Skip to content

DAG Constructor — Overview

The DAG Constructor is a visual, browser-based pipeline builder built into the Titan Dashboard. It lets you design, configure, and deploy DAGs without writing any code — and generates equivalent YAML or Python SDK output as you build.

Open it at: http://127.0.0.1:5000/dags/new

DAG Constructor canvas

Layout

┌─────────────────────────────────────────────────────┬──────────┐
│  Topbar — pipeline name, actions, deploy            │          │
├────────────────────────────────────────────────────┤ Sidebar  │
│                                                    │          │
│  Canvas — drag nodes, draw edges                  │  Node    │
│                                                    │  Config  │
│                                                    │          │
├────────────────────────────────────────────────────┤          │
│  Output panel — live YAML / Python SDK preview    │          │
└────────────────────────────────────────────────────┴──────────┘

Topbar — add nodes, name the pipeline, undo/redo, save draft, load a previous DAG, deploy.

Canvas — the work area. Nodes are draggable. Edges are drawn by dragging from a node's output port to another node.

Sidebar — appears when a node is selected. Configure all job properties: script, capability, args, HITL gate, and more.

Output panel — live YAML and Python SDK codegen that updates as you build. Copyable with one click.

Quick start

  1. Click + Add Task in the topbar to place a node on the canvas
  2. Select the node — the sidebar opens
  3. Set a Job ID and choose a Script File from the dropdown
  4. Hover the node — a port (●) appears on its right edge
  5. Drag from the port to another node to create a dependency edge
  6. Click Deploy to Titan when ready

After a successful deploy, the button becomes a "✓ Deployed — View DAG →" link that opens the DAG Visualizer for the run.

Node types

Type Use for
Task (Run Payload) Scripts that run once and exit — training, preprocessing, analysis
Service Long-running processes that stay alive — model servers, API endpoints

Capability routing

Each node has a Requirement field that controls which worker receives the job:

Requirement Worker needed
GENERAL Any available worker
GPU Worker started with GPU capability
HIGH_MEM Worker started with HIGH_MEM capability
PYTHON Any Python-capable worker

Workers advertise their capability at startup. The master only dispatches a job to a worker that matches its requirement.

Next steps