DeepWiki-Open (Grok-Wiki)

DeepWiki-Open (also known as Grok-Wiki) is a tool designed to automatically generate interactive, high-quality wikis for GitHub, GitLab, or BitBucket repositories.

When provided with a repository name, the tool performs the following workflow:

1. Code Structure Analysis: Analyzes the repository's architecture.
2. Documentation Generation: Creates comprehensive technical documentation.
3. Visual Diagramming: Generates diagrams to explain system logic and flow.
4. Wiki Organization: Structures the content into a navigable wiki format.
5. Codemap Generation: Creates code-centric guided tours to help users navigate the codebase.

The DeepWiki-Open repository is organized into a backend API and a frontend web application:

• api/: The backend API server.
  • main.py: API entry point.
  • api.py: FastAPI implementation.
  • rag.py: Retrieval Augmented Generation logic.
  • data_pipeline.py: Data processing utilities.
  • requirements.txt: Python dependencies.
• src/: The Next.js frontend application.
  • app/: Next.js application directory.
  • components/: React components (e.g., Mermaid.tsx for diagram rendering).
• public/: Static assets.
• .env: Environment variable configuration file.

Understanding the directory layout for development or customization:

• api/: The backend API server (FastAPI).
  • main.py: API entry point.
  • api.py: FastAPI implementation.
  • rag.py: RAG (Retrieval-Augmented Generation) logic.
  • data_pipeline.py: Data processing utilities.
• src/: The Next.js client application.
  • app/: Next.js application directory.
  • components/: React components (e.g., Mermaid.tsx for diagram rendering).
• public/: Static assets.
• .env: Environment variable configuration.

DeepWiki provides two primary ways to interact with your repository:

Ask (RAG-based Chat)

Uses Retrieval Augmented Generation (RAG) to chat with your repository. It provides context-aware answers based on actual code snippets, supports real-time streaming, and maintains conversation history.

DeepResearch (Multi-turn Research)

Performs an in-depth, multi-turn research process for complex topics. It follows a structured process:

1. Research Plan: Outlines the approach and initial findings.
2. Research Update: Adds new insights based on previous iterations.
3. Final Conclusion: Synthesizes all iterations into a comprehensive answer.

How to use DeepResearch: Before submitting a question in the Ask interface, toggle the "Deep Research" switch to ON. The AI may iterate up to 5 times automatically to complete the research.

DeepWiki-Open automates the creation of interactive wikis for GitHub, GitLab, or Bitbucket repositories through the following workflow:

1. Repository Access: Clones the repository. For private repositories, users must provide a Personal Access Token.
2. Analysis: Analyzes the code structure and creates embeddings for intelligent retrieval.
3. AI Generation: Uses a selected AI model (Google Gemini, OpenAI, OpenRouter, Azure OpenAI, or local Ollama) to generate context-aware documentation.
4. Visualization: Automatically creates Mermaid diagrams to visualize architecture and data flows.
5. Organization: Compiles everything into a structured, navigable wiki.
6. Interaction: Enables a "Ask" feature (RAG-based Q&A) and "DeepResearch" for complex topic exploration.

DeepWiki supports multiple model providers. You can switch between them by providing the corresponding API keys in your environment variables.

Supported Providers:

• Google: Default model gemini-2.5-flash. Also supports gemini-2.5-flash-lite, gemini-2.5-pro, etc.
• OpenAI: Default model gpt-5-nano. Also supports gpt-5, 4o, etc.
• OpenRouter: Access to various models (Claude, Llama, Mistral, etc.) via a single API.
• Azure OpenAI: Default model gpt-4o.
• Ollama: Local open-source models (e.g., llama3).

To use a custom OpenAI-compatible endpoint (like Alibaba Qwen) for embeddings, replace api/config/embedder.json with embedder_openai_compatible.json and set OPENAI_BASE_URL.

# Example for OpenAI-compatible embedding providers
OPENAI_API_KEY=your_key
OPENAI_BASE_URL=https://compatible-endpoint/v1

DeepWiki provides two primary ways to interact with your repositories:

Questioning (RAG-driven)

Allows you to chat with your repository using Retrieval-Augmented Generation (RAG). It provides context-aware responses based on actual code, supports real-time streaming, and maintains conversation history.

Deep Research

An advanced multi-turn research process that performs deep investigations into complex topics. It follows a structured plan:

1. Research Plan: Outlines the method and initial findings.
2. Research Updates: Adds new insights based on previous iterations.
3. Final Conclusion: Provides a comprehensive answer based on all iterations.

Deep Research can automatically iterate up to 5 times to reach a conclusion. To use it, toggle the "Deep Research" switch in the questioning interface before submitting your question.

DeepWiki provides two primary ways to interact with your repository:

Questions (RAG)

Allows you to chat with your repository using Retrieval Augmented Generation (RAG). It provides contextual answers based on actual code, uses real-time streaming, and maintains conversation history.

DeepResearch

An advanced, multi-step research process for complex topics. It follows a structured plan and automatically continues research for up to 5 iterations until a conclusion is reached.

Research Stages:

1. Research Plan: Describes the approach and initial findings.
2. Research Updates: Iterative insights built on previous steps.
3. Final Conclusion: A comprehensive response based on all iterations.

To use DeepResearch: Toggle the "Pesquisa Aprofundada" (Deep Research) switch in the Questions interface before sending your query.