SDKGen

Turn any API documentation URL into a fully-typed, live-tested SDK.

---

🚀 What is SDKGen?

SDKGen is an agentic, multi-agent LLM system that eradicates the developer integration tax.

Instead of spending 3–7 hours manually reading docs, writing boilerplate HTTP clients, defining type models, and debugging integration errors, a developer simply opens any API documentation page in Chrome and clicks the SDKGen extension.

Behind the scenes, our autonomous engineering team reads the docs, writes the code, tests it against the live API internet, auto-fixes any bugs, and injects the finished, fully-typed SDK directly into the developer's active VS Code workspace.

No copy-paste. No OpenAPI spec required. No manual HTTP wiring.

---

🧠 The Architecture (Hub-and-Spoke)

SDKGen uses a non-linear LangGraph State Machine. Unlike standard linear LLMs (like ChatGPT or Claude) which have no agency, SDKGen operates as a self-healing engineering team.

All communication routes through the Supervisor. No agent talks directly to another.

flowchart LR
    classDef supervisor fill:#00c8ff,stroke:#000,stroke-width:2px,color:#000,font-weight:bold
    classDef worker fill:#1e293b,stroke:#64748b,stroke-width:2px,color:#f8fafc
    classDef terminal fill:#22c55e,stroke:#000,stroke-width:2px,color:#000,font-weight:bold
    
    START((API URL)) --> Sup
    
    Sup{SUPERVISOR<br/>Decision Engine}
    class Sup supervisor

    Res[RESEARCHER<br/>Scrapes Docs]
    Arch[ARCHITECT<br/>Builds Schema]
    Eng[ENGINEER<br/>Writes Code]
    QA[QA TESTER<br/>Live HTTP Tests]
    
    class Res,Arch,Eng,QA worker

    Sup -- "Needs Docs" --> Res
    Res -- "Knowledge Base" --> Sup
    
    Sup -- "Needs Schema" --> Arch
    Arch -- "JSON Schema" --> Sup
    
    Sup -- "Needs SDK" --> Eng
    Eng -- "Source Files" --> Sup
    
    Sup -- "Needs Testing" --> QA
    
    QA -. "Fail: Code Bug" .-> Sup
    QA -. "Fail: Missing Docs" .-> Sup
    linkStyle 8,9 stroke:#ef4444,stroke-width:2px,stroke-dasharray: 5 5,color:#ef4444
    
    QA -- "Pass: 200 OK" --> Sup
    Sup -- "100% Verified" --> Pack[PACKAGER<br/>VS Code Injection]
    class Pack terminal
    
    Pack --> END((Done))

Loading

The 6 Agents:

1. Supervisor: The central brain. Reads state and decides which agent acts next.
2. Researcher: Plans the crawl and uses Playwright to build a knowledge base from HTML.
3. Architect: Converts human prose into strict Pydantic JSON schemas.
4. Engineer: Generates models, client code, and tests. Auto-repairs syntax errors.
5. QA Tester: Executes real HTTP requests against the live API to prove the code works.
6. Packager: Triggers ElevenLabs narration and injects the files into VS Code via WebSockets.

---

🛠️ Tech Stack

• Backend: Python 3.11, FastAPI, Uvicorn
• AI / Agentic: LangGraph, LangChain, Gemini 1.5 Pro, Gemini 1.5 Flash
• Web Scraping: Playwright (async headless Chromium), BeautifulSoup4
• Testing: httpx (async) with built-in safety blocks
• Voice / Audio: ElevenLabs TTS (eleven_flash_v2) & Conversational AI Agents
• Frontend / Bridge: Chrome Extension (Manifest V3), VS Code Extension (TypeScript), Server-Sent Events (SSE), WebSockets.
• Presentation: Gamma 4

---

📂 Project Structure

SDKGen/
├── backend/                  # FastAPI server and LangGraph agents
│   ├── agents/               # Supervisor, Researcher, Architect, Engineer, QA, Packager
│   ├── graph/                # LangGraph StateGraph compilation and Router
│   ├── tools/                # Playwright scraper, HTTP executor, ElevenLabs
│   └── prompts/              # System prompts for all agents
├── chrome-extension/         # Chrome Popup UI and Content Scraper
│   ├── popup/                # Dark terminal UI and SSE streaming
│   └── lib/                  # WebSocket/URI Bridge protocol
└── vscode-extension/         # VS Code integration
    └── src/                  # Webview panel, File Writer, WebSocket server

---

💻 Running the Project Locally

1. Start the Backend

You will need a Gemini API key. ElevenLabs is optional for the voice features.

cd backend
python -m venv venv && source venv/bin/activate
pip install -r requirements.txt

# Create a .env file:
# GEMINI_API_KEY=your_key_here
# ELEVENLABS_API_KEY=optional_key

uvicorn main:app --reload

The backend will run on http://localhost:8000

2. Load the Chrome Extension

1. Open Chrome and navigate to chrome://extensions.
2. Turn on Developer mode (top right).
3. Click Load unpacked and select the chrome-extension/ directory.

3. Load the VS Code Extension

cd vscode-extension
npm install
npx tsc

Open the vscode-extension folder in VS Code and press F5 to launch the Extension Development Host.

---

⚡ Demo Targets

SDKGen was built to handle both modern APIs and the "un-spec'd web":

• Modern Complex: Stripe API (Nested JSON objects, complex pagination).
• Classic Hacks: OpenWeatherMap API (Missing auth headers, large single-page dumps).
• The "Bad Docs": Old XML-based logistics APIs or poorly structured finance documentation pages.

---

Eliminating the integration tax, one API at a time.