Vibez

A multi-agent CLI coding assistant built in Rust that uses parallel workers for faster code generation.

Features

• 🚀 Parallel Tool Execution - Independent operations run concurrently for maximum speed
• 🤖 Multi-Agent Architecture - Manager coordinates multiple worker agents
• 🔒 Smart Conflict Detection - Automatic file locking prevents race conditions
• 📝 Contract-Based Coordination - Workers share interfaces through contracts
• ⚡ Streaming Responses - Real-time output as the AI generates code
• 🎯 Topological Task Scheduling - Dependency-aware task execution

Installation

From crates.io (recommended)

cargo install vibez

From source

git clone https://github.com/taradepan/vibez
cd vibez
cargo install --path .

Setup

On first run, Vibez will prompt you for your OpenAI API key:

vibez

🔑 OpenAI API Key Required
Vibez needs your OpenAI API key to function.
Get your API key from: https://platform.openai.com/api-keys

Enter your OpenAI API key: sk-...

💾 Saving API key to ~/.vibez/config.toml...
✓ API key saved to /Users/you/.vibez/config.toml

Your API key is securely stored in ~/.vibez/config.toml and will be used for all future runs.

Usage

Interactive mode (REPL)

vibez

Single prompt mode

vibez "Add authentication to my app"

Commands

• /help - Show available commands
• /config - Show current configuration
• /plan - Enable plan mode for complex tasks (uses multi-agent architecture)
• /clear - Clear screen
• /quit - Exit

Options

vibez --help

Configuration

Configuration Priority

Settings are loaded in this order (highest priority last):

1. Default values
2. User config file (~/.vibez/config.toml)
3. Project config file (./.vibez/config.toml)
4. Environment variables (VIBEZ_MODEL, VIBEZ_VERBOSE, VIBEZ_API_TIMEOUT)
5. CLI arguments (--model, --verbose)

Note: OpenAI API key is NOT loaded from environment variables for security. It must be in a config file.

Example Config File

~/.vibez/config.toml:

model = "gpt-5.1"
verbose = false

[api]
openai_api_key = "sk-..."
timeout = 60

Architecture

Multi-Agent System

Vibez features two modes of operation:

Single-Agent Mode (Default)

For straightforward tasks, Vibez uses a single agent that handles requests directly with intelligent tool execution.

Multi-Agent Mode (/plan)

For complex tasks, use /plan to activate the multi-agent architecture:

1. Manager Agent - Analyzes the task, breaks it down into subtasks, creates coordination contracts, and orchestrates worker agents
2. Worker Agents - Execute assigned subtasks in parallel, adhering to contracts and interfaces defined by the manager
3. Contract System - Workers share code interfaces through contracts, enabling parallel development without conflicts
4. Dependency Resolution - Topological sorting ensures tasks execute in the correct order based on dependencies
5. File Locking - Automatic detection and prevention of concurrent file access conflicts
6. Smart Grouping - Graph-based algorithm groups independent operations for parallel execution

The multi-agent system excels at:

• Large refactoring tasks spanning multiple files
• Feature implementations requiring coordination across components
• Complex workflows with multiple independent subtasks
• Parallel code generation across different modules

Core Components

• Tool Registry - Provides file operations (read, write, edit), bash commands, and API interactions
• Streaming Output - Real-time display of AI responses and progress
• Rate Limiting - Semaphore-based control to manage API request concurrency

Performance

Vibez is designed for speed:

• Parallel Tool Execution: Independent file reads/writes run concurrently
• Conflict Detection: Graph-based algorithm groups non-conflicting operations
• Topological Scheduling: Tasks execute as soon as dependencies are met
• Rate Limiting: Semaphore-based API request management

License

MIT - see LICENSE file for details