ReVibe is a high-performance command-line coding assistant powered by a wide range of Large Language Models. It provides a conversational interface to your codebase, enabling you to explore, refactor, and build complex features through natural language and a robust set of autonomous tools.

✨ Key Features

• 🚀 Multi-Provider Ecosystem: Support for OpenAI, Mistral, Qwen, Cerebras, Groq, HuggingFace, Ollama, and LlamaCPP.
• 🔄 Hot-Swapping: Switch providers and models instantly mid-session with /provider and /model.
• 🎨 Modern TUI: A polished, minimal interface inspired by leading AI coding tools for zero distraction.
• 🛠️ Autonomous Toolset: File system operations, advanced code search, git integration, and command execution.
• 🛡️ Safe by Design: Granular tool permissions with an interactive approval system.
• 🧩 MCP Support: Extend capabilities with Model Context Protocol servers.

🚀 Installation

Using uv (Recommended)

uv tool install revibe

Using pip

pip install revibe

From Source

git clone https://github.com/OEvortex/revibe.git
cd revibe
uv sync --all-extras
uv run revibe

🛠️ Setup

Quick Start

1. Navigate to your project directory.
2. Run revibe to start the onboarding process.
3. ReVibe will automatically create your configuration at ~/.revibe/config.toml and prompt for necessary API keys.

🔑 Authentication & Environment Variables

ReVibe manages API keys in ~/.revibe/.env. You can also set them directly in your shell.

[!TIP] For Qwen, install qwen-code if not installed: npm install -g @qwen-code/qwen-code@latest, then use /auth in qwen to authenticate, then you can close qwen and use qwencode provider in ReVibe.

📖 Usage

💬 Interactive Mode

Simply run revibe to enter the interactive TUI.

• Multi-line Input: Use Ctrl+J or Shift+Enter for newlines.
• File Referencing: Type @ to trigger fuzzy path autocompletion.
• Direct Commands: Prefix with ! to execute shell commands (e.g., !npm test).

🤖 Programmatic Mode

Execute single prompts directly from your shell:

revibe --prompt "Explain the logic in @revibe/core/agent.py"

⚡ Slash Commands

⚙️ Configuration

ReVibe uses TOML for configuration. It checks ./.revibe/config.toml first, then falls back to ~/.revibe/config.toml.

[[mcp_servers]]
name = "fetch_server"
transport = "stdio"
command = "uvx"
args = ["mcp-server-fetch"]

[[mcp_servers]]
name = "github"
transport = "http"
url = "https://mcp-github-server.com"
api_key_env = "GITHUB_TOKEN"

active_model = "gpt-4o"
system_prompt_id = "architect"
disabled_tools = ["bash"]

[tools.read_file]
permission = "always"

task(agent="explore", prompt="Find where agent profiles are loaded and explain the flow.")

agent_type = "subagent"
description = "Focused analysis agent"
system_prompt_id = "cli"

name = "analysis-model"
provider = "mistral"
alias = "analysis"
context = 123456
max_output = 32768
supported_formats = ["native", "xml"]

🖥️ Editor Integration

ReVibe supports the Agent Client Protocol (ACP), allowing it to act as a backend for compatible editors like Zed. See ACP Setup for instructions.

📄 License

Licensed under the Apache License, Version 2.0. See LICENSE for details.