💻 LuxurAI Coder — Autonomous Code Synthesis Studio

LuxurAI Coder is a full-featured, AI-powered engineering studio built directly into the platform. It is designed for developers who need to generate, refactor, debug, and architect production-grade code — across multiple languages and frameworks — using natural language prompts.

Unlike a simple chat-based coding assistant, Coder operates with a persistent multi-file project workspace, a built-in terminal, live syntax highlighting, and an intelligent web-grounded research layer.

---

🏗️ Studio Layout

The Coder interface is a three-panel engineering environment:

Panel	Purpose
Left Sidebar	File Explorer — lists all generated project files. Click any file to open it in the editor.
Center Editor	Monaco-powered code editor with full syntax highlighting and line numbers.
Right Panel	AI stream output — shows live thought logs and prose explanations as the AI synthesizes code.

A collapsible terminal at the bottom streams live output and web research logs.

---

🎛️ Generation Modes

The input bar exposes three synthesis mode pills that control how the AI approaches a task:

Mode	Behavior
Synthesis (default)	Full code generation from a natural-language description. Builds entire files and project structures from scratch.
Refactor	Takes existing code in context and surgically rewrites or optimizes it according to the prompt.
Explain	Analyzes the code in context and produces detailed inline commentary and architectural breakdowns.

---

🌐 DuckDuckGo Web Search Grounding

The DDG Search toggle (enabled by default) allows the AI to perform live internet searches before generating code:

• When enabled, the system sends a structured search query to DuckDuckGo and fetches the top live results.
• The results are displayed in a collapsible Live Grounding badge directly in the output panel, listing source URLs and snippet previews.
• The gathered context is silently injected into the generation prompt, ensuring the AI produces current, accurate code (e.g. using the latest library APIs or fresh documentation).
• Toggle off for purely offline, context-only generation.

---

📁 Multi-File Project Workspace

Coder maintains a persistent file tree for the current session:

• Initial file: main.py (default). The AI can generate additional files automatically based on prompt complexity.
• File Explorer: All generated files appear in the left sidebar. Clicking a filename loads it into the editor.
• Active File Context: The currently open file's content is automatically passed as context to the next generation, allowing iterative editing and patching.
• ZIP Export: The full multi-file project can be exported as a .zip archive directly from the header toolbar.

---

⚡ Quick-Start Presets

Four one-click preset prompts are available to accelerate common scaffolding tasks:

Preset	What It Builds
FastAPI + PostgreSQL	Production microservice with asyncpg, SQLAlchemy 2.0, Pydantic v2 schemas, JWT auth, and health endpoints
React + Tailwind	Full-stack TypeScript dashboard with responsive grid and analytics components
Multi-Agent Swarm	Autonomous multi-agent framework with Leader Planner, Specialist Coder, and AST Validator
WebSocket Server	High-concurrency async Pub/Sub server with heartbeat ping/pong and recovery

---

🔌 API Integration

Submit a Coder Generation Request

Stream code generation via SSE.

• Endpoint: POST /api/stream/coder
• Headers:

  Content-Type: application/json
  Authorization: Bearer <your_session_token>

• Request Body:

  {
    "prompt": "Create a FastAPI endpoint with JWT auth and role-based access control",
    "context": "",
    "language": "python",
    "mode": "synthesis",
    "reference_repo": "auto",
    "enable_web_search": true,
    "files": {},
    "history": [],
    "session_id": "cdr_abc123xyz"
  }

Request Fields

Field	Type	Description
prompt	string	Natural-language instruction for the AI
context	string	Current file content to pass as active editing context
language	string	Target language — python, typescript, javascript, rust, sql, etc.
mode	string	Generation mode: synthesis, refactor, or explain
reference_repo	string	Reference repository hint for the AI (auto lets the backend decide)
enable_web_search	boolean	If true, performs a live DDG search before generating
files	object	Current file tree map { "filename.py": "...content..." }
history	array	Prior conversation turns for multi-turn context
session_id	string	Client-generated session identifier (prefix cdr_)

SSE Event Stream

The endpoint returns a text/event-stream. Events follow the same SSE protocol as the main chat:

data: {"type": "routing", "text": "Engaged Base Model · Code Synthesis Engine"}

data: {"type": "step", "step": "Searching DuckDuckGo: 'FastAPI JWT role-based access 2025'"}

data: {"type": "web_search", "query": "FastAPI JWT...", "results": [...]}

data: {"type": "token", "text": "from fastapi import"}

data: {"type": "token", "text": " Depends, HTTPException"}

data: [DONE]

---

🗂️ Supported Languages

Selectable from the language dropdown in the input bar:

• Python
• TypeScript / JavaScript
• React (JSX/TSX)
• Rust
• Go
• Java
• PostgreSQL / SQLX

---

⭐ Feedback & Rating

Each generated response includes thumbs-up / thumbs-down buttons. Votes are submitted to POST /api/feedback/rate to help improve external model routing and prompt quality over time.

TIP

Multi-turn Iterations After the first generation, subsequent prompts automatically carry the full conversation history and the current files state. This lets you iteratively build on the same codebase — asking the AI to "add error handling to the existing route" or "refactor the authentication module" without losing prior context.