Add up what your team spends on AI coding tools. Copilot, $19 per seat per month. CodeRabbit, $24 per seat per month. Cursor Pro, $20 per seat per month. Now look at what those tools actually do: they call an API – the same APIs you already have access to – and add a layer of UX on top. You're paying for a middleman.
The markup is the product
Most AI developer tools make money on the spread between what they pay for API calls and what they charge you. This is a straightforward business model, and it's the reason these tools exist as subscription services rather than one-time purchases.
A typical AI-powered code review might consume $0.20 to $0.50 in API tokens. The tool charges $24 per month per seat for unlimited reviews. If a developer runs 10 reviews a month, the tool is collecting $24 for roughly $3 worth of API calls. That's an 8x markup. For lighter users, the ratio is even higher.
This model has a structural consequence that goes beyond pricing: it requires sending your code to someone else's servers. The tool is the one calling the API, so your code has to pass through their infrastructure. They need to see your code, process it, send it to the model provider, and return the results. This is the architecture that the middleman model demands.
The tool is the markup. The AI underneath is the same AI you can access directly through your own subscriptions.
The BYOK model
VibeRails doesn't run a cloud backend service, and it doesn't route your code through VibeRails servers.
Instead, it orchestrates your local installations of Claude Code and Codex CLI. These are tools you've already installed on your machine. They use your API keys and your subscriptions, and they talk directly to your chosen AI provider. VibeRails provides the orchestration layer – it structures the review, manages the analysis pipeline, presents findings, and coordinates fix sessions – but the actual model inference is performed by the AI provider you have configured in those tools and billed to your account.
This means three things:
- No VibeRails cloud. VibeRails runs locally and orchestrates your existing Claude Code/Codex setup – you're not uploading your code to a separate vendor platform. When analysis runs, relevant code is sent from your machine directly to the AI provider configured in those tools.
- Your API keys stay on your machine. VibeRails never sees them, stores them, or transmits them. The CLI tools you've already configured handle authentication directly.
- Your existing subscriptions pay for the compute. If you have a Claude Code subscription, that's what powers the review. VibeRails doesn't add any inference cost on top.
This is what we call BYOK – Bring Your Own Key. It's not a pricing trick. It's a fundamentally different architecture that changes what a code review tool costs and what it needs to know about your code.
The math
Here's what full-codebase code review costs with the conventional model versus BYOK.
CodeRabbit for a 20-person team: $24/seat/month × 20 seats × 12 months = $5,760 per year. And that covers PR-scoped review only – diffs, not the full codebase.
SonarQube Enterprise: starting at $20,000+ per year for self-hosted. Rule-based analysis only, no AI reasoning.
VibeRails: $299 per developer for a lifetime licence, or $19/mo per developer if you prefer monthly. One licence per machine. Full-codebase AI code review across 17 detection categories, triage mode, AI-assisted fix sessions.
The free tier gives you up to 5 issues per review session. No signup. No credit card. Download and run your first review in minutes.
The economics are only possible because of BYOK. When the tool doesn't bear inference costs, the tool doesn't need to be a subscription. See the full breakdown on the pricing page.
Local orchestration is the future
AI subscriptions are becoming as common as IDE licenses. Most professional developers already pay for at least one: Anthropic's Claude, OpenAI's ChatGPT or Codex, or both. These subscriptions include powerful CLI tools that can read files, reason about code, and generate changes – the same capabilities that cloud-hosted AI dev tools charge separately to access.
As this becomes the norm, the value proposition of tools that resell API access gets harder to justify. Why pay $24/seat/month for a tool to call Claude on your behalf when Claude Code is already installed on your machine? Why send your proprietary code to a third-party vendor platform when you can run the orchestration on your desktop and send code directly to the AI provider you already use?
The tooling layer still matters – orchestration, UX, structured output, triage workflows, fix pipelines. These are real problems that need good solutions. But the AI itself doesn't need to be bundled in. The tools that understand this will offer better economics, better privacy, and better alignment with how developers already work.
Download VibeRails and run your first full-codebase review. Your AI subscription does the heavy lifting. The tool just makes it useful.
And if you want to pay nothing at all for AI compute: open-weight models now achieve near-cloud-API coding quality. Run a local model with Ollama, point VibeRails at it, and the entire review pipeline costs only electricity. See the local AI code review guide.
Limits and tradeoffs
- It can miss context. Treat findings as prompts for investigation, not verdicts.
- False positives happen. Plan a quick triage pass before you schedule work.
- Privacy depends on your model setup. If you use a cloud model, relevant code is sent to that provider; local models can keep inference on your own hardware.