A terminal-based multi-agent coding assistant where AI agents collaborate, argue, and brainstorm together to help you code.
-
Multiple Specialized Agents: Each with a unique perspective
- 💻 The Coder: Writes clean, efficient code
- 🔍 The Critic: Reviews code for bugs, security issues, and best practices
- 🐛 The Debugger: Expert at finding and fixing bugs
- 🏗️ The Architect: Designs system architecture and structure
- 👤 The End User: Represents real user perspective, asks questions others might miss
-
Collaborative Discussion: All agents see the entire conversation and can brainstorm together
-
Review Cycles: Iterative code review process where agents work together to improve code
-
BYOK (Bring Your Own Key): Use your own API key with multiple providers
# Clone the repository
git clone https://github.com/Githubguy132010/devvy.git
cd devvy
# Install dependencies
bun install
# Build the project
bun run build
# Link globally (optional)
bun link devvyOn first run, Devvy will guide you through an interactive setup wizard:
devvyThe setup wizard will help you:
- Select your AI provider (OpenAI, Anthropic, OpenRouter, or custom)
- Enter your API key
- Choose your default model
You can also run the setup wizard anytime with:
devvy setup- OpenAI - GPT-4, GPT-4o, GPT-4-turbo
- Anthropic - Claude 3.5, Claude 3
- OpenRouter - Access to 100+ models from various providers
- Custom - Any OpenAI-compatible API endpoint
devvy setup# Set your API key
devvy config set-key your-api-key-here
# Set provider (openai, anthropic, openrouter, custom)
devvy config set-provider openrouter
# Select model interactively (fetches available models from API)
devvy config set-model
# Or set model directly
devvy config set-model gpt-4o
# For custom providers, set the base URL
devvy config set-base-url https://your-api-endpoint.com/v1
# View current configuration
devvy config showYou can also use environment variables:
export OPENAI_API_KEY=your-key # For OpenAI
export ANTHROPIC_API_KEY=your-key # For Anthropic
export OPENROUTER_API_KEY=your-key # For OpenRouterdevvy chat
# or just
devvy@coder Write a function to calculate fibonacci numbers
@critic Please review the code above
@debugger I'm getting an error: TypeError: undefined is not a function
@architect Design a REST API for a todo app
@enduser What questions would a user have about this feature?
@review - Start a review cycle (Critic reviews, agents fix issues)
@brainstorm - All agents collaborate on a topic
/model - Select AI model (scrollable list with arrow keys)
/config - Show current configuration
/clear - Clear conversation history
/history - Show conversation history
/help - Show help
/exit - Exit Devvy
- You describe a task → The Architect provides a design
- The Coder implements → Writes the code based on the design
- The Critic reviews → Checks for bugs, security issues, best practices
- The End User evaluates → Asks questions from a user perspective
- Iteration → Agents discuss and fix issues until the code is approved
The unique aspect is that all agents can see the entire conversation and respond to each other, creating a more human-like collaborative discussion.
You> @architect Design a simple REST API for managing tasks
🏗️ [Architect]
I'll design a simple REST API for task management...
[provides architecture]
You> @coder Implement the API based on the architecture
💻 [Coder]
Here's the implementation...
[provides code]
You> @review
🔍 [Critic]
Reviewing the code...
NEEDS CHANGES: Missing error handling for invalid task IDs...
👤 [End User]
What happens if a user tries to update a task that doesn't exist?
💻 [Coder]
Good points! Here's the updated implementation with proper error handling...
🔍 [Critic]
APPROVED: The code now handles all edge cases properly.
Currently, all conversation history is kept in memory during a session. Future improvements may include:
- Persistent conversation storage
- Context window management for long conversations
- Summarization for older messages
GPL-3.0 - See LICENSE for details.
Contributions are welcome! Please feel free to submit issues and pull requests.