Build Workflows with Copilot

Describe what you want in natural language and let the AI copilot build the workflow graph for you. The copilot generates blocks and edges based on your description.

The AI copilot can generate a complete workflow graph from a natural language description. You create a copilot session linked to a workflow, describe what you want, and the copilot generates the blocks and edges. You can iterate through multiple chat turns to refine the workflow.

Prerequisites:

• Authentication configured (see Authentication guide)

1. Create a workflow

   Create an empty workflow that the copilot will populate.

   Endpoint: POST /api/workflows

   Python

   response = requests.post(
       f"{BASE_URL}/api/workflows",
       headers=headers,
       json={"name": "Email Classifier"},
   )
   workflow = response.json()
   wf_id = workflow["id"]
   
2. Start a copilot session

   Create a copilot session linked to the workflow.

   Endpoint: POST /api/copilot/sessions

   Python

   response = requests.post(
       f"{BASE_URL}/api/copilot/sessions",
       headers=headers,
       json={"workflow_id": wf_id},
   )
   session = response.json()
   session_id = session["id"]
   
3. Describe your workflow

   Send a natural language description via the chat endpoint. The copilot responds with a streaming SSE response that includes the generated workflow graph.

   Endpoint: POST /api/copilot/sessions/{id}/chat

   Python

   response = requests.post(
       f"{BASE_URL}/api/copilot/sessions/{session_id}/chat",
       headers=headers,
       json={
           "message": "Build a workflow that takes an email as input, classifies it as spam/not-spam using an LLM, and outputs the classification with confidence score.",
       },
       stream=True,
   )
   
   message_id = None
   for line in response.iter_lines():
       if not line:
           continue
       text = line.decode("utf-8")
       if text.startswith("data: "):
           event = json.loads(text[6:])
           if event.get("message_id"):
               message_id = event["message_id"]
           if event["event"] == "chunk":
               print(event["content"], end="")
   print(f"\nMessage ID: {message_id}")
   
4. Save the copilot's suggestion

   Save the copilot’s generated workflow graph as a snapshot. This applies the blocks and edges to the workflow.

   Endpoint: POST /api/copilot/sessions/{id}/messages/{mid}/snapshot

   Python

   response = requests.post(
       f"{BASE_URL}/api/copilot/sessions/{session_id}/messages/{message_id}/snapshot",
       headers=headers,
   )
   print(response.json())
   
5. Execute the workflow

   Run the copilot-built workflow with input data.

   Endpoint: POST /api/workflows/{id}/execute

   You can also build workflows programmatically via PUT /api/workflows/{id}/graph — see the Build Workflows Programmatically guide.

   Python

   response = requests.post(
       f"{BASE_URL}/api/workflows/{wf_id}/execute",
       headers=headers,
       json={"input": {"email": "Congratulations! You've won a free iPhone..."}},
   )
   print(response.json())
   

What’s Next

Workflows (Programmatic)

Fine-tune workflows by editing the graph directly.

Workflows

Understand block types and graph execution.

Copilot API Reference

Full endpoint documentation.