%% Offline Developer AI Stack – Full Local First Ecosystem (No Cloud Required)
graph TD
  %% ==== Root ====
  Root[🖥️ Offline Developer AI Stack]:::entry

  %% ==== LLM Inference Engines ====
  Root --> Inference[🧠 LLM Inference Engines]:::section

  Inference --> LMStudio[LM Studio]:::tool
  Inference --> Ollama[Ollama]:::tool
  Inference --> GGML[GGML / llama.cpp]:::tool
  Inference --> vLLM[vLLM (for local GPU)]:::tool

  %% ==== Local Embedding Models ====
  Root --> Embedding[🔡 Embedding Engines]:::section
  Embedding --> BGE[BGE-M3 (local)]:::model
  Embedding --> MiniLM[MiniLM (int8)]:::model
  Embedding --> Gemma[Gemma 2B / 3n .task]:::model
  Embedding --> TFLite[TFLite converted models]:::model

  %% ==== Auto Indexing Pipeline ====
  Root --> Indexing[📚 Auto Repo Indexing]:::section
  Indexing --> Watcher[🕵️ File Watcher (inotify/fswatch)]:::tool
  Indexing --> EmbedPipe[⚙️ Embed + Store Pipeline]:::tool
  EmbedPipe --> LanceDB[LanceDB]:::vectordb
  EmbedPipe --> FAISS[FAISS (CPU)]:::vectordb

  %% ==== RAG Layer ====
  Root --> RAG[RAG Context Assembly]:::section
  RAG --> Chunking[📄 Smart Chunker (Markdown-aware)]:::tool
  RAG --> Retriever[🔍 Retriever: Top-k + rerank]:::tool
  RAG --> ContextBuilder[📎 Prompt Context Builder]:::tool
  RAG --> LocalChat[🗨️ Local Chat App]:::output

  %% ==== Frontend Interfaces ====
  Root --> UI[🧑‍💻 Dev Interfaces]:::section
  UI --> TUI[Terminal UI (eg. TerminalGPT)]:::tool
  UI --> WebUI[Web GUI (eg. LM Studio, Chatbot-UI)]:::tool
  UI --> VSCode[VSCode Plugin/CLI]:::tool

  %% ==== Use Cases ====
  Root --> UseCases[🚀 Use Cases]:::section
  UseCases --> PRSummary[📝 Auto PR Summary Generator]:::usecase
  UseCases --> DocSearch[🔍 Project Documentation QA]:::usecase
  UseCases --> Refactor[🔧 On-Device Code Refactoring]:::usecase
  UseCases --> QueryAnswer[💡 Prompt-aware Local Assistant]:::usecase
  UseCases --> DiffBot[🧾 Git Commit Diff Summarizer]:::usecase

  %% === File Structure Conventions ===
  Root --> FS[🗂️ File & Metadata Structure]:::section
  FS --> repos[📁 /repos/project-name]:::meta
  FS --> indexdb[📁 /embeddings/lancedb]:::meta
  FS --> config[📄 config.yaml (model paths, chunk size, repo names)]:::meta
  FS --> logs[📄 logs/*.jsonl (chat + indexing audit)]:::meta

  %% === Hardware Profile ===
  Root --> Devices[💻 Target Hardware]:::section
  Devices --> MacM4[MacBook Pro M4 Max]:::device
  Devices --> RyzenMiniPC[AMD Mini PC (32+ GB RAM)]:::device
  Devices --> Jetson[Jetson Xavier NX (Quantized)]:::device

  %% Styling
  classDef entry fill:#fff,stroke:#000,stroke-width:2px;
  classDef section fill:#eef,stroke:#88a,stroke-width:2px;
  classDef tool fill:#ccf,stroke:#55a,stroke-width:1px;
  classDef model fill:#cff,stroke:#399,stroke-width:1px;
  classDef vectordb fill:#fdf,stroke:#a3a,stroke-width:1px;
  classDef output fill:#efe,stroke:#3a3,stroke-width:1px;
  classDef usecase fill:#fcf,stroke:#a3a,stroke-width:1px;
  classDef meta fill:#ffe,stroke:#cc0,stroke-width:1px;
  classDef device fill:#ddf,stroke:#55f,stroke-width:1px;