Why RTK Wasn't Enough (And What I Added)

RTK (Reduce Toolkit) is a solid Rust CLI for reducing AI context size. I used it daily. Then I hit its limits.

RTK handles ANSI stripping and basic deduplication well. But real-world CLI output has patterns RTK doesn't catch. I forked it and built ContextZip.

What RTK Does Well

• ANSI escape code removal
• Basic line deduplication
• Character count reporting
• Clean Rust codebase

What Was Missing

1. Language-aware stack trace filtering. RTK treats stack traces as plain text. It doesn't know that node:internal/modules/cjs/loader is a framework frame and /app/src/server.ts is your code. ContextZip recognizes stack trace formats for Node.js, Python, Rust, Go, Java, and C# — and strips framework frames while keeping application frames.

2. Semantic duplicate grouping. RTK deduplicates exact matches. But 40 TypeScript errors with the same message but different file paths aren't exact duplicates. ContextZip groups by error pattern, not exact string match.

3. Command-specific patterns. npm install progress bars use ANSI cursor movement, not just color codes. Docker layer IDs are unique strings that RTK won't deduplicate. ContextZip has 102 command-specific patterns.

4. Savings tracking. contextzip gain tracks your cumulative savings over time. RTK only reports per-command stats.

5. Web content extraction. When AI agents fetch web pages, the HTML structure is noise. ContextZip extracts content from HTML.

The Numbers

RTK on a typical session: 30-50% reduction.
ContextZip on the same session: 60-90% reduction.

The difference is language-aware filtering and semantic grouping. Generic text processing gets you halfway. Understanding the meaning of the output gets you the rest.

cargo install contextzip
eval "$(contextzip init)"

GitHub: github.com/contextzip/contextzip

Part of the ContextZip Daily series. Follow for daily tips on optimizing your AI coding workflow.

Install: npx contextzip | GitHub: jee599/contextzip