Happy-LLM: giáo trình miễn phí xây dựng LLM từ đầu

This book by Li Bojie presents AI agents through the core formula Agent = LLM + Context + Tools, covering 10 chapters from fundamentals through production deployment. The material spans context engineering, memory systems, MCP protocol, coding agents, evaluation frameworks, model post-training, multi-agent collaboration, and multimodal interaction. What sets it apart is the 95 accompanying experiments—each designed to be run locally, not just read. The book is fully open source under Apache 2.0, with PDF and EPUB downloads available alongside the web version. Community translations cover 13 languages including English, Spanish, Japanese, and Korean. The level assumes familiarity with Python and basic machine learning concepts; some experiments require GPU hardware or Python 3.11+.

WeKnora is a self-hosted LLM knowledge platform written in Go that transforms raw documents into queryable RAG pipelines, autonomous ReAct agents, and self-maintaining wikis. It covers the full ingestion pipeline—parsing 10+ file formats, chunking with revision history, embedding into swappable vector stores, and surfacing answers with citations. The standout differentiator is Wiki Mode: agents auto-generate structured, interlinked Markdown pages with a visual knowledge graph, line-level diffs, and one-click rollback. On the operations side it ships with multi-workspace RBAC (Owner/Admin/Contributor/Viewer), scoped API keys, per-model concurrency governors, Langfuse tracing, and integrations with Feishu, Notion, Yuque, and many IM channels. Fully modular so you can swap LLMs, vector databases, and object storage backends without platform lock-in.

FastGPT is an AI Agent building platform that provides out-of-the-box capabilities for data processing, RAG retrieval, and model calling, all built on large language models. It allows for visual workflow orchestration via its Flow feature to create complex application scenarios. Users can deploy it via Docker or Sealos Cloud for self-hosting, use a cloud service version, or opt for a commercial version with additional features and support. Key functionalities include Agent Skill orchestration, knowledge base management with support for various file types (TXT, MD, HTML, PDF, Docx, PPTX, CSV, XLSX, URL), hybrid retrieval, and debugging tools like full call chain logs. The project is licensed under the FastGPT Open Source License, which permits direct commercial use as a backend service but prohibits SaaS offerings without specific commercial authorization.

NoteGen is a local-first Markdown application focused on separating information capture from organization. It lets users quickly save various inputs—like voice recordings, screenshots, or links—without immediate structuring. Later, selected records can be processed by configurable AI models to generate structured Markdown notes, reports, or articles, all of which remain editable. The tool works directly with standard Markdown files, includes a canvas for visual content such as flowcharts and mind maps, and creates a searchable knowledge base from local material using both semantic and text search. It offers synchronization through services like GitHub, S3, and WebDAV, and is available for Windows, macOS, Linux, Android, and iOS.