svibe

Agentic CLI tool with hierarchical agents, multi-provider support, and MCP protocol

svibe is an intelligent command-line assistant that uses AI agents to help with development tasks. It coordinates multiple AI models, provides built-in development tools, and supports the Model Context Protocol (MCP) for extensibility.

Features

• Multi-Provider Support: OpenAI (cloud) and Ollama (local) with bearer/basic authentication
• Hierarchical Agents: Supervisor agent coordinates specialized sub-agents
• Smart Model Routing: Different models for different task types (chat, code, planning)
• Built-in Tools: File operations, shell commands, git operations
• MCP Protocol: Connect to external MCP servers for additional tools/resources
• Project Configs: Local .svibe/config.toml overrides global settings

Quick Start

Installation

# Build from source
git clone https://github.com/yourusername/svibe.git
cd svibe
cargo build --release

# Binary location
./target/release/svibe --help

Setup

# Initialize global configuration
svibe init

# Set your API key
export OPENAI_API_KEY=your-api-key

# Or initialize with local project config
svibe init --local

Usage

# Interactive REPL mode
svibe repl

# Run a single task
svibe run "Fix the bug in main.rs"

# Use specific agent
svibe -a code run "Write a function to parse JSON"

# Show status
svibe status

# List tools
svibe tools

Configuration

Config Locations

Type	Path	Priority
Global	~/.config/svibe/config.toml	1 (lowest)
Local	.svibe/config.toml	2 (overrides global)
Env Vars	OPENAI_API_KEY, OLLAMA_HOST	3 (highest)

Example Configuration

# ~/.config/svibe/config.toml

[providers.openai]
api_key = "sk-..."  # Or set OPENAI_API_KEY env var
base_url = "https://api.openai.com/v1"
default_model = "gpt-4o"

[providers.ollama]
base_url = "http://localhost:11434"
default_model = "llama3"
# Bearer token auth (for cloud-hosted Ollama-compatible APIs)
bearer_token = "your-token"
# OR Basic auth
# username = "user"
# password = "pass"

[agents.default]
description = "Default general-purpose agent"
system_prompt = "You are a helpful AI assistant."
tools = ["file", "shell", "git"]

[agents.default.models]
chat = "ollama/llama3"
code = "openai/gpt-4o"
planning = "openai/gpt-4o"
tool_use = "openai/gpt-4o"
default = "ollama/llama3"

Commands

Command	Description
svibe init	Initialize global configuration
svibe init --local	Create project-specific config
svibe repl	Start interactive REPL mode
svibe run "<task>"	Execute a single task
svibe status	Show agent and provider status
svibe tools	List available tools
svibe agents	List available agents
svibe health	Health check providers
svibe mcp list	List connected MCP servers
svibe mcp connect <name> <cmd>	Connect MCP server

REPL Mode

svibe repl

REPL commands:

• /help - Show help
• /clear - Clear screen
• /status - Show agent status
• /exit, /quit, :q - Exit REPL

Built-in Tools

File Tools

• read - Read file contents
• write - Write content to file
• edit - Edit file at specific line
• search - Search for pattern in file
• list - List directory contents

Shell Tools

• Execute shell commands with safety checks
• Blocks dangerous patterns (rm -rf, etc.)
• Configurable timeout

Git Tools

• status, diff, commit
• branch, checkout, merge, rebase
• log, add

MCP Support

Connect to Model Context Protocol servers for additional tools:

# List connected servers
svibe mcp list

# Connect a server
svibe mcp connect filesystem npx -y @modelcontextprotocol/server-filesystem /path/to/allow

# Disconnect
svibe mcp disconnect filesystem

Architecture

svibe/
├── src/
│   ├── main.rs           # CLI entry point
│   ├── config/           # Configuration system
│   ├── providers/        # AI model providers
│   ├── tools/            # Built-in tools
│   ├── mcp/              # MCP protocol
│   ├── agents/           # Agent system
│   └── cli/              # CLI interface

Environment Variables

Variable	Description	Default
OPENAI_API_KEY	OpenAI API key	-
OLLAMA_HOST	Ollama host URL	http://localhost:11434
RUST_LOG	Logging level	svibe=info

Development

# Build debug
cargo build

# Build release
cargo build --release

# Run with args
cargo run -- run "test task"

# Run tests
cargo test

# Format code
cargo fmt

# Lint
cargo clippy

License

MIT