🐝 Bee

The open-source AI knowledge hive.

Learn AI. Build AI. Share AI.

A community-driven collection of modern AI engineering resources — theory, runnable examples, production patterns, and learning paths. Think MDN for AI Engineering: freeCodeCamp + Awesome-AI + the OpenAI Cookbook, in one repository.

---

Note

Bee is not an AI application. It is a knowledge base — a curated, beautifully organized library of everything a developer needs to learn and build with AI.

📖 Table of Contents

• Why Bee?
• What's Inside
• Quick Start
• Repository Structure
• Learning Paths
• Featured Examples
• How Content Is Organized
• Contributing
• Roadmap
• Community & Support
• FAQ
• License
• Acknowledgements

🍯 Why Bee?

AI engineering knowledge is scattered across blog posts that rot, framework docs that assume you already know the concepts, and tutorials that don't run. Bee gathers the good stuff in one place and holds it to a single quality bar:

• 🧠 Concepts explained from first principles — you'll understand why, not just how.
• 🏃 Every code example runs — each lives in its own folder with dependencies and a README.
• 🗺️ Structured learning paths — go from "never called an LLM" to "shipping agents."
• 🏭 Production patterns, not toys — evaluation, security, cost, observability, deployment.
• 🎨 Diagrams everywhere — Mermaid diagrams make architectures click.
• 🌍 Community-owned — designed for hundreds of contributors from day one.

🧰 What's Inside

🧠 Foundations How LLMs work Transformers & attention Tokenization Embeddings Context windows	✍️ Building Blocks Prompt engineering System prompts Structured outputs / JSON mode Function & tool calling Streaming	🔎 Retrieval (RAG) Chunking strategies Vector databases Hybrid search Reranking RAG evaluation
🤖 Agents Agent fundamentals Planning & reflection Memory Multi-agent systems MCP · LangGraph · CrewAI	🖼️ Multimodal Vision models & OCR Speech recognition Text-to-speech Image generation	🏭 Production Evaluation & guardrails Security & safety Deployment (Docker/K8s) Observability Inference optimization

🚀 Quick Start

Bee is a reading-and-doing repository. There are three ways to use it:

1. Browse on GitHub — every folder has a README.md. Start with docs/.

2. Read it as a website (nicer search, dark mode, diagrams):

git clone https://github.com/bee-ai-labs/bee.git
cd bee
pip install -r docs/requirements.txt
mkdocs serve          # open http://127.0.0.1:8000

3. Run an example — each example is self-contained:

cd examples/01-chatbot
cp .env.example .env          # add your API key
uv sync                       # or: pip install -e .
python -m app                 # run it

Tip

New to AI engineering? Don't read randomly — follow a Learning Path. It sequences the content so each concept builds on the last.

🗂️ Repository Structure

bee/
├── docs/               📚 The knowledge base (also the MkDocs website)
│   ├── concepts/          🟢 Foundations: LLMs, transformers, tokenization, embeddings
│   ├── prompting/         🟡 Prompting, system prompts, structured output, tool calling
│   ├── rag/               🟡 Chunking, vector DBs, hybrid search, reranking, eval
│   ├── agents/            🔴 Single/multi-agent, planning, memory, MCP, LangGraph, CrewAI
│   ├── vision/ speech/    🖼️ Multimodal: vision, OCR, speech-to-text, TTS
│   ├── evaluation/        📏 Evals, benchmarks, hallucination, guardrails
│   ├── security/          🛡️ Prompt injection, safety, auth, rate limiting
│   ├── deployment/        🚢 Docker, Kubernetes, serving, scaling
│   ├── mlops/             ⚙️ CI/CD, observability, monitoring, cost
│   └── learning-paths/    🗺️ Curated sequences for each skill level
├── examples/           🏃 Self-contained, runnable Python projects
├── architectures/      🏗️ Reference architectures + decision records (Mermaid)
├── templates/          📦 Copy-paste starters (FastAPI+LLM, RAG service, agent loop)
├── guides/             🧭 Task-oriented how-tos ("how do I stream responses?")
├── tutorials/          📝 End-to-end walkthroughs (build X from scratch)
├── datasets/ benchmarks/  📊 Curated data pointers + evaluation harnesses
├── assets/             🎨 Banner, logo, brand guide, diagram sources
├── scripts/            🛠️ Repo tooling (new-example scaffolder, link checks)
└── .github/            🤖 Issue/PR templates, CI workflows, automation

