First Workflow

In this guide you will build your first Canvas from scratch: a simple RAG flow that retrieves documents and generates a response. By the end you will know how to navigate the editor, configure the fundamental nodes, and publish the flow as an assistant.

Accessing the Canvas

From the left sidebar click Canvas (nodes icon). The list of your Canvases opens. You have two options:

1. From template: click New from Template, choose one of the 6 sector templates and click Use this template.
2. From scratch: click New Canvas, assign a name (e.g. "Basic RAG") and click Create.

This guide starts from scratch.

The Editor Interface

The editor is divided into three zones:

┌────────────┬─────────────────────────────┬──────────────────┐
│  Palette   │        Workspace            │  Config Panel    │
│  (left)    │        (center)             │   (right)        │
│            │                             │                  │
│ ○ Begin    │    [Begin] ──→ [Retrieval]  │  Component:      │
│ ○ LLM      │       ↓                     │  Retrieval       │
│ ○ Retrieval│    [LLM] ──→ [Message]      │  topK: 10        │
│ ○ Message  │                             │  ...             │
│ ○ Switch   │                             │                  │
│ ...        │                             │                  │
└────────────┴─────────────────────────────┴──────────────────┘

Zone	Function
Palette (left)	List of all 15 components. Drag a component into the workspace to add it to the flow.
Workspace (center)	Main working area. You can move nodes, draw edges between output and input ports, and zoom with the mouse wheel.
Config Panel (right)	Clicking on a node opens its configuration panel. All component parameters are editable here.

Workspace navigation:

• Pan: hold and drag to move the view.
• Zoom: mouse wheel or +/- buttons at the bottom right.
• Select node: single click to select it and open the config.
• Move node: drag the node to the desired position.
• Create edge: click on the output port of a node (dot on the right edge), hold and drag to the input port of the next node.
• Delete edge: click on the edge and press the Delete key.

Building the Basic RAG Flow

The flow you will build is:

Begin → Retrieval → LLM → Message

This is the minimal flow for a RAG assistant: it collects the question, retrieves relevant documents, generates a response, and presents it to the user.

Step 1 — Begin Node

The Begin node is already present in the workspace when you create a Canvas. Click it to open the configuration.

Parameter	Value
Mode	conversational
Input Fields	(leave empty for now — the question comes from the chat field)

With conversational mode, the Begin node automatically receives the user's chat text in the variable {{sys.query}}.

Task vs conversational mode

task mode launches the flow once with a structured form (fields defined in Input Fields). conversational mode integrates the Canvas as a chat assistant: the user writes messages and the flow responds. For this first example, conversational is the right choice.

Step 2 — Retrieval Node

From the palette, drag a Retrieval into the workspace. Position it to the right of Begin. Drag an edge from the output port of Begin to the input port of Retrieval.

Click on Retrieval and configure:

Parameter	Value
Query	{{sys.query}}
topK	10
Company Documents	Enabled
Knowledge Base	Enabled
External Sources	Disabled (for now)

The Retrieval node will search the company documents and Knowledge Base for the 10 passages most relevant to the user's question.

Step 3 — LLM Node

Drag an LLM into the workspace, position it to the right of Retrieval, and connect Retrieval → LLM.

Click on LLM and configure:

Parameter	Value
Model	writer
System Prompt	You are a precise corporate assistant. Answer in English based exclusively on the provided documents. If the information is not in the documents, say so explicitly. Always cite sources with [N].
User Prompt	Context documents:\n{{retrieval_1.formalized_content}}\n\nUser question:\n{{sys.query}}
Temperature	0.3
Max Tokens	2048

Variable reference

{{retrieval_1.formalized_content}} is the text retrieved by the Retrieval node, already formatted with document titles and relevant passages. The name retrieval_1 is automatically generated by the Canvas (first Retrieval node in the flow). You can verify it in the node header.

Step 4 — Message Node

Drag a Message into the workspace, position it to the right of LLM, and connect LLM → Message.

Click on Message and configure:

Parameter	Value
Content	{{llm_1.content}}
Format	markdown
Show Citations	Enabled

The Message node collects the LLM output and presents it to the user with Markdown formatting and interactive citations.

Step 5 — Verify the Flow

Before saving, visually check that:

• The edges are: Begin → Retrieval → LLM → Message.
• Every node has all input and output ports correctly connected (unconnected ports appear as empty dots).
• There are no isolated nodes (without edges).

Saving the Canvas

Click Save in the top right (or use Ctrl+S). The Canvas is saved with the current state of the flow.

Auto-save

Queria automatically saves changes every 30 seconds during editing. The save icon at the top indicates the state: full disk = saved, disk with yellow dot = unsaved changes.

Running and Testing

Test Execution from the Editor

Click Run Test (play button at the top). A bottom panel opens with a text field. Write a test question — for example "What are the conditions of the Rossi contract?" — and press Enter.

Watch the flow run:

• Nodes change state: gray (waiting) → yellow (running) → green (completed).
• The bottom panel shows the output of each node as it is produced.
• The final response appears in the Message with Markdown formatting.

If a node shows a red state, click on it to see the error message in the config panel.

Inspecting Node Outputs

During or after execution, click on any node to see:

• The inputs received (with actual variable values).
• The outputs produced (with full values).

This is useful for verifying that variables are populated correctly and that data flows as expected.

Publishing as an Assistant

When the flow works correctly, you can publish it:

1. Click Publish in the top right.
2. Enter an optional description for end users.
3. Choose Visibility:
   • Only me: visible only to you.
   • Team: visible to all users in the organization.
   • Specific topic: available as an assistant for a defined Topic.
4. Click Confirm publication.

Once published, the Canvas appears in the list of available assistants in the chat. Users find it in the Assistants menu at the top of the chat and can select it as a dedicated interlocutor.

Next Steps

You have built your first working Canvas. Now you can:

• Add routing logic: insert a Switch node to handle different types of questions differently.
• Enable external sources: in the Retrieval node, enable Legal Sources, Pharma Sources, or other external databases.
• Add an input form: change the Begin mode to task and define the form fields.
• Explore the examples: the Practical Examples page shows progressively more complex flows.
• Use a template: the Sector Templates page offers ready-made starting points for your sector.

---

Queria v3.1.2 -- Canvas Agent Builder