Anatomy of Claude Code
A visual, interactive deep-dive into Anthropic's AI coding assistant. 1,902 source files, 512,000 lines of code, one interactive visualization to understand it all.
🎮 Pixel Mini-Games →0
Source files
0K+
Lines of code
0+
Built-in tools
0
Core systems
Journey of a Conversation
When you type 'help me fix this bug', what does Claude Code do behind the scenes?
⌨️User Input
You type 'help me fix this bug' in the terminal. The PromptInput component captures your input, attaching context like the current directory and git status.
🛡️Permission Check
The three-layer defense activates: ML classifier quickly assesses safety → rule engine matches user-defined policies → confirmation dialog pops up if needed.
Learn more →📋System Prompt Construction
Assembles full context: default system prompt + user environment + 45 tool definitions + memory attachments + project rules. Like a waiter translating your simple order into a complete ticket the kitchen understands.
Learn more →🌊Streaming API Call
Sends the request via the Anthropic API. The response streams back token by token — each one rendered to the terminal immediately. Async generators power it all.
Learn more →🔧Tool Invocation
Claude decides it needs to read files, search code, or run commands. The tool scheduler kicks in: read ops run in parallel (fast), write ops run serially (safe).
Learn more →🔄Loop & Return
Tool results go back to Claude, which analyzes them and may call more tools. This loop continues until the task is complete, streaming the final result to your terminal.
System Architecture
15 subsystems, 4 architecture layers. Hover for details, scroll to zoom.