🗺️ Learning Paths

Instead of a pile of articles, Bee sequences content into paths. Pick where you are:

Path	For	Starts with
🟢 AI Engineer — Fundamentals	You can code, but LLMs are new	How LLMs Work
🟡 Building with LLMs	You've made API calls, want to build real features	Prompt Engineering
🟡 RAG Specialist	You need models to use your data	RAG Overview
🔴 Agent Engineer	You're building autonomous, tool-using systems	Agent Fundamentals
🔴 Production & MLOps	You need to ship, monitor, and scale	Deployment

➡️ See all learning paths →

⭐ Featured Examples

Every example is a complete, runnable project with its own README, dependencies, and tests.

Example	What it teaches	Level
01-chatbot	Streaming chat, conversation state, cost tracking	🟢
02-rag-document-qa	Chunking, embeddings, retrieval, cited answers	🟡
03-research-agent	Tool use, planning, multi-step reasoning	🔴

➡️ Browse all examples →

🎯 How Content Is Organized

Every section of the knowledge base follows the same structure, so you always know what to expect:

Overview → Learning Objectives → Theory → Practical Examples → Code Snippets → Diagrams → Best Practices → Common Mistakes → Exercises → References

Content is tagged by difficulty so you can meet yourself where you are:

• 🟢 Beginner — assumes general programming knowledge only
• 🟡 Intermediate — assumes you've built basic LLM features
• 🔴 Advanced — assumes production experience

🤝 Contributing

Bee is built by its community — and that means you. Whether you fix a typo, add a diagram, or write an entire tutorial, you're welcome here.

1. Read the Contributing Guide (5-minute read).
2. Find a good first issue or a [WANTED] topic in any section README.
3. Copy the relevant _TEMPLATE/ and fill it in.
4. Open a PR — our CI checks formatting, spelling, and links so you don't have to.

By participating you agree to our Code of Conduct.

📍 Roadmap

Bee grows in public. See the full Roadmap and Changelog.

• M0 — Foundation: structure, docs site, community & CI
• M1 — Flagship knowledge: Concepts, Prompting, RAG, Agents (deep)
• M2 — Runnable example fleet
• M3 — Breadth: vision, speech, eval, security, deployment
• M4 — Interactive notebooks & video companions

💬 Community & Support

• 💡 Questions & ideas → GitHub Discussions
• 🐛 Bugs & content errors → Open an issue
• 🛟 Need help? → SUPPORT.md
• 🔒 Security concern? → SECURITY.md

❓ FAQ

Is Bee affiliated with any AI company?

No. Bee is an independent, community-driven, open-source project. It uses vendor SDKs in examples but the concepts are vendor-neutral.

Which LLM provider do the examples use?

Concepts are provider-agnostic. Code examples default to the Anthropic SDK for concreteness, with notes on adapting to OpenAI, Google, Ollama, and others. You can run most examples against any provider.

Do I need a GPU?

No. Almost everything uses hosted APIs. The few local-model sections clearly say so and offer CPU-friendly alternatives.

Can I use Bee content in my own course/blog/book?

Yes — docs are CC-BY-4.0 (give credit) and code is MIT. See LICENSE.

How do I keep code examples from breaking as APIs change?

Each example pins its dependencies and runs in CI. See CONTRIBUTING.md for the maintenance policy.

📜 License

Bee is dual-licensed: documentation and prose under CC-BY-4.0, and all source code under the MIT License. In short — reuse freely, just give credit. 🐝

🙏 Acknowledgements

Bee stands on the shoulders of the open-source AI community and is inspired by the clarity of MDN Web Docs, the practicality of the OpenAI Cookbook, the breadth of the Awesome lists, and the welcoming ethos of freeCodeCamp.

Built with 🍯 by the Bee community.

If Bee helps you, consider giving it a ⭐ — it helps other developers find the hive.