Google has introduced Code Wiki, a new platform designed to keep software documentation continuously synchronized with the code it describes. The system generates a structured wiki for each repository, automatically updates it after every change, and powers an integrated chat interface that understands the entire codebase. The public preview, available now for open-source projects, aims to reduce one of the most expensive bottlenecks in software development: reading and understanding existing code.
Rather than relying on static files that become outdated, Code Wiki automatically reconstructs documentation—scanning the full repository, maintaining links to every symbol, and regenerating diagrams that reflect the current state of the code. Developers can move fluidly between high-level explanations and the exact files, classes, and functions referenced in the wiki. If something is still not clear, a Gemini-powered chat agent answers questions using the full, always-current wiki as its knowledge source.
The public preview website ingests open-source repositories and hosts interactive documentation for them. Architecture diagrams, class relationships, and sequence flows are generated on the fly and updated every time the code changes. Google says the goal is to enable new contributors to understand a system’s framework within minutes, while helping experienced developers onboard into unfamiliar modules without the need for slow, manual exploration.
A local version is also on the way. Google is developing a Gemini CLI extension for teams that want to run the same system privately on internal repositories—particularly valuable in companies where legacy codebases are poorly documented and institutional knowledge has faded. Interested teams can now join the waitlist.
Early reactions highlight both the enthusiasm and the questions surrounding the approach. On Reddit, one user wrote:
This is actually huge. I’m guessing that in the future, the idea is to have Gemini CLI query the wiki for instant context/information instead of going through the files and figuring things out manually.
Another user pushed back, noting the balance between local context and external tooling:
The CLI needs to read a file before editing it, so it’s in the model’s context anyway—there is no added value in having an external tool tell the CLI about your codebase while the CLI has access to your codebase.
And on YouTube, one viewer pointed to a common workflow request:
It would be good if you could print an MD or PDF of the complete documentation generated, including code and citations.
With Code Wiki, Google is betting that documentation should be living, structured, and continuously regenerated—not a side task developers struggle to maintain. The company frames it as a shift toward “instant understanding,” where navigating unfamiliar code becomes less about deciphering and more about exploring.
