Skip to main content

Getting Started with ThinkThing

Getting Started with ThinkThing

This guide walks you through creating your first graph, adding nodes, connecting them, and running the workflow.

Creating a New Graph

  1. Open ThinkThing from the QUI Core dashboard (Services → ThinkThing) or from the system tray
  2. Click New Graph in the gallery
  3. Give your graph a name
  4. A blank canvas opens with a Start node already placed

[Screenshot: New empty graph with Start node]

Adding Nodes

  1. Open the node palette on the left side of the canvas
  2. Browse categories or search for a specific node type
  3. Click a node to add it to the canvas, or drag it to a specific position

Your First Graph: Start → Prompt → End

The simplest useful graph:

  1. You already have a Start node — click it to set the initial input text (this is what flows into the first processing node). For webhook-triggered or scheduled graphs, the input comes from the trigger payload instead.
  2. From the palette, add a Prompt node (under Cognition)
  3. Add an End node (under Control Flow)
  4. Add an Anima node — this connects the Prompt node to an AI character

Connecting Nodes

Nodes have handles — small connection points on their edges:

  • Output handle (right side) — where content leaves the node
  • Input handle (left side) — where content enters the node
  • Tool handle (colored, typically top) — special connections for Anima, tools, or services

To connect two nodes:

  1. Click and drag from an output handle on the source node
  2. Drop onto the input handle of the target node
  3. An edge (connection line) appears between them

Connect your graph: Start → Prompt → End, and connect the Anima node to the Prompt node's tool handle.

Configuring Nodes

Click any node to open its configuration panel. Each node type has different settings:

Anima Node

Select which AI character powers the LLM calls in this graph. Choose the character from the dropdown — this determines the model, personality, and tool access for all connected cognition nodes.

Optional: Enable Agentic Mode to let connected nodes autonomously iterate with tools (max steps, token budget, timeout configurable).

Prompt Node

Enter the task or question you want the LLM to process. The prompt can include:

  • Static text — "Summarize the following content"
  • Variable references — {{input}} for the incoming content from the previous node
  • Context variables — {{var_name}} for values set by other nodes

Running a Graph

  1. Click the Run button (or use the execution endpoint)
  2. The Execution Monitor opens on the right side, showing real-time progress
  3. Watch each node execute step by step:
    • Pending — waiting to execute
    • Running — currently processing
    • Completed — finished successfully
    • Failed — error occurred

Content flows through the graph following your connections. Each node processes its input and passes the output to the next node.

[Screenshot: Execution monitor showing step-by-step progress with content preview]

Monitoring Execution

The execution monitor shows:

  • Node name and status — which node is running and whether it succeeded
  • Input content — what the node received
  • Output content — what the node produced
  • Duration — how long each step took
  • LLM details — model used, tokens consumed (for cognition nodes)

You can pause, resume, or cancel an execution at any time.

Saving and Loading

Graphs are saved automatically. You can:

  • Clone a graph to create a copy for experimentation
  • Export a graph as JSON for backup or sharing
  • Import a graph from JSON

Next Steps

Now that you've run your first graph, explore:

  • Cognition nodes — Summarize, Compare, Extract, Classify, Rewrite
  • Control flow — Gate (yes/no decisions), Choice (multi-path routing), Loop (iteration)
  • Terminal nodes — execute commands on the host machine
  • Human checkpoints — pause for approval before critical steps

See Node Reference for the full catalog.

Tip: Start with simple linear graphs (Start → process → End), then add branching and loops as you get comfortable. The most powerful graphs are often just 5-10 well-chosen nodes.

Updated on Mar 21, 2026
QUI