[PROPOSAL] Golang Rewrite to Address Critical Memory Usage (1.5-2GB per session)

Problem: Excessive Memory Usage Makes Claude Code Unusable on Standard Hardware

Current State on macOS (2.0.17):

Each Claude Code session consumes 1.5-2GB of active memory:

Process             Active  Compressed
claude heap-keeper  2.16GB  371MB
claude iterm2-fork  1.76GB  295MB  
claude gw-audit     1.66GB  271MB
claude iterm2-fork  1.46GB  251MB
claude hubspot      1.02GB  173MB

Real-World Impact:

• 16GB Mac: Cannot run more than 3-4 concurrent Claude sessions
• Had to upgrade to Mac Studio 64GB just to work with multiple projects
• Memory pressure forces constant session restarts
• Swap usage degrades performance significantly

Related Issues:

• #4953: Process grows to 120GB+ RAM and gets OOM killed (Linux)
• #8706: Resource exhaustion with Sonnet 4.5 (WSL2)
• #9579: Autocompacting loop causing massive token usage
• Multiple other perf:memory tagged issues

Proposal: Golang Rewrite

I'm offering to rewrite the Claude Code client in Go (under NDA if required) to address these fundamental performance issues:

Why Golang?

Memory Efficiency:

• Go binaries typically use 10-50MB vs Node.js 500MB-2GB
• Efficient garbage collection with lower overhead
• No V8 engine overhead (100-200MB baseline)
• Static compilation eliminates runtime dependencies

Performance:

• 5-10x faster startup time (no JIT warmup)
• Lower CPU usage (compiled vs interpreted)
• Better concurrency primitives (goroutines vs async/promises)

Production Examples:

• Docker CLI: ~30MB memory footprint
• kubectl: ~50MB memory footprint
• GitHub CLI (gh): ~40MB memory footprint
• All handle complex terminal UIs, API interactions, and concurrent operations

Estimated Improvements

• Memory: 1.5-2GB → 30-100MB per session (15-50x reduction)
• Startup: 2-3s → 100-300ms (10x faster)
• CPU: Significantly lower idle and active usage
• Disk: Single ~20MB binary vs 200MB+ node_modules

Implementation Approach

Phase 1: Core Rewrite (3-4 months)

• API client (using official anthropic-sdk-go)
• Terminal UI (using bubbletea or tview)
• File operations and git integration
• MCP server protocol

Phase 2: Feature Parity (2-3 months)

• Hooks system
• All current features and flags
• Migration tooling for existing configs

Phase 3: Testing & Migration (1-2 months)

• Beta testing with community
• Performance benchmarking
• Gradual rollout

About Me

• Experienced Go developer (can provide references)
• Willing to work under NDA
• Can commit 20-30 hours/week
• No compensation required (contribution to open source)

Questions for Anthropic Team

1. Would you be open to a Golang rewrite?
2. What are the technical constraints/requirements?
3. How can we structure this collaboration?
4. Is there existing internal work in this direction?

---

The current Node.js implementation is a major barrier to adoption. Multiple users have reported needing to upgrade hardware just to use Claude Code effectively. A memory-efficient rewrite would make this tool accessible on standard development machines.

I'm ready to start immediately if there's interest.