Creating a Template

Templates are your product listing on the AgentDeploy marketplace. This guide covers the SDK-based workflow for creating, publishing, and managing templates.

SDK-first workflow

ad init  →  build agent  →  ad deploy  →  create template listing  →  publish

With the SDK, your agent code and template configuration live in the same project:

my-agent/
├── .agentdeploy/
│   └── config.json       # Template metadata, secrets, model config
├── src/
│   ├── server.ts          # Agent entry point
│   └── tools.ts           # Tool definitions
├── package.json
├── tsconfig.json
└── wrangler.jsonc

Step 1: Configure your project

Edit .agentdeploy/config.json:

{
  "projectId": "customer-support-bot",
  "name": "Customer Support Bot",
  "template": "chat-agent",
  "entrypoint": "src/server.ts",
  "model": "openai/gpt-4o-mini",
  "secrets": [
    {
      "key": "KNOWLEDGE_BASE_API_KEY",
      "required": true,
      "description": "API key for the knowledge base service"
    },
    {
      "key": "TICKETING_WEBHOOK",
      "required": false,
      "description": "Webhook URL for ticket escalation"
    }
  ]
}

Step 2: Deploy

ad deploy

The CLI bundles your TypeScript, uploads it, validates, and publishes a new edge version.

Step 3: Create marketplace listing

After your first deploy, go to Dashboard > Seller > Templates > Create Template:

Basic info

Field	Description	Example
Name	Display name in the marketplace	Customer Support Bot
Agent ID	URL-safe identifier	customer-support-bot
Description	Short description for listings	AI-powered support agent with RAG and ticket escalation
Price	One-time purchase price (USD)	9.99
Category	Marketplace category	Productivity, Development, Business
Documentation URL	Link to your docs	https://github.com/you/repo#readme

Deployment configuration

Field	Value	Description
Deploy Runtime	edge	Agent runs as a Cloudflare Worker (recommended)
Hosting Compatible	true	Buyers can deploy on AgentDeploy hosting

When using the SDK, your agent is always edge runtime with nodejs_compat flag. The platform auto-detects Durable Object classes and creates appropriate bindings.

Supported LLM providers

Select which LLM providers your agent supports. All providers use OpenAI-compatible endpoints via the platform gateway:

Provider	Example model
OpenRouter	openai/gpt-4o-mini
Anthropic	anthropic/claude-3-sonnet
Google	google/gemini-flash
Groq	groq/llama-3.1-70b

Features

Add feature tags based on your agent's capabilities:

• streaming — Uses createChatAgent() with WebSocket streaming
• stateful — Uses createAgent() with SQLite storage
• scheduled — Uses scheduled tasks (scheduleEvery, schedule)
• tools — Uses defineTool() for tool calling
• mcp — Connects external MCP servers
• multi-agent — Multiple agent classes in one worker

Step 4: Security scan

Connect a GitHub repo for automated scanning, or rely on the platform's built-in code validation during deploy:

GitHub-connected scanning

1. Click Connect GitHub App and authorize
2. Select your repository
3. Click Scan Repository — runs gitleaks, semgrep, OSV, prompt injection, permission analysis
4. Need a C rating or higher to publish

CLI-only validation

• The platform validates your bundled script during ad deploy
• Checks for dangerous patterns (eval(), process.env, etc.)
• Verifies Durable Object class declarations are valid
• Validates script size (3MB limit for SDK-bundled agents)

Step 5: Publish

Once scanning passes:

1. Go to your template's Edge tab
2. Your latest deployed version appears
3. Click Publish
4. Your template is live on the marketplace

Template visibility

State	Visible?	Buyers can purchase?
Draft	No	No
Published (no scan)	No	No
Published (scan C+)	Yes	Yes
Published (scan A/B)	Yes, with badge	Yes
Yanked	No	Existing deployments unaffected

Updating your agent

To push an update:

# Make changes to your code, then:
ad deploy --version 1.1.0

The CLI creates a new edge version. New deployments automatically use the latest version. Existing deployments stay on their current version until the buyer updates.

Managing multiple agents

If you have multiple agent templates:

# Each project has its own .agentdeploy/config.json
~/projects/support-bot/     # Template: customer-support
~/projects/sales-agent/     # Template: sales-assistant
~/projects/billing-bot/     # Template: billing-specialist

Each project deploys independently to its own template on the marketplace.