ChatGPT reads your browser state. Copilot injects ads. I'm using a $2/month Claude API instead.

This week was rough for AI trust.

First: ChatGPT won't let you type until Cloudflare reads your React state — someone decrypted the fingerprinting code that runs before every ChatGPT session. It's reading your browser's internal React state tree. Before you type a single character.

Then: Copilot edited an ad into a pull request. A Microsoft-owned coding tool injected promotional content into code review.

Both stories hit the top of Hacker News simultaneously. Both have hundreds of comments from developers who are done.

I left months ago. Here's what I use instead.

The problem with Big Tech AI

When the product is free or cheap, you're not the customer — you're the data source. ChatGPT at $20/month is still OpenAI's product, trained on your conversations, fingerprinting your browser, optimizing for engagement over accuracy.

Copilot is Microsoft. Microsoft's revenue model is not "help developers." It's enterprise contracts, ad revenue, and platform lock-in.

The incentive structures are misaligned. That's not a conspiracy — it's just business.

What I actually use

I pointed Claude Code at a cheap API proxy instead of Anthropic's official endpoint.

One environment variable:

export ANTHROPIC_BASE_URL=https://api.simplylouie.com

That's it. Claude Code works exactly the same. Same models, same tool calls, same MCP support.

The API costs $2/month. Not $20. Not $200. $2.

No browser fingerprinting. No ad injection. No React state surveillance. Just API calls.

The actual config

If you want to lock down the whole setup, here's my .claude/settings.json:

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.simplylouie.com"
  },
  "denyPermissions": [
    "Bash(git reset*)",
    "Bash(git push --force*)",
    "Bash(rm -rf*)"
  ],
  "confirmWrite": true
}

This does three things:

1. Routes Claude Code through the cheap API instead of direct Anthropic
2. Blocks the destructive git commands that caused last week's incident
3. Confirms before writing files

The denyPermissions list is non-negotiable after the git reset issue. The ANTHROPIC_BASE_URL override saves me $18/month.

Why Claude specifically

I've tested all the main models. Claude is the one that tells me when I'm wrong.

ChatGPT validates. Gemini hedges. Claude pushes back.

For code review, that matters. I don't want an AI that agrees with my bad architecture choices. I want one that flags them before I ship.

The math

• ChatGPT Plus: $20/month
• GitHub Copilot: $10/month
• Claude Pro: $20/month
• Claude via SimplyLouie API: $2/month

If you're using any of the Big Three AI tools, you're paying 5-10x more than you need to for the same underlying model access.

How to switch

1. Sign up at simplylouie.com — 7-day free trial, card required
2. Get your API key
3. Set ANTHROPIC_BASE_URL=https://api.simplylouie.com in your environment
4. Everything that worked before still works

Claude Code, LiteLLM, any OpenAI-compatible client — they all work with a base URL override.

One more thing

50% of SimplyLouie's revenue goes to animal rescue. Not to browser fingerprinting R&D.

The other 50% keeps the servers running.

That's the whole business model. No surveillance, no ads, no enterprise sales funnel. Just cheap API access and rescue dogs.

---

If you switched away from ChatGPT or Copilot recently, what was your breaking point? Drop it in the comments.