Tokens saved 0 0 0 0 0 0 0 0 0

Vibe Rails

An opinionated framework that keeps vibe coding from going off the rails

Run the install command for your platform
Windows
irm https://raw.githubusercontent.com/robstokes857/vibe-rails/main/Scripts/install.ps1 | iex
Linux
wget -qO- https://raw.githubusercontent.com/robstokes857/vibe-rails/main/Scripts/install.sh | bash
Mac
curl -fsSL https://raw.githubusercontent.com/robstokes857/vibe-rails/main/Scripts/install.sh | bash
GitHub
github.com/robstokes857/vibe-rails
Works with your stack

Multiple LLMs and Code Editors

Claude

Codex

Z.AI

Grok

Copilot

AntiGravity

OpenCode

VS Code

Tokens Saved

tokens

Credit where it's due: we got this idea from rtk

Stay on track

Key Features

Run AI Tasks in Parallel

Give each model its own task and sandbox. Claude can plan while Codex refactors and AntiGravity runs tests, all without touching the same working files.

  • Keep every task in a separate sandbox
  • Run different models and tasks at the same time
  • Keep files, prompts, and branch state separate
  • Compare the results before merging into main
3x
Concurrent model lanes
Faster
Shorter cycle time than serial AI runs
Safer
No agents overwriting each other
Parallel execution board live
Claude lane Architecture + planning Running
Codex lane Refactor implementation Running
AntiGravity lane Tests + validation Running
Three focused agents working at the same time in one coordinated workflow.

Environment Isolation, Like Conda for LLMs

Give each AI workflow its own settings, permissions, and reusable profile.

  • Custom settings per environment: prompts, args, and tool behavior
  • Grant special permissions only where needed, keep sensitive flows locked down
  • Separate config directories and identity per environment
  • Clone proven profiles and reuse them across projects
Environment profiles managed
production_safe Strict rules, limited commands, protected paths Safe
refactor_lane Expanded tool access for deep code transformations Balanced
research_fast High-speed experimentation in isolated sandbox Fast
Like Conda, but for AI coding workflows and shared team defaults.

Remote Terminal Access

Let your AI write code while you're away from your desk. Open a secure, real-time terminal in your browser and connect to your local machine through an encrypted WebSocket relay.

  • API key authentication limits access to your sessions
  • Monitor your LLM's progress from your phone, tablet, or any browser
  • Use the interactive terminal to type commands at any time
  • Auto-synced terminal size so output renders perfectly
viberails: remote session
$ claude "refactor the auth module"
Claude is working...
Editing src/auth/middleware.ts
Editing src/auth/tokens.ts
Running npm test -- --watch=false
All 47 tests passed.
---
You watched this happen from your phone.

More Features

Use shared defaults, enforce project rules, and keep prompts focused.

Convention Over Configuration

Start with sensible defaults instead of configuring every project from scratch.

  • Use the same environment, terminal, and guardrail conventions in every project.
  • Get new contributors running without a long setup checklist.
  • Keep workflows consistent across repos and teams.

Spend more time shipping code and less time maintaining tool configuration.

Rule Enforcement

Define coding rules once and have the model fix violations before it moves on.

  • Set requirements for tests, complexity, naming, and reliability.
  • Choose WARN, COMMIT, or STOP for each rule.
  • Catch risky changes before review.

Checks run while the code is being written.

Token Savings

Send the model only the files and context it needs for the current task.

  • Prioritize the files that matter for each request.
  • Avoid resending unchanged files and project history.
  • Use fewer tokens and get responses sooner.

Smaller, focused prompts cost less and are easier to review.