Triggers - Ensemble Edge

Overview

Triggers define how ensembles are invoked. Conductor supports nine trigger types, all configured using the unified trigger: array in your ensemble YAML:

HTTP - Full web routing with path params, CORS, rate limiting, HTML/JSON responses
Webhook - Simple HTTP endpoints for external integrations
MCP - Model Context Protocol tool exposure
Email - Email routing and processing
Queue - Cloudflare Queues message processing
Cron - Scheduled execution with cron expressions
Build - Static generation at build time
CLI - Custom developer commands
Startup - Execute on Worker cold start (initialization)

All triggers use the same configuration pattern:

name: my-ensemble

trigger:
  - type: <trigger-type>
    # trigger-specific configuration

HTTP Triggers

Full web routing with path parameters, CORS, rate limiting, authentication, and HTML or JSON responses. Use HTTP triggers for building APIs, web pages, and complex web applications.

Basic JSON API

name: users-api
trigger:
  - type: http
    path: /api/users/:id
    methods: [GET]
    public: true
    responses:
      json: {enabled: true}

flow:
  - agent: fetch-user
    input: {userId: ${input.params.id}}

agents:
  - name: fetch-user
    operation: data
    config:
      backend: d1
      binding: DB
      query: "SELECT * FROM users WHERE id = ?"
      params: [${input.userId}]

outputs:
  user: ${fetch-user.output[0]}

Access: GET /api/users/123 → Returns JSON

Server-Rendered HTML Page

name: blog-post
trigger:
  - type: http
    path: /blog/:slug
    methods: [GET]
    public: true
    responses:
      html: {enabled: true}
    templateEngine: liquid

flow:
  - agent: fetch-post
    input: {slug: ${input.params.slug}}

  - operation: html
    config:
      template: |
        <!DOCTYPE html>
        <html>
        <head><title>{{ fetch-post.title }}</title></head>
        <body>
          <article>
            <h1>{{ fetch-post.title }}</h1>
            <div>{{ fetch-post.content }}</div>
          </article>
        </body>
        </html>
      data: ${fetch-post}

agents:
  - name: fetch-post
    operation: data
    config:
      backend: d1
      binding: DB
      query: "SELECT * FROM posts WHERE slug = ?"
      params: [${input.slug}]

outputs:
  html: ${html.output}

Access: GET /blog/my-post → Returns HTML page

HTTP with Authentication & Rate Limiting

trigger:
  - type: http
    path: /api/chat
    methods: [POST]
    auth:
      type: bearer
      secret: ${env.API_KEY}
    rateLimit:
      requests: 10
      window: 60  # 10 requests per minute
    cors:
      origin: "https://myapp.com"
      credentials: true
    responses:
      json: {enabled: true}

HTTP Request Context

HTTP triggers automatically parse request data and make it available to your ensemble:

Field	Description
`input.method`	HTTP method (GET, POST, etc.)
`input.params`	Path parameters (`/users/:id` → `input.params.id`)
`input.query`	Query string parameters
`input.headers`	Request headers
`input.body`	Request body (parsed JSON)
`input.cookies`	Parsed cookies from Cookie header

Cookie Access:

agents:
  - name: check-session
    condition: ${input.cookies.session_id}
    operation: storage
    config:
      type: kv
      action: get
      key: session-${input.cookies.session_id}

To set cookies in responses, use the cookies operation:

agents:
  - name: create-session
    operation: cookies
    config:
      action: set
      name: session_id
      value: ${generate-id.output}
      httpOnly: true
      secure: true
      sameSite: strict
      maxAge: 86400

The cookies operation integrates with Location Context for GDPR/CCPA consent-aware cookie management.

HTTP vs Webhook

Feature	HTTP	Webhook
Path params	✅ `/users/:id`	❌ No
Multiple methods	✅ GET, POST, etc.	✅ Yes
Rate limiting	✅ Yes	❌ No
CORS	✅ Yes	❌ No
HTML rendering	✅ Yes	❌ No
Use case	Web apps, APIs	Simple webhooks

Rule of thumb: Use http for web routing and pages. Use webhook for receiving webhooks from external services.

Multi-Path HTTP Triggers

Handle multiple related endpoints in a single ensemble using the paths array. This allows one ensemble to serve multiple routes with different HTTP methods and path parameters.

name: users-api

trigger:
  - type: http
    paths:
      - path: /api/v1/users
        methods: [GET, POST]
      - path: /api/v1/users/:id
        methods: [GET, PUT, DELETE]
    public: true

flow:
  - operation: code
    config:
      handler: |
        const { method, params } = context.input

        if (method === 'GET' && params.id) {
          // GET /api/v1/users/:id - Fetch single user
          return { action: 'fetch-user', userId: params.id }
        } else if (method === 'GET') {
          // GET /api/v1/users - List users
          return { action: 'list-users' }
        } else if (method === 'POST') {
          // POST /api/v1/users - Create user
          return { action: 'create-user', data: context.input.body }
        } else if (method === 'PUT' && params.id) {
          // PUT /api/v1/users/:id - Update user
          return { action: 'update-user', userId: params.id, data: context.input.body }
        } else if (method === 'DELETE' && params.id) {
          // DELETE /api/v1/users/:id - Delete user
          return { action: 'delete-user', userId: params.id }
        }

