====== DeepWiki ======

DeepWiki is a free, AI-powered documentation platform created by [[devin|Cognition Labs]] (the company behind the autonomous AI software engineer Devin) that automatically generates structured, wiki-style documentation for any GitHub repository. ((Source: [[https://cognition.ai/blog/deepwiki|Cognition AI Blog - DeepWiki: AI docs for any repo]]))

Launched on April 27, 2025, DeepWiki acts as a "Wikipedia of code," allowing developers to replace "github.com" with "deepwiki.com" in any repository URL to instantly receive AI-generated documentation complete with architecture diagrams, navigation, and interactive Q&A capabilities. ((Source: [[https://dev.to/czmilo/deepwiki-ai-driven-revolution-in-code-documentation-1jb4|DEV Community - DeepWiki: AI-Driven Revolution in Code Documentation]]))

===== Background =====

Cognition Labs, founded in August 2023 by competitive programmers Scott Wu, Steven Hao, and Walden Yan, first gained prominence with [[devin|Devin]], their autonomous AI software engineer, which was demoed in March 2024 and became generally available in December 2024. ((Source: [[https://en.wikipedia.org/wiki/Cognition_AI|Wikipedia - Cognition AI]]))

DeepWiki evolved from two internal Cognition tools: **Devin Wiki** and **Devin Search**, which were built to help developers quickly understand complex codebases. The public launch of DeepWiki made these capabilities freely available for all public GitHub repositories. ((Source: [[https://cognition.ai/blog/deepwiki|Cognition AI Blog - DeepWiki: AI docs for any repo]]))

To prepare for launch, Cognition pre-indexed over 50,000 of the top public GitHub repositories, reportedly spending more than $300,000 in compute costs to process over 4 billion lines of code. ((Source: [[https://dev.to/czmilo/deepwiki-ai-driven-revolution-in-code-documentation-1jb4|DEV Community - DeepWiki: AI-Driven Revolution in Code Documentation]]))

===== How It Works =====

DeepWiki combines large language models with graph-style analysis of a repository's structure to extract key concepts, relationships, and workflows from source code, configuration files, and existing documentation. ((Source: [[https://ai.miraheze.org/wiki/DeepWiki|Learn AI Wiki - DeepWiki]]))

The system operates through several stages:

  - **Repository Indexing**: DeepWiki clones and analyzes the repository, identifying file structure, dependencies, and code relationships.
  - **Content Generation**: LLMs process the indexed data to produce structured documentation pages covering architecture, modules, usage patterns, and API references.
  - **Diagram Generation**: Architecture and component diagrams are automatically generated using Mermaid.js, providing visual representations of system structure and data flows.
  - **RAG-Powered Q&A**: A retrieval-augmented generation (RAG) system enables users to ask natural language questions about the codebase and receive context-rich answers with code snippets and source references. ((Source: [[https://docs.devin.ai/work-with-devin/deepwiki|Devin Docs - DeepWiki]]))

For private repositories, DeepWiki is integrated into the Devin platform and automatically indexes connected repos, refreshing documentation periodically. ((Source: [[https://cognition.ai/blog/devin-2|Cognition AI Blog - Devin 2.0]]))

===== Key Features =====

  * **Instant Documentation**: Replace "github.com" with "deepwiki.com" in any public repository URL to generate a complete wiki.
  * **Architecture Diagrams**: Auto-generated Mermaid.js diagrams visualizing system architecture, component relationships, and data flows.
  * **Ask Devin Integration**: An interactive Q&A feature powered by Devin's code search capabilities, allowing developers to ask complex questions about a codebase and receive detailed, source-grounded answers. ((Source: [[https://docs.devin.ai/work-with-devin/deepwiki|Devin Docs - DeepWiki]]))
  * **Deep Research Mode**: Advanced querying that performs codebase-wide analysis for tasks such as security vulnerability identification and migration planning.
  * **Structured Navigation**: Generated wikis include hierarchical tables of contents, module-level pages, and cross-references between components.
  * **Source Links**: All documentation pages link back to the relevant source code lines in the original repository.
  * **Badge Maker**: Repository maintainers can generate DeepWiki badges to embed in their README files.

===== MCP Server =====

On May 22, 2025, Cognition launched the official DeepWiki Model Context Protocol (MCP) server, providing free programmatic access to DeepWiki's documentation and Q&A capabilities with no authentication required. ((Source: [[https://cognition.ai/blog/deepwiki-mcp-server|Cognition AI Blog - The DeepWiki MCP Server]]))

The MCP server exposes three tools:

  * **ask_question**: Ask natural language questions about any indexed repository and receive AI-generated answers.
  * **read_wiki_contents**: Retrieve the contents of specific wiki pages.
  * **read_wiki_structure**: Discover the structure and navigation of a repository's generated wiki.

The server is available as an official Docker MCP catalog entry and can be integrated with AI assistants such as Claude Desktop, VS Code, and other MCP-compatible clients. OpenAI also published examples of using the DeepWiki MCP server with their API. ((Source: [[https://cognition.ai/blog/deepwiki-mcp-server|Cognition AI Blog - The DeepWiki MCP Server]]))

===== Adoption =====

Within two weeks of launch, teams at Google, Hugging Face, Perplexity, Accenture, and other companies had added DeepWiki links to their repositories, making it easier for developers to understand and contribute to their projects. ((Source: [[https://www.linkedin.com/posts/cognition-ai-labs_developer-activity-7326657159860015104-Ied_|LinkedIn - Cognition post on DeepWiki adoption]]))

DeepWiki has been used to document repositories ranging from small open-source utilities to massive enterprise codebases. Cognition reported that one banking client used Devin Wiki (the private version) to generate documentation across more than 400,000 internal repositories. ((Source: [[https://cognition.ai/blog/devin-annual-performance-review-2025|Cognition AI Blog - Devin Annual Performance Review 2025]]))

===== Open Source Alternatives =====

The launch of DeepWiki inspired the creation of **DeepWiki-Open**, an open-source alternative created by AsyncFuncAI and released on April 30, 2025, just days after DeepWiki's announcement. ((Source: [[https://github.com/AsyncFuncAI/deepwiki-open|GitHub - AsyncFuncAI/deepwiki-open]]))

DeepWiki-Open replicates the core concept using a Python/FastAPI backend and TypeScript/Next.js frontend, supporting multiple LLM providers including OpenAI, Google Gemini, and local models via Ollama. It supports GitHub, GitLab, and Bitbucket repositories and can be fully self-hosted. The project gained over 15,000 GitHub stars and attracted 70 contributors. ((Source: [[https://github.com/AsyncFuncAI/deepwiki-open|GitHub - AsyncFuncAI/deepwiki-open]]))

===== See Also =====

  * [[devin|Devin]]
  * [[code_generation_agents|Code Generation Agents]]
  * [[how_to_build_a_coding_agent|How to Build a Coding Agent]]

===== References =====