Nut.new: Time-Travel Debugging AI for Vibe Coding

TL;DR

TL;DR: Nut.new is an open source vibe coding tool by Replay.io that records your app’s execution, then uses AI with time-travel debugging data to fix bugs that normally stump LLMs — free to use, built on Bolt.DIY.

Source and Accuracy Notes

• Project page: nut.new
• Source repository: github.com/replayio/devtools (718 stars)
• Technical blog post: blog.replay.io/the-nut-api
• License: Custom (see replayio/devtools LICENSE)
• HN launch thread: news.ycombinator.com/item?id=43258585
• Source last checked: 2026-06-16

What Is Nut.new?

Nut.new is a vibe coding platform built by Replay.io that combines their time-travel debugging technology with AI to solve the biggest weakness of AI code generation: debugging.

As Replay.io CEO Brian Hackett explained in the Show HN post:

AIs are really good at writing code but really bad at debugging — it’s amazing to use Claude to prompt an app into existence, and pretty frustrating when that app doesn’t work right and Claude is all thumbs fixing the problem.

The core insight: when an AI coding assistant gets stuck on a bug, it’s typically only looking at source code. Nut gives the AI access to runtime behavior — what actually happened when the app ran — through Replay’s recording technology.

How It Works

The Recording Pipeline

1. Record execution — Nut captures everything that happens when your app runs (DOM changes, network requests, console output, React state, etc.) using Replay’s browser or simulation tooling
2. AI explores with tools — Claude gets specialized tools to query the recording like a database: inspect console messages, trace React component trees, follow control dependencies and dataflow
3. Precise explanations — Instead of guessing from source code alone, the AI produces accurate root-cause explanations that the coding LLM can use to write correct fixes

The Nut API

Under the hood, Nut exposes a chat API for querying recordings:

[
  {
    "role": "system",
    "type": "text",
    "content": "The following user message describes a bug..."
  },
  {
    "role": "user",
    "type": "text",
    "content": "app goes blank getting directions"
  }
]

The API returns precise explanations of what went wrong at runtime — context that isn’t apparent from source code alone.

Setup Workflow

Step 1: Use Nut.new Directly (No Install Required)

Nut.new is a hosted web application. Visit nut.new and start prompting:

1. Go to https://nut.new
2. Describe the app you want to build
3. Nut generates the code using Bolt.DIY
4. If bugs appear, Nut automatically records execution and debugs

Step 2: Add Your Own API Key (Optional)

Nut is completely free with limited uses. For unlimited access:

1. Open Nut.new settings
2. Add your own Anthropic API key
3. Use Nut without usage limits

Step 3: Import Existing Projects

To debug apps built elsewhere:

1. Click "Import" in Nut.new
2. Upload your project files
3. Describe the bug you're encountering
4. Nut records the execution and explains the root cause

Step 4: Use the Nut API Directly

For integration into your own tools:

# Contact Replay.io for API access
# Email: [email protected]
# Or use the contact form at replay.io

Deeper Analysis

What Makes Nut Different from Bolt.new / V0 / Lovable

Nut is a fork of Bolt.DIY (the open source Bolt.new fork). The key difference:

| Feature | Bolt.new / V0 | Nut.new | |---------|---------------|---------| | Code generation | Yes | Yes (via Bolt.DIY) | | Runtime recording | No | Yes (Replay recordings) | | AI debugging context | Source code only | Source + runtime behavior | | Bug fix reliability | Gets stuck frequently | Resolves via execution data | | Open source | Partially | Yes |

Bug Categories Nut Handles Well

From the launch examples, Nut excels at:

• Exception handling bugs — e.g., exception thrown from a catch block unmounts the entire app
• Conditional rendering issues — e.g., modal component not always created on certain routes
• CSS/layout problems — e.g., icon sizing constrained by parent elements
• State initialization bugs — e.g., loading never finishes due to missing responsive UI state
• React performance issues — e.g., infinite rendering loop from missing useCallback

Limitations

• Currently best suited for frontend applications; full-stack features are in development
• Recording fidelity depends on the method: Replay browser captures everything, simulation may miss timing-sensitive bugs
• Chat links are not yet shareable (planned)
• The tool is still early-stage and most effective on problems similar to what it has seen before

Practical Evaluation Checklist

Use this to decide if Nut.new fits your workflow:

• [ ] You build frontends primarily with React/Next.js
• [ ] You frequently get stuck on bugs when using AI coding assistants
• [ ] You want a free alternative to Bolt.new with better debugging
• [ ] You’re comfortable with an early-stage open source tool
• [ ] You can provide your own API key for unlimited usage
• [ ] You don’t need full-stack (backend + database) features yet

Security Notes

• Nut.new is hosted by Replay.io — your code runs on their infrastructure
• Adding your own Anthropic API key means API calls go directly to Anthropic (not through Replay)
• Recordings are stored on Replay’s servers; review their privacy policy before using with sensitive code
• The open source Bolt.DIY base means the frontend code is auditable

FAQ

Q: Is Nut.new really free? A: Yes. You get free uses by default, or add your own Anthropic API key for unlimited access. Replay.io is also offering unlimited free access for users who provide feedback.

Q: How is Nut different from Bolt.new? A: Nut is a fork of Bolt.DIY (open source Bolt.new fork) with Replay’s time-travel debugging integrated. When the AI gets stuck on a bug, Nut records the app’s execution and gives the AI runtime context to produce accurate fixes.

Q: What AI models does Nut use? A: Nut uses Claude (Anthropic) for both code generation and debugging. You can add your own API key to use your own Anthropic account.

Q: Can I use Nut for backend/full-stack apps? A: Currently Nut is best suited for frontend applications. Full-stack features including database integration are planned for future releases.

Q: Is Nut.new open source? A: Yes. Nut is built on Bolt.DIY which is open source. The underlying Replay recording technology is developed at github.com/replayio/devtools.

Q: How do recordings work? A: Nut captures your app’s execution using either the Replay browser (a Chrome replacement with perfect recording fidelity) or simulation (lower fidelity but works in any environment). The recording becomes a queryable database the AI explores to understand runtime behavior.

Conclusion

Nut.new addresses the most frustrating part of vibe coding: when AI generates code that doesn’t work and can’t figure out why. By giving AI access to actual runtime behavior through time-travel recordings, Nut provides the debugging context that source code alone cannot offer.

For developers and non-developers alike who build with AI, Nut.new represents a meaningful step toward reliable AI-assisted development. The tool is free, open source, and backed by Replay.io’s years of work on web debugging infrastructure.

If you’ve ever been stuck in a loop of “fix this bug” → “that didn’t work” → “fix it again” with your AI coding assistant, Nut.new is worth trying.