  - agent: users-handler
    input: ${code.output}

agents:
  - name: users-handler
    operation: data
    config:
      backend: d1
      binding: DB
      query: ${input.action === 'fetch-user' ? 'SELECT * FROM users WHERE id = ?' :
              input.action === 'list-users' ? 'SELECT * FROM users' :
              input.action === 'create-user' ? 'INSERT INTO users (name, email) VALUES (?, ?)' :
              input.action === 'update-user' ? 'UPDATE users SET name = ?, email = ? WHERE id = ?' :
              'DELETE FROM users WHERE id = ?'}

outputs:
  result: ${users-handler.output}

Benefits of Multi-Path Triggers:

Organize related endpoints in one ensemble
Share authentication and middleware across paths
Reduce configuration duplication
Keep related business logic together
Support RESTful API patterns naturally

Path Parameters:

Use :param syntax for dynamic segments (e.g., /users/:id, /posts/:slug)
Access via ${input.params.id}, ${input.params.slug}, etc.
Works with any HTTP method

Example: Blog API

trigger:
  - type: http
    paths:
      - path: /blog
        methods: [GET]
      - path: /blog/:slug
        methods: [GET]
      - path: /blog/:slug/comments
        methods: [GET, POST]
    public: true
    rateLimit:
      requests: 100
      window: 60

This single ensemble handles:

GET /blog - List all posts
GET /blog/:slug - View single post
GET /blog/:slug/comments - List comments
POST /blog/:slug/comments - Add comment

Complex Website Structure

For full-blown websites with sitemaps, robots.txt, dynamic pages, etc., organize ensembles by route:

ensembles/
├── pages/
│   ├── home.yaml              # GET /
│   ├── about.yaml             # GET /about
│   ├── blog-list.yaml         # GET /blog
│   ├── blog-post.yaml         # GET /blog/:slug
│   ├── user-dashboard.yaml    # GET /dashboard/:userId
│   └── contact-form.yaml      # GET /contact, POST /contact
├── api/
│   ├── users.yaml             # GET /api/users/:id
│   ├── posts.yaml             # GET /api/posts, POST /api/posts
│   └── search.yaml            # GET /api/search
├── static/
│   ├── robots.yaml            # GET /robots.txt
│   ├── sitemap.yaml           # GET /sitemap.xml
│   └── health.yaml            # GET /health
└── auth/
    ├── login.yaml             # POST /auth/login
    └── logout.yaml            # POST /auth/logout

Each file is an ensemble with trigger: {type: http}: Example: ensembles/static/robots.yaml

name: robots-txt
trigger:
  - type: http
    path: /robots.txt
    methods: [GET]
    public: true

flow:
  - operation: code
    handler: |
      return {
        output: `User-agent: *
Allow: /
Sitemap: https://yoursite.com/sitemap.xml`
      }

outputs:
  content: ${code.output}

Example: ensembles/static/sitemap.yaml

name: sitemap-xml
trigger:
  - type: http
    path: /sitemap.xml
    methods: [GET]
    public: true

flow:
  - agent: fetch-all-posts
    operation: data
    config:
      backend: d1
      query: "SELECT slug, updated_at FROM posts WHERE status = 'published'"

  - operation: code
    handler: |
      const urls = input.fetchAllPosts.map(post =>
        `<url>
          <loc>https://yoursite.com/blog/${post.slug}</loc>
          <lastmod>${post.updated_at}</lastmod>
        </url>`
      ).join('\n')

      return {
        output: `<?xml version="1.0" encoding="UTF-8"?>
<urlset xmlns="http://www.sitemaps.org/schemas/sitemap/0.9">
  ${urls}
</urlset>`
      }

outputs:
  xml: ${code.output}

This approach gives you:

✅ Full control over every route
✅ Each route is testable independently
✅ Easy to add auth, rate limiting per route
✅ Auto-discovery finds all ensembles
✅ SEO-friendly (sitemaps, robots.txt)
✅ Dynamic content from database
✅ AI-powered pages via think agents

Webhook Triggers

Expose ensembles as HTTP endpoints for external services.

You own your webhook paths. You can define any path you want. We recommend using /webhooks/* paths for clarity (e.g., /webhooks/github, /webhooks/stripe).

