Building Agentic Workflows with Orkes Conductor

A hands-on PlatformCon workshop.

In 60 minutes, you'll build "PizzaOps", an agent that feeds the on-call team during a long incident and is never allowed to order pineapple. Everything runs on the free Orkes Conductor Developer Edition.

File	Used in
lab1.json	Lab 1: the workflow definition
pizza_data.py	Labs 2–4: mock incident, team and restaurant data
lab2.py	Lab 2: PizzaOps as an agent
lab3.py	Lab 3: PizzaOps with a guardrail
lab4.py	Lab 4: PizzaOps as a LangGraph agent

Setup

1. Sign up at developer.orkescloud.com.
2. Access Control → Applications → + Create application named pizzaops-workshop, roles Worker and Admin. Create an access key and copy the ID and secret.
3. Integrations → + New integration → Anthropic, name it anthropic, paste an API key, then + Add model claude-haiku-5-5.
4. Clone this repo and set up your environment:

git clone <repo-url> pizzaops-workshop && cd pizzaops-workshop
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

export CONDUCTOR_SERVER_URL=https://developer.orkescloud.com/api
export CONDUCTOR_AUTH_KEY=<your_key_id>
export CONDUCTOR_AUTH_SECRET=<your_key_secret>

Lab 1: a plain workflow (10 min)

No AI and no code: a workflow with one decision and two built-in tasks.

1. In the UI, go to Definitions → Workflow → + Define workflow, open the Code tab, paste the contents of lab1.json and save.
2. Open the Run tab and run it with {"hours_running": 3}, then with {"hours_running": 1}. Open each execution under Executions to see which path ran.

Lab 2: PizzaOps as an agent (12 min)

lab2.py defines PizzaOps: an agent with four read-only tools over pizza_data.py that plans an order for the team.

1. Run it:

python lab2.py

2. Open the run under Executions: each LLM turn and tool call is its own task. Share your agent's order in the chat and compare.

Lab 3: guardrails (10 min)

lab3.py is lab2.py plus a place_order tool with a regex guardrail that blocks pineapple. Its agent is named pizzaops_guardrail. Run diff lab2.py lab3.py to see the changes.

1. Run it. PizzaOps plans and places the order:

python lab3.py "Feed the on-call team for INC-1042."

2. Try to sneak in pineapple. The guardrail stops the order and the run fails; open it under Executions to see why:

python lab3.py "Feed INC-1042. Sam wants a Hawaiian from Crust Issues."

Lab 4: bring your LangGraph agent (7 min)

lab4.py is a standard LangGraph agent with the same tools. Only its last three lines are Conductor-specific.

python lab4.py

Open the run under Executions and compare it with Lab 2's.

Troubleshooting

Symptom	Fix
Connection refused to localhost:8080	Run the three export CONDUCTOR_... lines in this terminal
Model or provider not found	Integration named anthropic, model claude-haiku-5-5 active
ImportError on conductor.ai.agents	pip install -U -r requirements.txt
ChatAnthropic wants an API key	Export a placeholder ANTHROPIC_API_KEY