MAD (Multi-Agents-Debate) is an open-source framework exploring the collective debating capabilities of LLMs. It addresses the 'Degeneration of Thoughts' (DoT) issue—where self-reflection leads to biased or rigid outputs—by implementing a tit-for-tat interaction between multiple agents. By assigning roles like 'devil' and 'angel' to provide mutual external feedback, MAD corrects distorted thinking and breaks cognitive rigidity. It achieves significant performance gains in Counterintuitive QA and Commonsense Machine Translation tasks.
AgentScope is a production-ready multi-agent framework designed to simplify the development of LLM-powered applications. It provides high-level abstractions for agent communication, memory management, and tool usage. Rather than using restrictive prompts, AgentScope leverages the inherent reasoning and tool-use capabilities of LLMs. Key features include Real-time Voice, Human-in-the-loop steering, Agentic RL for model tuning, and support for MCP and A2A protocols. It is engineered for scalability, supporting local, serverless, and K8s deployments with integrated OTel monitoring for enterprise-grade observability.
The official repository for the book 'Build a Large Language Model (From Scratch)' by Sebastian Raschka. It provides a comprehensive, step-by-step guide to developing, pretraining, and finetuning a GPT-like LLM from the ground up. By using pure PyTorch instead of high-level LLM libraries, it explains the inner workings of components like attention mechanisms and transformer blocks. The project includes end-to-end pipelines for pretraining on unlabeled data and finetuning for instruction-following, designed to be accessible on standard consumer hardware.
DeepTutor, developed by HKUDS, is an agent-native personalized intelligent tutoring platform. It integrates various AI agent modes like problem-solving, quiz generation, and deep research within a unified chat workspace. Key features include an AI Co-Writer for collaborative content creation, a Book Engine to transform materials into interactive "living books," and a Knowledge Hub for RAG-powered learning, all supported by persistent memory for evolving personalized guidance.
Langflow is a versatile platform for building and deploying AI-powered agents and workflows. It features a visual builder interface for rapid prototyping, combined with built-in API and MCP servers for seamless integration. Supporting major LLMs and vector databases, Langflow allows multi-agent orchestration and conversation management. Developers can customize components with Python and use an interactive playground for testing. It’s designed to be enterprise-ready with observability integrations like LangSmith.