Basic Webhook

name: data-processor

trigger:
  - type: webhook
    path: /webhooks/process  # Recommended: /webhooks/* prefix
    methods: [POST]
    public: true

flow:
  - agent: process-data

agents:
  - name: process-data
    operation: think
    config:
      provider: anthropic
      model: claude-sonnet-4
      prompt: "Process this data: ${input.data}"

outputs:
  result: ${process-data.output}

Invoke via HTTP:

curl -X POST https://your-worker.workers.dev/webhooks/process \
  -H "Content-Type: application/json" \
  -d '{"data": "hello world"}'

Authenticated Webhook

trigger:
  - type: webhook
    path: /webhooks/secure-endpoint
    methods: [POST, PUT]
    auth:
      type: bearer
      secret: ${env.API_TOKEN}

Invoke with authentication:

curl -X POST https://your-worker.workers.dev/webhooks/secure-endpoint \
  -H "Authorization: Bearer ${API_TOKEN}" \
  -H "Content-Type: application/json" \
  -d '{"action": "process"}'

Webhook Authentication Types

Bearer Token:

trigger:
  - type: webhook
    path: /webhooks/bearer-auth
    auth:
      type: bearer
      secret: ${env.WEBHOOK_SECRET}

HMAC Signature (GitHub-style):

trigger:
  - type: webhook
    path: /webhooks/github
    auth:
      type: signature
      secret: ${env.GITHUB_SECRET}

Sender must include:

X-Webhook-Signature: sha256=abc123...
X-Webhook-Timestamp: 1705315200

Basic Authentication:

trigger:
  - type: webhook
    path: /webhooks/basic-auth
    auth:
      type: basic
      secret: ${env.BASIC_CREDS}  # Format: username:password

Async Webhook Execution

For long-running ensembles, return immediately and process in background:

trigger:
  - type: webhook
    path: /webhooks/long-task
    methods: [POST]
    async: true
    timeout: 300000  # 5 minutes
    public: true

Returns immediately with execution ID:

{
  "executionId": "exec-abc123",
  "status": "processing"
}

MCP Triggers

Expose ensembles as Model Context Protocol tools for AI assistants. Conductor automatically generates MCP tool schemas from your ensemble’s inputs definition.

Basic MCP Tool

name: search-docs
description: "Search documentation for answers"

inputs:
  query:
    type: string
    description: "Search query"
    required: true
  limit:
    type: number
    description: "Max results to return"
    optional: true

trigger:
  - type: mcp
    auth:
      type: bearer
      secret: ${env.MCP_TOKEN}  # Supports $env.VAR_NAME syntax

flow:
  - agent: search

agents:
  - name: search
    operation: data
    config:
      backend: vectorize
      binding: DOCS_INDEX
      operation: query
      vector: ${input.query}
      topK: ${input.limit || 5}

outputs:
  results: ${search.output}

MCP Endpoints:

GET /mcp/tools - List all ensembles exposed as MCP tools (with auto-generated input schemas)
POST /mcp/tools/{name} - Invoke an ensemble via MCP protocol

The ensemble becomes available as an MCP tool with auto-generated schema:

// AI assistants can call via MCP
{
  "tool": "search-docs",
  "parameters": {
    "query": "how to configure webhooks",
    "limit": 10
  }
}

Input Schema Generation

Conductor automatically converts your inputs block to MCP’s JSON Schema format:

# Your ensemble inputs
inputs:
  code:
    type: string
    description: "Code to analyze"
    required: true
  language:
    type: string
    description: "Programming language"
    optional: true

Becomes:

{
  "inputSchema": {
    "type": "object",
    "properties": {
      "code": {
        "type": "string",
        "description": "Code to analyze"
      },
      "language": {
        "type": "string",
        "description": "Programming language"
      }
    },
    "required": ["code"]
  }
}

Authentication Options

Bearer Token (simple or JWT):

trigger:
  - type: mcp
    auth:
      type: bearer
      secret: $env.MCP_TOKEN  # Simple token comparison

If JWT_SECRET is configured in your environment, bearer tokens are validated as JWTs. OAuth (coming soon):

trigger:
  - type: mcp
    auth:
      type: oauth  # Validates JWT format, full OAuth requires external provider

Public MCP Tool

trigger:
  - type: mcp
    public: true  # No authentication required

See MCP Integration for complete guide.

Email Triggers

Trigger ensembles via Cloudflare Email Routing. Conductor fully parses RFC822 emails including MIME multipart content and attachments.

Basic Email Trigger

name: support-ticket-router

trigger:
  - type: email
    to: "support@*"  # Wildcard matching for address patterns
    # Or use specific addresses:
    # addresses:
    #   - support@example.com
    #   - help@example.com
    public: false
    auth:
      from:
        - "*@example.com"  # Only accept from company domain
    reply_with_output: true

flow:
  - agent: classify-ticket
  - agent: route-ticket

agents:
  - name: classify-ticket
    operation: think
    config:
      provider: anthropic
      model: claude-sonnet-4
      prompt: |
        Classify this support email:
        From: ${input.from}
        Subject: ${input.subject}
        Body: ${input.body}

  - name: route-ticket
    operation: http
    config:
      url: ${env.TICKET_SYSTEM_API}/tickets
      method: POST
      body:
        category: ${classify-ticket.output.category}
        priority: ${classify-ticket.output.priority}
        content: ${input.body}

outputs:
  ticketId: ${route-ticket.output.id}
  category: ${classify-ticket.output.category}

Configure Cloudflare Email Routing to forward to your Worker.

