feat: migrate to new google-genai SDK with Gemini 3 support

.cursor/rules/backend-rag-conventions.mdc

@@ -0,0 +1,31 @@
+---
+description: Rules for Python backend development, Adalflow RAG pipeline, and API conventions.
+globs:
+  - "api/**/*.py"
+alwaysApply: false
+---
+
+# Backend & RAG Conventions
+
+This project uses FastAPI and the `adalflow` framework to build a robust RAG pipeline.
+
+## Python & FastAPI Patterns
+- **Pydantic Models**: Use Pydantic for all request/response schemas. Define them in `api/api.py` or `api/websocket_wiki.py`.
+- **Async Handlers**: Use `async def` for API endpoints and WebSocket handlers to maintain high concurrency.
+- **Error Handling**: Use `fastapi.HTTPException` for REST errors. For WebSockets, send a descriptive text message before closing the connection.
+
+## Adalflow & RAG Logic
+- **Component Pattern**: The `RAG` class in `api/rag.py` must inherit from `adal.Component`.
+- **Memory**: Use the `Memory` class (from `adalflow`) for conversation history management.
+- **Data Pipeline**: Use `api/data_pipeline.py` for repository cloning and document parsing. Respect the exclusion filters defined in `repo.json`.
+- **Embedders**: Always initialize embedders via `get_embedder` in `api/tools/embedder.py`.
+- **Prompts**: Centralize all LLM prompts in `api/prompts.py`. Use templates for dynamic content.
+
+## Vector Database (FAISS)
+- Vector indices are managed by `DatabaseManager` in `api/data_pipeline.py`.
+- Ensure embedding dimensions are validated before insertion (see `_validate_and_filter_embeddings` in `api/rag.py`).
+
+## Anti-Patterns
+- Do not hardcode API keys; use environment variables via `api/config.py`.
+- Avoid blocking synchronous calls in async routes (use `run_in_threadpool` if necessary).
+- Do not bypass the `DatabaseManager` for file system operations in `~/.adalflow`.

.cursor/rules/frontend-conventions.mdc

@@ -0,0 +1,33 @@
+---
+description: Rules for Next.js frontend, TypeScript types, and UI component development.
+globs:
+  - "src/**/*.{ts,tsx}"
+alwaysApply: false
+---
+
+# Frontend & UI Conventions
+
+The frontend is a Next.js application focused on repository visualization and AI interaction.
+
+## Next.js & React Patterns
+- **App Router**: Use the `src/app` directory for routing. Dynamic repository routes follow the `[owner]/[repo]` pattern.
+- **Server vs. Client**: Use `"use client"` only for components requiring state, effects, or browser APIs (e.g., `Ask.tsx`, `WikiTreeView.tsx`).
+- **TypeScript**: Strictly type all component props and API responses. Sync with backend Pydantic models.
+
+## AI & Chat UI
+- **WebSockets**: Use `src/utils/websocketClient.ts` for all chat interactions. Do not implement raw WebSocket logic in components.
+- **Markdown Rendering**: Use the `Markdown` component for AI responses to support syntax highlighting and Mermaid diagrams.
+- **Streaming**: Handle partial message updates in the UI to provide a "typing" effect.
+
+## State & Data
+- **Local Storage**: Cache generated wiki structures in `localStorage` to allow instant navigation.
+- **i18n**: Use `next-intl` for all user-facing strings. Access via `useTranslations` hook.
+- **Context**: Use `LanguageContext` for global language state.
+
+## Styling
+- **Tailwind CSS**: Use Tailwind for all styling. Follow the existing design system (dark/light mode support via `next-themes`).
+- **Icons**: Use `lucide-react` for consistent iconography.
+
+## Anti-Patterns
+- Do not fetch data directly from the backend in components; use the Next.js API proxy routes in `src/app/api/`.
+- Avoid large monolithic components; break down complex views like the Wiki page into smaller sub-components.

.cursor/rules/project-overview.mdc

@@ -0,0 +1,36 @@
+---
+description: General project context, architecture, and tech stack overview.
+globs:
+  - "**/*"
+alwaysApply: true
+---
+
+# DeepWiki-Open Project Overview
+
+DeepWiki-Open is a full-stack RAG (Retrieval-Augmented Generation) application designed to generate and interact with documentation for software repositories.
+
+## Tech Stack
+- **Frontend**: Next.js 15 (App Router), React 19, TypeScript, Tailwind CSS, Mermaid.js.
+- **Backend**: FastAPI (Python 3.10+), Adalflow (RAG framework), FAISS (Vector DB).
+- **AI**: Provider-agnostic (OpenAI, Google Gemini, Ollama, Azure, OpenRouter, etc.).
+
+## Core Architecture
+- **Modular Monolith**: The backend is organized into functional modules within `api/`.
+- **RAG Pipeline**: Uses `adalflow` to orchestrate document ingestion, embedding, and retrieval.
+- **Communication**: REST for configuration/metadata; WebSockets (`/ws/chat`) for real-time streaming chat.
+- **Persistence**: 
+  - Repositories: `~/.adalflow/repos/`
+  - Vector DBs: `~/.adalflow/databases/`
+  - Wiki Cache: `~/.adalflow/wikicache/` (JSON)
+
+## Key Directories
+- `api/`: FastAPI backend logic.
+- `api/tools/`: Embedders and utility clients.
+- `src/app/`: Next.js pages and API routes.
+- `src/components/`: Reusable React components.
+- `src/utils/`: Frontend utilities (WebSocket client, etc.).
+
+## Development Principles
+- **Provider Agnostic**: Always use the factory patterns in `api/config.py` and `api/tools/embedder.py` when adding LLM support.
+- **Type Safety**: Maintain parity between Pydantic models in `api/` and TypeScript interfaces in `src/types`.
+- **Streaming First**: Prioritize streaming responses for AI interactions to improve UX.

api/api.py

@@ -7,7 +7,6 @@
 import json
 from datetime import datetime
 from pydantic import BaseModel, Field
-import google.generativeai as genai
 import asyncio
 
 # Configure logging

api/config/embedder.json

@@ -18,8 +18,9 @@
     "client_class": "GoogleEmbedderClient",
     "batch_size": 100,
     "model_kwargs": {
-      "model": "text-embedding-004",
-      "task_type": "SEMANTIC_SIMILARITY"
+      "model": "gemini-embedding-001",
+      "task_type": "SEMANTIC_SIMILARITY",
+      "output_dimensionality": 3072
     }
   },
   "embedder_bedrock": {

api/config/generator.json

@@ -20,9 +20,15 @@
       }
     },
     "google": {
-      "default_model": "gemini-2.5-flash",
+      "default_model": "gemini-3-flash-preview",
       "supportsCustomModel": true,
       "models": {
+        "gemini-3-flash-preview": {
+          "temperature": 1.0,
+          "top_p": 0.8,
+          "top_k": 20,
+          "thinking_level": "high"
+        },
         "gemini-2.5-flash": {
           "temperature": 1.0,
           "top_p": 0.8,