Warden

Interactive CLI for bootstrapping Warden agents with support for both A2A and LangGraph protocols.

Getting Started

Install the CLI globally:

npm install -g warden-code

Then run it with:

warden

Or just run it directly:

npx warden-code

This launches an interactive CLI where you can create new agents.

Commands

Command	Description
/new [path]	Create a new agent interactively [optionally provide a path]
/build [path]	Enter AI-powered build mode to modify your agent via chat
/chat <url>	Chat with a running agent via A2A or LangGraph
/help	Show available commands
/clear	Clear the terminal
/exit	Exit the CLI

Creating an Agent

Run /new to start the agent creation wizard:

1. Agent name - a name for your agent
2. Description - what your agent does
3. Model - Echo (just a demo that echoes input) or OpenAI (GPT-powered)
4. Capability - Streaming or Multi-turn conversations
5. Skills - Define agent capabilities (optional)

After generation, your agent will be ready at src/agent.ts.

Models

Model	Description
Echo + Streaming	Minimal streaming agent that echoes input
Echo + Multi-turn	Minimal multi-turn conversation agent
OpenAI + Streaming	GPT-powered agent with streaming responses
OpenAI + Multi-turn	GPT-powered agent with conversation history

All options use AgentServer from @wardenprotocol/agent-kit, which exposes both:

• A2A Protocol
• LangGraph Protocol

Generated Project Structure

my-agent/
├── src/
│   ├── agent.ts      # Your agent logic (handler function)
│   └── server.ts     # Server setup and configuration
├── agent-card.json   # Agent identity, capabilities, and skills (A2A protocol)
├── package.json
├── tsconfig.json
├── Dockerfile
├── .env.example
└── .gitignore

Running Your Agent

cd my-agent
npm run build
npm run agent

Your agent will be available at http://localhost:3000.

Build Mode

Run /build inside a scaffolded project to enter an AI-powered chat session. Describe the changes you want and the LLM will modify your agent code directly. Requires an OpenAI or Anthropic API key (configured on first run).

While in build mode you can type /chat to talk to your running agent without leaving the session. The URL is resolved automatically from your project's .env (AGENT_URL), or you can pass it explicitly (e.g. /chat http://localhost:3000). Type /exit inside the chat sub-session to return to build mode.

Chatting with an Agent

Run /chat http://localhost:3000 to interact with a running agent. The CLI auto-detects whether the agent supports A2A, LangGraph, or both, and prompts you to choose when multiple protocols are available.

Hosting Your Agent

To host your agent on a server, you can use a cloud provider like AWS, Google Cloud, or Azure. You can also use a containerization platform like Render.com to deploy your agent as a Docker container.

License

Apache-2.0