Email Input Fields (RFC822 Parsed)

Conductor parses RFC822 emails and provides structured data to your ensemble:

Field	Type	Description
`input.from`	`string`	Sender email address
`input.to`	`string`	Recipient email address
`input.subject`	`string`	Email subject line
`input.body`	`string`	Plain text body
`input.html`	`string \| null`	HTML body (if present)
`input.headers`	`Record<string, string>`	All email headers as key-value pairs
`input.attachments`	`Array<Attachment>`	File attachments (see below)

Attachment format:

{
  filename: string;      // Original filename
  contentType: string;   // MIME type (e.g., "application/pdf")
  content: string;       // Base64-encoded content
}

Example: Processing attachments

agents:
  - name: process-invoice
    operation: code
    config:
      handler: |
        const { attachments } = context.input
        const pdfAttachment = attachments.find(a => a.contentType === 'application/pdf')
        if (pdfAttachment) {
          // Process PDF content (base64 encoded)
          return { hasPdf: true, filename: pdfAttachment.filename }
        }
        return { hasPdf: false }
    input:
      attachments: ${input.attachments}

Reply with Output

When reply_with_output: true, ensemble outputs are sent back via email:

trigger:
  - type: email
    addresses: [assistant@example.com]
    reply_with_output: true

outputs:
  response: ${process-email.output.response}

Queue Triggers

Process Cloudflare Queue messages in batches.

Basic Queue Consumer

name: task-processor

trigger:
  - type: queue
    queue: TASK_QUEUE
    batch_size: 10
    max_retries: 3
    max_wait_time: 5  # seconds

flow:
  - agent: process-batch

agents:
  - name: process-batch
    operation: queue
    config:
      mode: consume
      queue: TASK_QUEUE

outputs:
  processed: ${process-batch.output.count}

Queue Configuration

queue - Cloudflare Queue binding name (must match wrangler.toml)
batch_size - Maximum messages per batch (default: 10)
max_retries - Retry failed messages (default: 3)
max_wait_time - Max seconds to wait for batch to fill

Note: To send messages to queues, use the queue operation - see Queue Operation documentation.

Cron Triggers

Schedule ensemble execution with cron expressions.

Basic Cron Trigger

name: daily-report

trigger:
  - type: cron
    cron: "0 8 * * *"  # Daily at 8 AM UTC
    timezone: "America/New_York"
    enabled: true

flow:
  - agent: generate-report
  - agent: send-email

agents:
  - name: generate-report
    operation: storage
    config:
      type: d1
      query: |
        SELECT COUNT(*) as orders,
               SUM(total) as revenue
        FROM orders
        WHERE created_at >= strftime('%s', 'now', '-1 day') * 1000

  - name: send-email
    operation: email
    config:
      to: [team@example.com]
      subject: "Daily Report - ${new Date().toDateString()}"
      body: |
        Orders: ${generate-report.output[0].orders}
        Revenue: $${generate-report.output[0].revenue}

outputs:
  sent: ${send-email.success}

Cron Expression Format

Standard cron syntax (5 or 6 fields):

* * * * *
│ │ │ │ │
│ │ │ │ └─ Day of week (0-7, 0 and 7 are Sunday)
│ │ │ └─── Month (1-12)
│ │ └───── Day of month (1-31)
│ └─────── Hour (0-23)
└───────── Minute (0-59)

Examples:

"0 0 * * *" - Daily at midnight UTC
"0 */4 * * *" - Every 4 hours
"0 9 * * 1-5" - Weekdays at 9 AM
"0 0 1 * *" - First day of month
"0 0 * * 0" - Every Sunday

Cron with Custom Input

Pass data to scheduled executions:

trigger:
  - type: cron
    cron: "0 8 * * *"
    timezone: "America/New_York"
    input:
      report_type: "daily"
      recipients: ["team@example.com"]
    metadata:
      description: "Daily morning report"
      team: "analytics"

Access in ensemble:

agents:
  - name: process
    operation: think
    config:
      prompt: "Generate ${input.report_type} report"

Multiple Cron Triggers

Ensembles can have multiple schedules:

trigger:
  # Daily report
  - type: cron
    cron: "0 8 * * *"
    timezone: "America/New_York"
    input:
      frequency: "daily"

  # Weekly summary
  - type: cron
    cron: "0 9 * * 1"
    timezone: "America/New_York"
    input:
      frequency: "weekly"

Disable Cron Trigger

Temporarily disable without removing:

trigger:
  - type: cron
    cron: "0 0 * * *"
    enabled: false  # Disabled

Schedule Metadata

Access schedule information in ensemble:

flow:
  - agent: process
    input:
      cron: ${input._schedule.cron}
      timezone: ${input._schedule.timezone}
      triggered_at: ${input._schedule.triggeredAt}

Build Triggers

Run ensembles at build time to generate static content. Build triggers execute during the ensemble conductor build command and are useful for generating documentation, static pages, or pre-computing data.

