Skip to main content

​
Quick Start

Get a complete AI workflow running on the edge in 5 minutes. No fluff, no 20-step tutorials. Just the essentials.

​
What You’ll Build

A simple company intelligence workflow that:
  1. Fetches company data from an API
  2. Analyzes it with Claude
  3. Returns structured results
Running on Cloudflare Workers at 200+ locations globally.

​
Prerequisites

# You need:
- Node.js 18+ installed
- A Cloudflare account (free tier works)
- 5 minutes

# That's it.

​
Step 1: Install Conductor

npm install -g @ensemble-edge/conductor
conductor --version

​
Step 2: Create a Project

mkdir my-ai-workflow
cd my-ai-workflow
conductor init
This creates: 
my-ai-workflow/
├── ensembles/
│   └── hello-world.yaml      # Your first ensemble
├── components/
│   └── prompts/
│       └── hello.md          # Your first prompt
├── wrangler.toml             # Cloudflare config
├── package.json
└── conductor.config.ts       # Conductor config

​
Step 3: Write Your First Ensemble

Edit ensembles/company-intel.yaml: 
ensemble: company-intel
description: Analyze a company from its domain

agents:
  # Fetch company data
  - name: fetch
    operation: http
    config:
      url: https://api.company-data.com/lookup?domain=${input.domain}
      method: GET
      cache_ttl: 3600  # Cache for 1 hour

  # Analyze with AI
  - name: analyze
    operation: think
    config:
      model: claude-3-5-sonnet-20241022
      prompt: |
        Analyze this company data and provide:
        - Industry classification
        - Key products/services
        - Market position (1-5 scale)
        - Growth indicators

        Company data:
        ${fetch.output}

        Respond in JSON format.
      response_format:
        type: json_object

  # Return results
output:
  company: ${input.domain}
  analysis: ${analyze.output}
  cached: ${fetch.cached}

​
Step 4: Configure Cloudflare

You need two things:

​
1. API Tokens

Add to .dev.vars (local development): 
# .dev.vars
ANTHROPIC_API_KEY=sk-ant-...
Add to Cloudflare dashboard (production): 
# Via dashboard: Workers & Pages > Your Worker > Settings > Variables

ANTHROPIC_API_KEY: <your-key>
Edit wrangler.toml: 
name = "my-ai-workflow"
main = "src/index.ts"
compatibility_date = "2024-11-01"

# KV for caching (optional but recommended)
[[kv_namespaces]]
binding = "CACHE"
id = "your_kv_namespace_id"

# AI Gateway for observability (optional)
[ai]
binding = "AI_GATEWAY"
Create KV namespace: 
wrangler kv:namespace create "CACHE"
# Copy the ID to wrangler.toml

​
Step 5: Test Locally

# Start local dev server
conductor dev

# In another terminal, test it
curl http://localhost:8787/ensembles/company-intel \
  -H "Content-Type: application/json" \
  -d '{"domain": "stripe.com"}'
Response: 
{
  "company": "stripe.com",
  "analysis": {
    "industry": "Financial Technology / Payments",
    "products": ["Payment processing", "Billing", "Connect"],
    "market_position": 5,
    "growth_indicators": "Strong - expanding globally"
  },
  "cached": false
}
Run it again - see "cached": true and <10ms response time.

​
Step 6: Deploy to Production

# Deploy to Cloudflare
conductor deploy

# Get your worker URL
# Example: https://my-ai-workflow.your-subdomain.workers.dev
Test production: 
curl https://my-ai-workflow.your-subdomain.workers.dev/ensembles/company-intel \
  -H "Content-Type: application/json" \
  -d '{"domain": "openai.com"}'
You’re live. Globally. With caching. In 5 minutes.

​
What Just Happened?

  1. Edge Execution - Your workflow runs on Cloudflare’s network at 200+ locations
  2. Sub-50ms Cold Starts - First request takes ~40ms, cached requests <10ms
  3. Built-in Caching - HTTP responses cached in KV automatically
  4. AI Gateway - All AI calls tracked and cached via Cloudflare AI Gateway
  5. Structured Outputs - JSON response format enforced

​
Next Steps

​
Add Component Versioning (Edgit)

# Install Edgit
npm install -g @ensemble-edge/edgit
edgit init

# Version your prompt
edgit tag create company-analysis-prompt v1.0.0

# Deploy to production
edgit deploy set company-analysis-prompt v1.0.0 --to prod
Now reference it in your ensemble: 
agents:
  - name: analyze
    operation: think
    component: company-analysis-prompt@v1.0.0  # Versioned!
    config:
      model: claude-3-5-sonnet-20241022

​
Add A/B Testing

ensemble: company-intel

# Test two prompt versions simultaneously
agents:
  - name: analyze-v1
    operation: think
    component: company-analysis-prompt@v1.0.0
    config:
      model: claude-3-5-sonnet-20241022

  - name: analyze-v2
    operation: think
    component: company-analysis-prompt@v2.0.0
    config:
      model: claude-3-5-sonnet-20241022

output:
  v1_result: ${analyze-v1.output}
  v2_result: ${analyze-v2.output}
  # Compare results and pick winner

​
Add More Operations

agents:
  # Database storage
  - name: store
    operation: storage
    config:
      type: d1
      database: ANALYTICS
      query: |
        INSERT INTO companies (domain, analysis, created_at)
        VALUES (?, ?, ?)
      params:
        - ${input.domain}
        - ${analyze.output}
        - ${Date.now()}

  # Send email notification
  - name: notify
    operation: email
    config:
      to: team@company.com
      subject: "New company analyzed: ${input.domain}"
      body: ${analyze.output}

​
Use Pre-built Agents

agents:
  # Web scraping with 3-tier fallback
  - name: scrape
    agent: scraper
    config:
      url: https://${input.domain}
      extract:
        title: "h1"
        description: "meta[name=description]"

  # Quality validation
  - name: validate
    agent: validator
    config:
      evaluator_type: completeness
      threshold: 0.8
    input:
      content: ${analyze.output}

​
Learn More

Core Concepts

Understand components, agents, and ensembles

Operations Reference

All available operations (think, code, http, storage, etc.)

Pre-built Agents

Ready-to-use agents (scraper, validator, RAG, etc.)

Playbooks

Real-world patterns and recipes

​
Troubleshooting

Add your API key to .dev.vars for local development:
echo 'ANTHROPIC_API_KEY=sk-ant-...' >> .dev.vars
For production, add it in Cloudflare dashboard under Workers > Settings > Variables.
Create a KV namespace first:
wrangler kv:namespace create "CACHE"
Copy the ID to your wrangler.toml file.
First request always takes longer (~100-200ms). Subsequent requests should be <50ms cold start + execution time.Enable caching to get <10ms for repeated requests:
config:
  cache_ttl: 3600  # Cache for 1 hour
Make sure you’ve configured AI Gateway in Cloudflare dashboard:
  1. Go to AI > AI Gateway
  2. Create a gateway
  3. Add the gateway ID to your wrangler.toml
[ai]
binding = "AI_GATEWAY"
gateway_id = "your-gateway-id"
That’s it. You’ve got a production AI workflow running on the edge with caching, versioning, and infinite scale.No Docker, no Kubernetes, no server management. Just Git, YAML, and Cloudflare Workers.