AI is writing your code and making decisions that affect your systems. VIBES helps you track every action, prove it's authentic, and stay in control — so you can ship and operate with confidence.
AI is writing your code and making decisions that affect your systems. Can you answer these questions?
Which lines did AI write — and what decisions did it make? If a model gets compromised, can you find every line it touched and every action it took across your entire system?
What was it asked to do? When a bug appears or an agent makes a bad deployment call, do you know what prompt or context led to that outcome?
How did it reason? When AI makes a surprising choice — whether an architectural decision in code or an agent approving a production deployment — can you see why?
This isn't hypothetical. CrowdStrike researchers found that AI-generated code vulnerability rates jump nearly 50% under certain conditions. Without a record of what AI wrote, you can't find affected code when threats like this emerge.
Right now, most teams can't answer any of these. AI-generated code blends invisibly into your codebase, and agent decisions vanish the moment they're made — no trail, no accountability. That's a problem that grows every day you ship code and run operations without tracking what AI did.
VIBES is an open standard that tracks AI involvement — both code generation and agent decision-making. Think of it like a flight recorder for everything AI does in your workflow.
When you use an AI tool — Claude Code, Gemini, Cursor, Copilot, or autonomous agents — VIBES records what it did, what it was asked, and how it reasoned. Code it wrote, decisions it made, actions it took. That data lives in a .ai-audit/ folder right alongside your code, tracked in git like everything else.
No cloud service required. No vendor lock-in. Just structured, portable data about every AI action — yours to keep, query, and act on.
VIBES gives you three levels of detail. Start simple and increase as your needs grow.
Like a receipt
Which AI tool, which model, and what action was taken — whether it wrote code or made a decision. The minimum viable audit trail, lightweight enough to always leave on.
Like a work order
Everything in Basic, plus what it was asked to do and what context it had. Reproduce results, audit prompts, and trace agent inputs.
Like showing its work
Everything in Detailed, plus the AI's full chain-of-thought reasoning. For security-critical code, high-stakes agent decisions, and when you need to understand why.
VIBES is the foundation. Three companion standards extend it to cover security, risk, and AI agent governance.
Want the full technical details? Explore the Technical track →
VIBES is the foundation, and every extension builds on the same data — so you can start simple and add capabilities as you need them. Adopt just VIBES for basic tracking, add VERIFY when you need tamper-proof records, layer on PRISM for risk scoring, or bring in EVOLVE for agent governance. Each one works independently, but combining them creates compounding value — and the ecosystem is open for future extensions.
Ready to know what AI did? Three ways to start today.
Add VIBES tracking to your AI tool in minutes. Works with Claude Code, Gemini, Codex, and more.
Install for your tool →Ask your AI tool provider if they support VIBES. The more people who ask, the faster this becomes standard.
How to ask →Full VIBES ecosystem — VIBES, VERIFY, PRISM, and EVOLVE — built in and ready to go.
runmaestro.ai →Projects across the community are already adopting VIBES to track AI code generation and agent decisions. See who's on board and join them.
Browse projects that have adopted VIBES transparency.
Explore the Registry →