Basic Build Trigger

name: generate-docs

trigger:
  - type: build
    enabled: true
    output: ./dist/docs

flow:
  - agent: docs
    input: { action: generate-openapi }

agents:
  - name: docs
    operation: docs
    config:
      format: openapi
      version: 3.0.0
      includeSchemas: true

outputs:
  path: ${trigger.output}/openapi.json
  content: ${docs.output}

Run with: ensemble conductor build

Build with Custom Input

Pass data to build-time executions:

trigger:
  - type: build
    enabled: true
    output: ./dist/static
    input:
      format: json
      includeExamples: true
    metadata:
      description: "Generate API documentation"
      version: "1.0.0"

flow:
  - agent: docs
    input:
      format: ${trigger.input.format}
      examples: ${trigger.input.includeExamples}

Access trigger metadata in ensemble:

agents:
  - name: process
    operation: think
    config:
      prompt: "Generate ${trigger.metadata.description} v${trigger.metadata.version}"

Multiple Build Triggers

Generate different static assets:

trigger:
  # Generate OpenAPI docs
  - type: build
    enabled: true
    output: ./dist/docs
    input: { action: generate-openapi }

  # Generate static site
  - type: build
    enabled: true
    output: ./dist/site
    input: { action: generate-site }

  # Pre-compute analytics
  - type: build
    enabled: true
    output: ./dist/data
    input: { action: compute-analytics }

Conditional Build

Use enabled to skip builds conditionally:

trigger:
  - type: build
    enabled: ${env.BUILD_DOCS === 'true'}
    output: ./dist/docs

CLI Triggers

Create custom CLI commands that execute ensembles. CLI triggers are invoked via ensemble conductor run <command> and support options with defaults and validation.

Basic CLI Trigger

name: generate-docs

trigger:
  - type: cli
    command: generate-docs
    description: Generate documentation

flow:
  - agent: docs
    input: { action: generate-openapi }

agents:
  - name: docs
    operation: docs
    config:
      format: openapi

outputs:
  result: ${docs.output}

Run with: ensemble conductor run generate-docs

CLI with Options

Define command-line options with types and defaults:

trigger:
  - type: cli
    command: generate-docs
    description: Generate documentation in various formats
    options:
      - name: format
        type: string
        default: yaml
        description: Output format (yaml, json, html)
      - name: output
        type: string
        required: true
        description: Output file path
      - name: verbose
        type: boolean
        default: false
        description: Enable verbose logging

flow:
  - agent: docs
    input:
      format: ${trigger.options.format}
      outputPath: ${trigger.options.output}
      verbose: ${trigger.options.verbose}

Run with:

ensemble conductor run generate-docs --format=json --output=./docs/api.json --verbose

CLI Options Types

Supported option types:

options:
  # String option
  - name: format
    type: string
    default: yaml

  # Number option
  - name: limit
    type: number
    default: 100

  # Boolean flag
  - name: verbose
    type: boolean
    default: false

  # Required option
  - name: output
    type: string
    required: true

Access Options in Flow

CLI options are available via ${trigger.options.*}:

flow:
  - operation: code
    config:
      handler: |
        const format = context.input.format
        const output = context.input.outputPath
        console.log(`Generating ${format} to ${output}`)
        return { success: true }
    input:
      format: ${trigger.options.format}
      outputPath: ${trigger.options.output}

Startup Triggers

Run ensembles on Worker cold start, before HTTP routes are registered. Startup triggers are ideal for cache warming, health checks, and initialization tasks.

Cold start semantics: Cloudflare Workers naturally cold start after a few minutes of inactivity. Startup triggers run once per cold start - not on every request.

Basic Startup Trigger

name: cache-warmer
description: Pre-warm caches on Worker startup

trigger:
  - type: startup
    input:
      cache_keys: ['products', 'categories', 'users']

flow:
  - agent: warm-cache

agents:
  - name: warm-cache
    operation: code
    config:
      handler: |
        for (const key of context.input.cache_keys) {
          const data = await context.env.DB.prepare(`SELECT * FROM ${key} LIMIT 100`).all()
          await context.env.KV.put(`cache:${key}`, JSON.stringify(data.results))
        }
        return { warmed: context.input.cache_keys.length }

outputs:
  result: ${warm-cache.output}

Health Check on Startup

Verify dependencies are available before serving requests:

name: startup-health-check
description: Verify dependencies on startup

trigger:
  - type: startup
    metadata:
      priority: high

flow:
  - agent: check-db
  - agent: check-kv

agents:
  - name: check-db
    operation: code
    config:
      handler: |
        const result = await context.env.DB.prepare('SELECT 1').first()
        if (!result) throw new Error('Database not responding')
        return { db: 'ok' }

