ReactAgent, which chains multiple tool calls with intermediate reasoning steps for complex multi-step tasks. Both types can be extended with custom tools by subclassing BaseTool and implementing a run() method. The framework ships with 12+ built-in tools including Tavily web search, Wikipedia, arXiv lookup, Python and Bash code execution, weather querying, and YOLO object detection. RAG (Retrieval-Augmented Generation) pipelines are supported out of the box with Chroma, FAISS, and Pinecone vector stores, and ingestion of PDF, TXT, CSV, JSON, and URL sources. DemoGPT Core uses GPT-3.5-turbo (or compatible LLMs) to auto-generate LangChain-based Python code and Streamlit interfaces from natural language app descriptions. The generated apps are interactive and immediately runnable. The project is licensed under MIT and installable via pip. It requires a Python 3.8+ environment, an OpenAI API key, and optionally a Tavily API key for web search functionality. The project gained attention in 2023 as a rapid prototyping tool and has been cited in academic research through 2024. Its latest tracked release is v1.2.6, which added bug fixes for document loader and finalizer modules and preliminary Azure OpenAI URL support. Activity on the repository appears to have slowed in 2025-2026 relative to its initial burst of popularity. Key features: - ToolCallingAgent for single-step tool invocations and ReactAgent for multi-step reasoning chains - 12+ built-in tools: Tavily search, Wikipedia, arXiv, Python/Bash execution, weather, YOLO object detection - RAG pipelines with Chroma, FAISS, and Pinecone; ingests PDF, TXT, CSV, JSON, and URLs - Auto-generates complete Streamlit apps from plain-English natural language descriptions - LangChain-based code generation powered by GPT-3.5-turbo or compatible models - Extensible tool system via BaseTool subclassing - MIT-licensed, installable via pip, requires Python 3.8+
Free and open-source (MIT). Requires own OpenAI API key (billed by OpenAI) and optionally a Tavily API key.