Agent-Generated Architectural Wikis are really neat

We all love good documentation. We all want good documentation. But documentation maintenance is expensive, and even with the best intentions, things go stale.

A month-old README might be accurate. A six-month-old architecture doc? Maybe. That onboarding guide from last year? Good luck.

There's a new category of tooling that generates comprehensive architectural wikis from your codebase. They trace every claim back to source code and can be refreshed whenever you need. The tooling exists today, and it's genuinely useful for onboarding to and understanding complex codebases.

DeepWiki

DeepWiki from the Devin team generates comprehensive documentation sites from your codebase. Think Docusaurus or GitBook, but automatically generated and continuously refreshed by an AI agent.

Key features from their VSCode wiki:

• Traceability: Every statement links back to the actual source code that supports it
• Visual docs: Auto-generated Mermaid diagrams for architecture, data flow, and relationships
• Smart organization: Sidebar navigation that mirrors how developers think about the codebase
• Chat interface: Query the knowledge store with answers grounded in source code
• Auto-refresh: Configure a GitHub flag to re-index on significant changes

Google's CodeWiki: Video Intros

Google's CodeWiki automatically generates continuously updated architecture docs with diagram support and a chat interface that links answers to source code. The key innovation is NotebookLM-generated ~10-minute video overviews of your codebase that explain organization, component responsibilities, and architectural decisions. Watch the intro, then explore the wiki with full traceability back to source.

Building in the Open: ted-mosby

I was so impressed by this category of tooling that I decided to build my own open-source entry: ted-mosby.

Why "ted-mosby"?

Because good architecture documentation tells a story. It explains not just what your code does, but why it's organized the way it is. (And yes, the name is a How I Met Your Mother reference. Ted was an architect.)

What It Does

Built with buildanagentworkshop.com (also my tool), ted-mosby is capable of:

• Digesting your codebase and generating markdown documentation
• Creating content suitable for both human reading and agent consumption
• Generating static sites from the documentation
• Being refreshed and re-run as your codebase evolves

The Open Source Alternative

DeepWiki and CodeWiki are phenomenal tools, but there's value in an open-source alternative that developers can customize for specific project needs, integrate into existing documentation workflows, and extend with their own analysis and visualization tools.

ted-mosby is built using the Claude Agent SDK, so it's essentially a custom Claude Code agent geared specifically towards this documentation task. I'm continuing to iterate on it, and I'm exploring making it into a Claude Code Agent Skill. Imagine running /generate-wiki in your terminal and having comprehensive architectural documentation appear in your docs/ folder.

The Meta Layer: Wikis as Agent Context

These architectural wikis can serve as context for coding agents themselves.

If you're using Claude Code, Cursor, Gemini CLI, or a custom agent built with something like buildanagentworkshop.com, you can point the agent at your DeepWiki, CodeWiki, or ted-mosby-generated documentation.

Accelerated Agent Onboarding

Instead of running /init with Claude Code to generate an initial doc (which is a fantastic start, but usually just the essentials):

# With /init (standard approach)
Agent: *generates basic README with project overview*
Coverage: High-level essentials
Detail: Broad but shallow

# With architectural wiki
Agent: *reads comprehensive wiki with traced citations, diagrams, and deep context*
Coverage: Complete architectural understanding
Detail: Deep and verifiable

The architectural wiki is pre-digested context. Another agent has already done the work of understanding the codebase structure, and now your coding agent can onboard with comprehensive understanding rather than just the basics.

Better Agent Collaboration

When working with open-source tools, the architectural wiki becomes a shared knowledge base:

• Human reads wiki to understand the codebase
• Agent reads wiki to understand the codebase
• Both can reference the same conceptual model
• Collaboration becomes more efficient

This creates a positive feedback loop: better documentation leads to better agent understanding, better agent contributions, better codebase, and back to better documentation.

Why This Matters

As always with these tools, this wiki approach is a fantastic development for anyone consuming or contributing to open source. Key benefits:

• Lower barriers: Complex projects become approachable with 10-minute intros and traceable wikis
• Continuous updates: Docs regenerate automatically on significant changes
• Preserved knowledge: Architectural decisions get captured with links to implementing code
• Better agent context: Comprehensive wikis help coding agents onboard faster

Getting Started

Hosted: Try DeepWiki. CodeWiki is currently in private beta, but worth watching.

Generate it yourself!: Install ted-mosby via npm

npm install -g ted-mosby
ted-mosby generate

ted-mosby is early in development. Feel free to give it a star or contribute if you have ideas!

Point your coding agent at the generated wiki for faster onboarding.

The tools exist today. DeepWiki is production-ready, CodeWiki is in private beta, and ted-mosby is growing. We're moving from manually written, quickly stale docs to agent-generated, continuously fresh documentation that traces back to source.

I'm excited to see this approach and tooling evolve. If you're working on documentation tooling, agent systems, or just care about making codebases more approachable, I'd love to hear from you.

Links and Resources

• DeepWiki: deepwiki.com
• Google CodeWiki: codewiki.dev
• ted-mosby (my open-source tool): npmjs.com/package/ted-mosby
• buildanagentworkshop.com (also my tool): Platform I used to build ted-mosby