  - name: check-kv
    operation: code
    config:
      handler: |
        await context.env.KV.put('health:startup', Date.now().toString())
        return { kv: 'ok' }

outputs:
  health: ${check-kv.output}

Startup with Custom Input

Pass static input data to startup ensembles:

trigger:
  - type: startup
    enabled: true
    input:
      warmCaches: true
      preloadUsers: 100
    metadata:
      description: "Initialize application state"
      priority: "high"

Access in ensemble:

agents:
  - name: init
    operation: code
    config:
      handler: |
        if (context.input.warmCaches) {
          // Pre-warm caches
        }
        return { initialized: true }

Disable Startup Trigger

Temporarily disable without removing:

trigger:
  - type: startup
    enabled: false  # Disabled
    input:
      action: warm-cache

Performance Considerations

Keep startup triggers fast (under 5 seconds). Cloudflare Workers have a 30-second initialization timeout. While startup triggers run non-blocking via waitUntil(), slow triggers delay background task completion.

Good use cases:

Cache warming (KV reads/writes)
Health checks (database ping)
Configuration loading
Metrics initialization

Avoid:

Heavy data processing
Long-running API calls
Complex AI inference
Large file operations

Startup vs Cron

Feature	Startup	Cron
When	On cold start	On schedule
Frequency	Variable (depends on traffic)	Predictable
Blocking	Non-blocking (waitUntil)	Blocking
Use case	Initialization	Recurring tasks

If you need predictable timing, use cron. If you need “run once when Worker starts”, use startup.

Multiple Triggers

Ensembles can have multiple triggers of different types:

name: data-processor

trigger:
  # HTTP API
  - type: webhook
    path: /process
    methods: [POST]
    auth:
      type: bearer
      secret: ${env.API_TOKEN}

  # MCP Tool
  - type: mcp
    public: false
    auth:
      type: bearer
      secret: ${env.MCP_TOKEN}

  # Email
  - type: email
    addresses: [process@example.com]
    reply_with_output: true

  # Queue
  - type: queue
    queue: PROCESS_QUEUE
    batch_size: 10

  # Scheduled
  - type: cron
    cron: "0 */6 * * *"  # Every 6 hours
    timezone: "UTC"

flow:
  - agent: process-data

agents:
  - name: process-data
    operation: think
    config:
      provider: anthropic
      model: claude-sonnet-4
      prompt: "Process: ${input}"

outputs:
  result: ${process-data.output}

This ensemble can be invoked via:

POST to /webhooks/process
MCP tool call data-processor
Email to process@example.com
Queue message to PROCESS_QUEUE
Cron schedule every 6 hours

Trigger Security

Default-Deny Policy

All triggers (except queue and cron) require either:

Authentication (auth configuration), OR
Explicit public access (public: true)

✅ Valid:

trigger:
  # Has authentication
  - type: webhook
    path: /secure
    auth:
      type: bearer
      secret: ${env.TOKEN}

  # Explicitly public
  - type: webhook
    path: /public
    public: true

  # Queue/cron don't need auth (internal triggers)
  - type: queue
    queue: TASK_QUEUE

❌ Invalid:

trigger:
  # ERROR: No auth and not marked public
  - type: webhook
    path: /unsafe

Best Practices

Use environment variables for secrets:

auth:
  type: bearer
  secret: ${env.API_TOKEN}  # Never hardcode!

Verify webhook signatures: Use signature auth type for external webhooks

Limit email senders:

trigger:
  - type: email
    addresses: [support@example.com]
    auth:
      from: ["*@trusted-domain.com"]

Use async for long operations:

trigger:
  - type: webhook
    async: true
    timeout: 300000

Configuration Reference

HTTP Trigger

Field	Type	Required	Description
`type`	`"http"`	Yes	Trigger type
`path`	`string`	No	URL path with params (default: `/{ensemble-name}`)
`paths`	`array`	No	Multiple paths configuration (alternative to `path`)
`paths[].path`	`string`	Yes	URL path with params
`paths[].methods`	`string[]`	Yes	HTTP methods for this path
`methods`	`string[]`	No	HTTP methods (default: `["GET"]`)
`auth`	`object`	Conditional	Authentication config (required unless `public: true`)
`public`	`boolean`	No	Allow unauthenticated access (default: `false`)
`rateLimit`	`object`	No	Rate limiting configuration
`rateLimit.requests`	`number`	Yes	Max requests per window
`rateLimit.window`	`number`	Yes	Time window in seconds
`rateLimit.key`	`"ip"\|"user"`	No	Rate limit by IP or user (default: `"ip"`)
`cors`	`object`	No	CORS configuration
`cors.origin`	`string\|string[]`	No	Allowed origins
`cors.credentials`	`boolean`	No	Allow credentials
`responses`	`object`	No	Response type configuration
`responses.html`	`object`	No	HTML response config
`responses.json`	`object`	No	JSON response config
`templateEngine`	`"liquid"\|"handlebars"\|"simple"`	No	Template engine for HTML (default: `"liquid"`)
`middleware`	`function[]`	No	Custom Hono middleware

