From this post, I learned that Cognition, the company behind Devin created DeepWiki. They’re calling it a “Deep Research for Github.” They created explainer-wiki style web pages for the top 30,000 most popular Github repos.
From Silas Alberti, he said:
- 30k repos already indexed
- processed 4 billion+ lines of code
- the indexing alone cost $300k+ in compute
Some interesting points he makes in his thread:
- indexing a single repo costs $12 on average
- models are very good at understanding code locally. the hard part is understanding the global structure of the codebase.
- DeepWiki breaks down a codebase into a hierarchy of high-level systems and then generates a wiki page for every system
- one interesting signal that we’re able to leverage is the commit history. who tends to touch which files together? you can represent this as a graph.
You can also do Q&A on each project after clicking into it. I looked at a handful of examples and can see how valuable this is.
Some ones I briefly looked at:
- https://deepwiki.com/pandas-dev/pandas
- https://deepwiki.com/langchain-ai/langgraph
- https://deepwiki.com/BerriAI/litellm
- https://deepwiki.com/modelcontextprotocol/modelcontextprotocol
The generated diagrams are nice: 
A lot of MCP servers are already in the top 30k repos: 