AI & LLM Workflows

Overview

Modern development increasingly involves AI assistants that write, test, and deploy code. ZeroDeploy is designed from the ground up to support these AI-first workflows:

• Zero-auth Drop API — Deploy with one HTTP call, no login required
• Machine-readable output — Every CLI command supports --json for structured output
• Predictable exit codes — Specific codes for auth, validation, rate limits, and billing errors
• Agent skills — Structured workflows for deploy, rollback, and status
• LLM discovery — llms.txt helps AI assistants understand our API
• OpenAPI spec — Full API specification for code generation
• CLI introspection — zerodeploy inspect describes commands in JSON

Drop API (Zero Auth)

Deploy with one HTTP call — no login, no CLI, no account. Ideal for AI agents that need to deploy immediately.

curl -X POST https://api.zerodeploy.dev/drop \
  -H "Content-Type: text/html" \
  --data-binary @index.html

Returns a live URL, claim token, and expiration. Full details: Drop API reference.

Agent Skills

Skills give AI agents step-by-step instructions for common tasks. When you say “deploy my site”, the agent:

1. Checks if the CLI is installed (installs if needed)
2. Checks if you’re logged in (opens browser for OAuth if needed)
3. Runs zerodeploy deploy --json for structured output
4. Parses the result and reports the deployment URL

All through the CLI — no direct API calls needed.

Install Skills

npx skills add zerodeploy-dev/skills

This installs skills to your project and works across Claude Code, Codex, Copilot, Cursor, and other agents.

Use --yes to skip prompts, or --global to install for all projects.

Available Skills

Deploy

Triggered by /zerodeploy-deploy or when the user asks to deploy, publish, ship, or host their project.

You: Deploy this site
Agent: Running zerodeploy deploy --json...
       Deployed! URL: https://my-site.zerodeploy.app

Options the agent can use:

• --build — Run build command before deploying
• --install — Install dependencies first
• --dir <path> — Specify the build output directory
• --preview — Create a preview deployment

Rollback

Triggered by /zerodeploy-rollback. Lists recent deployments and rolls back to a previous version.

You: Rollback my site to the previous version
Agent: Found 3 deployments. Rolling back to abc12345...
       Rolled back! Site is now serving the previous deployment.

Setup

Triggered by /zerodeploy-setup. Handles first-time setup:

1. Installs the CLI globally
2. Opens browser for GitHub authentication
3. Creates zerodeploy.json in the project root

Status

Auto-triggered when you ask about deployment status, traffic, or site health.

You: Is my site up?
Agent: Your site my-site.zerodeploy.app is live.
       Last deployed: 2 hours ago (deployment abc12345)
       Status: ready

Example Prompts

These prompts trigger agent skills:

• “Deploy this site to ZeroDeploy”
• “Ship my project”
• “Rollback to the previous deployment”
• “Is my site up?”
• “Show me deployment status”
• “Set up ZeroDeploy for this project”
• “Deploy with build”

Using with Claude Code

Claude Code can deploy your sites directly. Just ask it to deploy and it will use the ZeroDeploy CLI:

You: "Deploy this site to ZeroDeploy"

Claude: I'll deploy your site using the ZeroDeploy CLI.
        Running: zerodeploy deploy

        Deployment successful!
        URL: https://my-site.zerodeploy.app

Claude Code can also:

• Create and manage sites and organizations
• Set up custom domains
• Configure CI/CD with deploy tokens
• Check deployment status and rollback if needed
• View analytics and form submissions

Using with Cursor

Cursor’s AI can deploy your projects using the integrated terminal. The --json flag helps Cursor parse responses:

# Cursor can run these commands and parse the JSON output
zerodeploy deploy --json
zerodeploy site list --json
zerodeploy deployments list --json

Machine-Readable Output

All data-returning commands support --json for structured output that AI assistants can parse:

$ zerodeploy site list --json
{
  "sites": [
    {
      "id": "019b...",
      "slug": "my-site",
      "url": "https://my-site.zerodeploy.app"
    }
  ]
}

The deploy command returns structured JSON:

Success:

{
  "success": true,
  "deployment_id": "abc123",
  "url": "https://my-site.zerodeploy.app",
  "preview_url": "https://abc123-my-site.zerodeploy.app",
  "is_preview": false,
  "file_count": 42,
  "total_size_bytes": 1048576,
  "site": "my-site"
}

Error:

{
  "success": false,
  "error": "not_authenticated",
  "message": "Not logged in. Run: zerodeploy login",
  "exit_code": 1
}

Structured Exit Codes

The CLI uses specific exit codes so AI assistants can handle errors programmatically:

CLI Introspection

The zerodeploy inspect command outputs CLI metadata in JSON, helping AI assistants discover available commands:

$ zerodeploy inspect
{
  "name": "zerodeploy",
  "version": "1.0.0",
  "commands": [
    {
      "name": "deploy",
      "description": "Deploy a site",
      "options": ["--org", "--site", "--dir", "--build", "--json"]
    },
    ...
  ]
}

# Inspect a specific command
$ zerodeploy inspect "site create"
{
  "name": "site create",
  "usage": "zerodeploy site create <slug> [--org <org>]",
  "description": "Create a new site",
  "arguments": [...],
  "options": [...]
}

LLM Discovery (llms.txt)

We provide llms.txt files that help AI assistants understand ZeroDeploy:

• zerodeploy.dev/llms.txt — Quick summary with links
• zerodeploy.dev/llms-full.txt — Complete documentation in one file
• api.zerodeploy.dev/openapi.json — Full OpenAPI 3.0 specification

AI assistants can fetch these files to learn about ZeroDeploy’s capabilities, CLI commands, and API endpoints without needing prior training data.

OpenAPI Specification

The full REST API is documented with OpenAPI 3.0, enabling AI assistants to:

• Generate API client code
• Understand request/response schemas
• Make direct API calls when needed

# Fetch the OpenAPI spec
curl https://api.zerodeploy.dev/openapi.json

Example: Full AI Workflow

For quick one-off deploys, use the Drop API. For ongoing projects with CI/CD, custom domains, and rollbacks, use the CLI:

# 1. Check if logged in
zerodeploy whoami --json
# Exit code 1? Run: zerodeploy login

# 2. Build the project
npm run build

# 3. Deploy
zerodeploy deploy --json
# Parse JSON for deployment URL

# 4. If deploy fails with exit code 7 (billing)
zerodeploy billing --json
# Inform user about balance

# 5. Verify deployment
curl -s https://my-site.zerodeploy.app | head -20

Best Practices for AI Assistants

• Use Drop for quick deploys — one HTTP call, no auth, no setup
• Use CLI for ongoing projects — custom domains, rollbacks, analytics
• Always use --json when parsing CLI output programmatically
• Check exit codes to handle errors appropriately
• Use inspect to discover command syntax
• Fetch llms.txt for up-to-date documentation