Webhook Trigger

Field	Type	Required	Description
`type`	`"webhook"`	Yes	Trigger type
`path`	`string`	No	Endpoint path (default: `/{ensemble-name}`)
`methods`	`string[]`	No	HTTP methods (default: `["POST"]`)
`auth`	`object`	Conditional	Authentication config (required unless `public: true`)
`public`	`boolean`	No	Allow unauthenticated access (default: `false`)
`async`	`boolean`	No	Background execution (default: `false`)
`timeout`	`number`	No	Timeout in milliseconds

MCP Trigger

Field	Type	Required	Description
`type`	`"mcp"`	Yes	Trigger type
`auth`	`object`	Conditional	Authentication config (required unless `public: true`)
`public`	`boolean`	No	Allow unauthenticated access (default: `false`)

Email Trigger

Field	Type	Required	Description
`type`	`"email"`	Yes	Trigger type
`addresses`	`string[]`	Yes	Email addresses to receive
`auth`	`object`	No	Sender whitelist patterns
`public`	`boolean`	No	Allow any sender (default: `false`)
`reply_with_output`	`boolean`	No	Send output via email reply (default: `false`)

Queue Trigger

Field	Type	Required	Description
`type`	`"queue"`	Yes	Trigger type
`queue`	`string`	Yes	Queue binding name
`batch_size`	`number`	No	Max messages per batch (default: `10`)
`max_retries`	`number`	No	Retry failed messages (default: `3`)
`max_wait_time`	`number`	No	Max seconds to wait for batch

Cron Trigger

Field	Type	Required	Description
`type`	`"cron"`	Yes	Trigger type
`cron`	`string`	Yes	Cron expression
`timezone`	`string`	No	IANA timezone (default: `"UTC"`)
`enabled`	`boolean`	No	Enable trigger (default: `true`)
`input`	`object`	No	Default input data
`metadata`	`object`	No	Schedule metadata

Build Trigger

Field	Type	Required	Description
`type`	`"build"`	Yes	Trigger type
`enabled`	`boolean`	No	Enable trigger (default: `true`)
`output`	`string`	Yes	Output directory path
`input`	`object`	No	Default input data
`metadata`	`object`	No	Build metadata

CLI Trigger

Field	Type	Required	Description
`type`	`"cli"`	Yes	Trigger type
`command`	`string`	Yes	Command name
`description`	`string`	No	Command description
`options`	`array`	No	Command-line options
`options[].name`	`string`	Yes	Option name
`options[].type`	`"string"\|"number"\|"boolean"`	Yes	Option type
`options[].default`	`any`	No	Default value
`options[].required`	`boolean`	No	Whether option is required (default: `false`)
`options[].description`	`string`	No	Option description

Startup Trigger

Field	Type	Required	Description
`type`	`"startup"`	Yes	Trigger type
`enabled`	`boolean`	No	Enable trigger (default: `true`)
`input`	`object`	No	Static input data for startup execution
`metadata`	`object`	No	Additional metadata (e.g., priority, description)

Response Formats

The format field in the output block controls response serialization and Content-Type headers. Use this for non-JSON responses like CSV, XML, YAML, etc.

Format Types

Type	Content-Type	Description
`json`	`application/json`	JSON serialization (default)
`text`	`text/plain`	Plain text
`html`	`text/html`	HTML content
`xml`	`application/xml`	XML content
`csv`	`text/csv`	CSV serialization from arrays
`markdown`	`text/markdown`	Markdown content
`yaml`	`application/x-yaml`	YAML serialization
`ics`	`text/calendar`	iCalendar format
`rss`	`application/rss+xml`	RSS feed
`atom`	`application/atom+xml`	Atom feed

CSV Export Example

name: export-users
trigger:
  - type: http
    path: /export/users.csv
    methods: [GET]
    public: true

agents:
  - name: fetch-users
    operation: data
    config:
      backend: d1
      query: SELECT id, name, email FROM users

output:
  status: 200
  headers:
    Content-Disposition: attachment; filename="users.csv"
  format:
    type: csv
    extract: users
  body:
    users: ${fetch-users.output}

YAML Config Example

name: app-config
trigger:
  - type: http
    path: /config.yaml
    methods: [GET]
    public: true

agents:
  - name: build-config
    operation: transform
    config:
      expression: |
        {
          "version": "1.0",
          "features": ["auth", "analytics"]
        }

output:
  status: 200
  format:
    type: yaml
    extract: config
  body:
    config: ${build-config.output}

iCalendar Event

name: calendar-event
trigger:
  - type: http
    path: /event.ics
    methods: [GET]
    public: true

output:
  status: 200
  headers:
    Content-Disposition: attachment; filename="event.ics"
  format: ics
  body:
    calendar: ${generate-event.output}

The extract option specifies which field from the body should be serialized. If not specified, the entire body is serialized.

