Codemap

Codemap began as a simple internal tool to help me with git commits, but it evolved into something much more ambitious - a comprehensive AI-powered coding assistant that represents the culmination of everything I learned from Drool, Learnway, and my journey into development and AI.

From Internal Tool to Coding Assistant

The story of Codemap really begins with frustration. As I was working on Learnway and diving deeper into development, I found myself constantly struggling with git commit messages. I’d make changes, stare at the terminal, and waste time crafting commit messages that captured what I actually did.

So I built a small script to analyze my git changes and generate commit messages automatically. It was nothing fancy - just a personal productivity hack that saved me a few minutes each day.

But then something interesting happened. As I continued working on complex codebases, I realized the same AI techniques that helped with commits could solve much bigger problems. What if I could ask questions about my code? What if I could search semantically through my repositories? What if I could generate documentation automatically?

The Evolution of Features

Codemap grew organically based on actual needs:

# Smart commit messages
cm commit

# AI-powered pull requests
cm pr

# Ask questions about your codebase
cm ask "Which module handles authentication?"

# Generate documentation
cm gen src/

# Semantic code search
cm index && cm search "error handling patterns"

The tool became a comprehensive developer toolkit with features I never originally planned:

• Multi-LLM Support: Integration with various AI providers through PydanticAI
• Semantic Search: Vector-based code understanding using embeddings
• Documentation Generation: Automatic creation of project documentation
• Git Integration: Smart commits, PR management, and workflow optimization
• Code Analysis: Deep understanding of codebase structure and dependencies

The CodeStruct Innovation

One of the most interesting developments was creating CodeStruct - a domain-specific language for representing code structure in a compressed, hierarchical format designed specifically for LLM context compression.

module: user_management
  class: User
    doc: Represents a user in the system...
    attr: name [type: STRING]
    attr: age [type: INTEGER]
    func: greet
      doc: Greets the user...
      param: name [type: STRING]
      returns: STRING

This notation can be minified for efficient AI processing:

m:user_management;cl:User;at:name[t:STR],at:age[t:INT];fn:greet|p:name[t:STR],r:STR

CodeStruct emerged from the practical need to feed large codebases to AI models efficiently while preserving semantic meaning - a problem I encountered repeatedly while building AI-powered developer tools.

The Reality Check

Here’s the honest truth: while Codemap became a sophisticated tool with comprehensive features, the landscape changed quickly. The advent of Claude Code and similar AI coding assistants has made many of Codemap’s features somewhat redundant.

I still maintain the project, but I don’t update it as frequently as I once did. The market moved faster than a solo developer could keep up with, and the big tech companies built similar (and often better) solutions with massive teams and resources.

The Real Value

Despite not becoming a commercial success, Codemap delivered something invaluable: it completed my transformation from a designer who could dabble in code to a developer who could build complex, AI-powered systems from scratch.

The project taught me:

• Advanced Python development with proper testing, CI/CD, and package management
• AI orchestration at scale with multiple model providers
• Domain-specific language design through CodeStruct
• Developer tool UX principles that actually matter to users
• Open source project management with proper documentation and community building

Current Status

Codemap exists in a interesting space - it’s still somewhat active and functional. But it’s also a reminder that timing and market fit matter as much as technical execution.

Sometimes the most valuable projects aren’t the ones that achieve market success, but the ones that push you to grow in unexpected directions. Codemap represents the culmination of my journey from designer to developer - a journey that started with frustration over git commits and ended with a deep understanding of AI, developer tools, and the challenges of building in rapidly evolving markets.