Multi-Agent Development System

AI that multiplies
to build your software

One chat. Multiple specialized agents working in parallel. Smith orchestrates planning, implementation, testing, and review—all from your terminal.

smith
$smith
💬 You: Add user authentication with JWT
🏛️ The Architect: Breaking down into tasks...
🔑 The Keymaker: Creating auth middleware...
🦑 Sentinel: Writing test suite...
✓ Feature complete and tested

Smith meets you where you work

One system. Multiple interfaces. Your choice.

INTERACTIVE

Chat while agents work

Launch Smith in your terminal and have a natural conversation. Behind the scenes, specialized agents coordinate in the background—planning, implementing, testing. Watch their progress in the sidebar while you chat, with full control over safety levels and execution permissions.

Get started
# Start Smith
$ cd my-project
$ smith
# Chat while agents coordinate
💬 You: Add user authentication
🏛️ The Architect: Breaking into tasks...
🔑 The Keymaker: Writing code...
✓ Feature complete
REAL-TIME

See agents coordinate

Watch agents work in parallel while you chat. Each agent announces what it's doing, and you can see the coordination happening in real-time. File conflicts are prevented automatically—agents know what's locked and coordinate their work.

Learn more
# Agents coordinate through chat
💬 You: Add authentication to the API
🏛️ Planning: Breaking into 3 tasks...
🔑 Implementation: Creating auth middleware
🔑 Implementation: Acquired lock on auth.go
🦑 Testing: Waiting for auth.go...
✓ auth.go written (lock released)
🦑 Testing: Writing test suite...
ONE-SHOT

Execute and exit

Need to run a specific task without interaction? Use exec mode. Perfect for CI/CD pipelines, automated refactoring, or any scenario where you need agents to complete work and exit. Same multi-agent coordination, just non-interactive.

See examples
# Execute and exit
$ smith exec "migrate to Go 1.22"
🕶️ Analyzing codebase...
🕶️ Updating go.mod...
🕶️ Updating syntax...
✓ Migration complete

Built for developers who demand control

🕶️

Multi-Agent Architecture

Specialized agents for planning, implementation, testing, and code review work together seamlessly. Each agent is an expert in its domain.

🔒

Safety First

Configurable auto-levels (off, low, medium, high) ensure AI actions match your comfort level. You decide how autonomous agents can be.

Provider Agnostic

Works with GitHub Copilot, OpenRouter, and OpenAI. Use your preferred provider and models. No vendor lock-in.

🎯

Conflict-Free Execution

Advanced file locking system prevents agents from stepping on each other. Multiple agents can work in parallel safely.

📦

Project-Centric

Each project has its own configuration. No global defaults forcing choices on you. Customize per project.

🎨

Beautiful Interface

Modern terminal interface with streaming responses, real-time updates, and intuitive slash commands for power users.

Get started in seconds

No complex setup. No configuration files. Just install and start building.

macOS / Linux

curl -fsSL https://agentsmith.sh/install | bash

Quick Start

1
cd your-project

Navigate to your project directory

2
smith

Launch the interactive session

3
Start chatting with your AI team

Describe what you want to build

Ready to multiply your development velocity?

Join developers who are shipping faster with AI agents that actually understand their code. Stop context-switching between tools—let specialized agents handle planning, implementation, and testing while you focus on what to build next.