Why I use a $2/month Claude API instead of the official one — for Linux kernel contributions
This week, Linus Torvalds merged official documentation into the Linux kernel about using AI coding assistants. The kernel project now has an official stance: AI tools are allowed, but with strict guidelines.
That's a big deal. The most conservative codebase in the world just opened the door to Claude Code, Copilot, and friends.
But there's a problem nobody talks about: kernel contributions mean long, deep sessions.
The real cost of kernel-level AI assistance
Kernel work isn't like writing a React component. You're reading through:
- Driver subsystems with decades of history
- Memory management code with careful invariants
- Network stack implementations with subtle protocol requirements
A single kernel contribution session might involve:
- Reading 50+ files for context
- Cross-referencing commit history
- Understanding 3-4 subsystems simultaneously
- Multiple review iterations with maintainers
That's not a 10-minute session. That's a 2-3 hour deep dive. Multiple times per week.
The rate limit wall
Here's what the Anthropic pricing page doesn't tell you upfront:
Claude Code's $20/month plan has rate limits. When you're 90 minutes into a kernel debugging session — stack trace analyzed, subsystem context loaded, finally approaching the fix — you hit the wall:
Rate limit exceeded. Please wait before making more requests.
For kernel work, this isn't just annoying. It's potentially hours of lost context.
What I do instead
I use the Claude API directly via a $2/month proxy that gives me unlimited usage for a flat monthly fee.
Here's the setup:
# Install Claude Code
npm install -g @anthropic-ai/claude-code
# Set your API base URL to the proxy
export ANTHROPIC_BASE_URL=https://simplylouie.com
export ANTHROPIC_API_KEY=your-key-here
# Now run Claude Code normally
claude
That's it. Claude Code doesn't know the difference. It sends requests to the proxy, the proxy forwards to Anthropic, you get responses back. Same model, same quality, no rate limits, $2/month.
Real workflow for kernel contributions
Here's how I actually use this for kernel work:
1. Subsystem orientation
claude "I'm looking at the scheduler subsystem, specifically the CFS implementation in kernel/sched/fair.c. Give me a map of the key data structures and how they interact."
This is a long context load. With rate limits, this alone might eat your quota for the session.
2. Bug investigation
claude "Here's a stack trace from a kernel oops in the network stack: [paste stack]. Walk me through what each frame means and what state the system was in."
3. Patch review
claude "Review this patch for correctness before I submit to LKML. Check for: locking violations, memory leak potential, style compliance with kernel coding standards."
4. Commit message drafting
Kernel commit messages have specific requirements. Claude is excellent at structuring them correctly:
claude "Draft a kernel commit message for this change. Include: what problem it solves, why this approach, any relevant Fixes: tags."
The economics
If you're contributing to the Linux kernel, you're probably a professional engineer. Your time is worth something.
- Claude Code subscription: $20/month with rate limits
- Getting blocked 90 minutes into a debugging session: priceless (in the bad way)
- SimplyLouie API proxy: $2/month, unlimited usage
For kernel contributors specifically — long sessions, deep context, multiple subsystems — the flat-rate model makes much more sense than a usage-throttled subscription.
About the project
SimplyLouie is a $2/month Claude API proxy. No rate limits, same Anthropic models. 7-day free trial, no charge until you decide to continue.
50% of revenue goes to animal rescue. The founder's rescue dog inspired the project — the idea being that if you save a little money on AI tools, maybe some of that goodwill flows to animals who need it.
The kernel's AI assistance docs are worth reading: github.com/torvalds/linux/blob/master/Documentation/process/coding-assistants.rst
Top comments (0)