DemoGPT

DemoGPT is an open-source Python library created by Melih Ünsal that lets developers build LLM-powered agents quickly using a comprehensive suite of built-in tools, prompts, frameworks, and model integrations. It is available on GitHub under the MIT License and installable via pip. The project has accumulated over 1,900 stars and 224 forks on GitHub and has been cited in multiple academic papers on autonomous agents.

What It Is

DemoGPT is an agent development framework centered on the demogpt_agenthub package. It provides everything needed to create, customize, and run AI agents: a library of pre-built tools, support for Retrieval Augmented Generation (RAG), multiple vector store backends, and two agent types — ToolCallingAgent and ReactAgent. The original DemoGPT pipeline also auto-generates LangChain code and Streamlit applications from natural language instructions, translating user prompts into interactive web apps through a plan → task → code-snippet → final-code pipeline.

Agent Architecture and Core Workflow

DemoGPT's pipeline follows four stages:

• Planning: Generates a structured plan from the user's instruction.
• Task Creation: Breaks the plan into specific, executable tasks.
• Code Snippet Generation: Converts each task into a code snippet.
• Final Code Assembly: Combines snippets into a runnable Streamlit application.

The ReactAgent adds a visible reasoning loop — at each step it decides whether to call a tool or return a final answer, showing its decision and reasoning before each tool invocation. The ToolCallingAgent provides a simpler interface for direct tool dispatch.

Built-In Tools and RAG Support

DemoGPT AgentHub ships with eleven built-in tools out of the box:

• TavilySearchTool, WikipediaTool, WikiDataTool
• WeatherTool (OpenWeatherMap)
• BashTool, PythonTool
• ArxivTool, PubmedTool
• YouTubeSearchTool, StackOverFlowTool
• RequestUrlTool, YoloTool (object detection)

The BaseRAG class supports Chroma, Pinecone, and FAISS vector stores, accepts PDF, TXT, CSV, and JSON files, and can be embedded directly into agent tool chains. Embedding models can be configured via sentence-transformers or OpenAI models.

Extensibility and Custom Tools

Developers extend DemoGPT by subclassing BaseTool, implementing a run() method, and passing the tool instance to any agent. The framework is designed to be model-agnostic — any LLM meeting code-generation performance criteria can be used, with OpenAIChatModel (wrapping GPT-4o-mini, GPT-3.5-turbo, etc.) provided as the default LLM wrapper.

Academic Recognition and Current Status

DemoGPT has been cited in several peer-reviewed and preprint papers on LLM-based autonomous agents, including a 2023 survey on large language model agents (Wang et al., arXiv:2308.11432) and a 2024 exploration of intelligent agents (Cheng et al., arXiv:2401.03428). The latest GitHub release is v1.2.6 (published September 2023), with the repository last pushed in April 2026. The README roadmap lists planned integrations including Gorilla for autonomous API calls, Llama 2, and a publicly available example database to accelerate generation — several of which remain open to-do items.