Triggers vs API Routes

Conductor provides two ways to execute ensembles:

Triggers (This Page)

Triggers are defined in ensemble YAML and provide:

Path-based routing with parameters (/users/:id)
Per-trigger authentication configuration
Rate limiting and CORS settings
Auto-discovery from ensemble definitions

trigger:
  - type: http
    path: /api/users/:id
    public: true  # Explicit public access

API Execute Routes

The /api/v1/execute/* routes provide programmatic access:

# Execute via API (requires authentication by default)
curl -X POST https://your-worker.workers.dev/api/v1/execute/ensemble/my-workflow \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"input": {}}'

API Execution Control: You can control which ensembles are accessible via the Execute API using:

Project-level policy in conductor.config.ts:

api: {
  execution: {
    ensembles: { requireExplicit: true }  // Require opt-in
  }
}

Per-ensemble control via apiExecutable:

name: internal-workflow
apiExecutable: false  # Prevent Execute API access

Key Differences:

Feature	Triggers	API Routes
Configuration	Per-trigger in YAML	Global in `conductor.config.ts`
Default Auth	Requires `public: true` or `auth`	Requires auth (secure by default)
Use Case	Public APIs, webhooks	Service-to-service, internal
Permissions	Per-trigger auth	Permission-based scoping
Access Control	Trigger config	`apiExecutable` flag

See Security & Authentication for complete auth documentation.

Next Steps

Security & Auth

API keys, permissions, and authentication

MCP Integration

Expose ensembles as MCP tools

Queue Operation

Queue message processing

Event-Driven

Event-driven patterns

​Overview

​HTTP Triggers

​Basic JSON API

​Server-Rendered HTML Page

​HTTP with Authentication & Rate Limiting

​HTTP Request Context

​HTTP vs Webhook

​Multi-Path HTTP Triggers

​Complex Website Structure

​Webhook Triggers

​Basic Webhook

​Authenticated Webhook

​Webhook Authentication Types

​Async Webhook Execution

​MCP Triggers

​Basic MCP Tool

​Input Schema Generation

​Authentication Options

​Public MCP Tool

​Email Triggers

​Basic Email Trigger

​Email Input Fields (RFC822 Parsed)

​Reply with Output

​Queue Triggers

​Basic Queue Consumer

​Queue Configuration

​Cron Triggers

​Basic Cron Trigger

​Cron Expression Format

​Cron with Custom Input

​Multiple Cron Triggers

​Disable Cron Trigger

​Schedule Metadata

​Build Triggers

​Basic Build Trigger

​Build with Custom Input

​Multiple Build Triggers

​Conditional Build

​CLI Triggers

​Basic CLI Trigger

​CLI with Options

​CLI Options Types

​Access Options in Flow

​Startup Triggers

​Basic Startup Trigger

​Health Check on Startup

​Startup with Custom Input

​Disable Startup Trigger

​Performance Considerations

​Startup vs Cron

​Multiple Triggers

​Trigger Security

​Default-Deny Policy

​Best Practices

​Configuration Reference

​HTTP Trigger

​Webhook Trigger

​MCP Trigger

​Email Trigger

​Queue Trigger

​Cron Trigger

​Build Trigger

​CLI Trigger

​Startup Trigger

​Response Formats

​Format Types

​CSV Export Example

​YAML Config Example

​iCalendar Event

​Triggers vs API Routes

​Triggers (This Page)

​API Execute Routes

​Next Steps

Security & Auth

MCP Integration

Queue Operation

Event-Driven

Overview

HTTP Triggers

Basic JSON API

Server-Rendered HTML Page

HTTP with Authentication & Rate Limiting

HTTP Request Context

HTTP vs Webhook

Multi-Path HTTP Triggers

Complex Website Structure

Webhook Triggers

Basic Webhook

Authenticated Webhook

Webhook Authentication Types

Async Webhook Execution

MCP Triggers

Basic MCP Tool

Input Schema Generation

Authentication Options

Public MCP Tool

Email Triggers

Basic Email Trigger

Email Input Fields (RFC822 Parsed)

Reply with Output

Queue Triggers

Basic Queue Consumer

Queue Configuration

Cron Triggers

Basic Cron Trigger

Cron Expression Format

Cron with Custom Input

Multiple Cron Triggers

Disable Cron Trigger

Schedule Metadata

Build Triggers

Basic Build Trigger

Build with Custom Input

Multiple Build Triggers

Conditional Build

CLI Triggers

Basic CLI Trigger

CLI with Options

CLI Options Types

Access Options in Flow

Startup Triggers

Basic Startup Trigger

Health Check on Startup

Startup with Custom Input

Disable Startup Trigger

Performance Considerations

Startup vs Cron

Multiple Triggers

Trigger Security

Default-Deny Policy

Best Practices

Configuration Reference

HTTP Trigger

Webhook Trigger

MCP Trigger

Email Trigger

Queue Trigger

Cron Trigger

Build Trigger

CLI Trigger

Startup Trigger

Response Formats

Format Types

CSV Export Example

YAML Config Example

iCalendar Event

Triggers vs API Routes

Triggers (This Page)

API Execute Routes

Next Steps