> ## Documentation Index
> Fetch the complete documentation index at: https://docs.wisdom.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Visual Mode
This guide walks through building an agent using Visual Mode in WisdomAI, from selecting a trigger to publishing your workflow for delivery. If you prefer to describe what you want and let the agent figure out the steps on every run, see [Create an agent - Prompt Mode](/advanced-features/agents/tutorial-create-agent).
The Agents feature is currently in beta.
## Before you begin
Make sure you have the following:
* An active WisdomAI account with access to the **Agents** tab.
* A domain connected to the data you want the agent to work with.
* A delivery channel ready to configure: email or Slack.
## Step 1: Open the Agent Builder
Visual Mode is the default view when you open the Agent Builder. The node canvas opens automatically and is ready for you to build.
1. Navigate to the **Agents** tab in the left sidebar and click **Create Agent**.
2. Select the **domain** your agent will run in.
3. (Optional) Edit your agent's name by clicking on the pencil icon. Add it in the **Agent Settings** modal and click **Save**.
## Step 2: Generate your workflow with the WisdomAI Builder (recommended)
Rather than placing nodes manually, you can describe your agent in plain language and let the **WisdomAI Builder** generate the structure for you. The WisdomAI Builder is an AI-powered chat panel built into the canvas. Type what you want your agent to do in the **Describe your agent** bar at the bottom of the screen. WisdomAI Builder reads your connected domain and data sources, constructs the complete node structure, and renders it on the canvas for you to review, adjust, and publish.
You can use the WisdomAI Builder at any point, such as to create an agent from scratch or to modify an existing workflow. For example: *add a conditional branch that skips delivery if revenue is above target.*
If you prefer to place and configure nodes yourself, or want to refine what the Builder generated, continue to Step 3.
## Step 3: Build manually
Use manual building when you want to refine the workflow the Builder generated — swap a data source, add a condition, change an instruction — or when you prefer to place and configure each node yourself. You can build manually from the start or at any time after the Builder generates your workflow.
Every agent starts with a trigger node, which defines when the agent runs.
1. Click the trigger node on the canvas. The **Select trigger** panel opens on the right.
2. Choose a trigger type.
| Trigger | When to use |
| ------------------ | -------------------------------------------------------- |
| **Manual trigger** | Use this when you need this analysis to run on demand. |
| **Schedule** | Use this when you want a recurring analysis on schedule. |
3. If you select the Schedule trigger and want it to act only when a condition is met, enter it in the **Trigger Criteria** field. Leave it empty to run unconditionally every time the trigger fires.
Start with a single condition. Once you confirm it fires correctly, add more. Debugging a multi-condition trigger from scratch is significantly harder than layering conditions one at a time.
Here, also define when the agent runs: set a start date and time, and a frequency. Match the frequency to how often your data changes and how quickly you need to act on it.
4. Click **Done**.
### Add and connect nodes
Nodes are the building blocks of a Visual Mode agent. Each node performs a specific task. They are connected through Edges, and these also define both the order of execution and how data flows between them.
For a full description of node types and how they behave, see the [Nodes Reference](/advanced-features/agents/nodes-reference).
To add a node:
1. Click the **+** button on the trigger node to open the **Add node** menu.
2. Select the node type you want to add.
3. Click on it to write an instruction in the node's settings panel describing what it should do with its input.
4. Click **Done**.
5. To add another node, click the circular button on any node's output and extend the arrow to add the next node in the flow. Repeat until your workflow is complete.
You don't have to finish the entire workflow before testing. Click **Preview** at any point to run the agent to its current state and see what each node produces. This makes it easier to catch problems early, before adding more nodes on top.
When you hover the mouse over a node, the **Delete node** icon appears. Click on it. In the modal that opens, select **Delete** to confirm the operation. **This action is irreversible**.
## Step 4: Configure the Artifact Builder
The Artifact Builder formats your workflow's output into a finished report for delivery.
1. Add the **Artifact Builder** node to the canvas, then click it to open its settings.
2. Write an instruction that covers two things separately: how to organize the data into sections, lists, and tables; and how to format the final output: headings, number formats, and layout.
Be explicit about formatting. The Artifact Builder does not infer structure on its own. Include details like:
* Section headings: "Section 1: Overview", "Section 2: Details"
* Table columns: "Include columns: Deal Name, Stage, Days Stuck, Owner"
* Number formats: "Show currency as \$X,XXX" and "Show percentages as X.X%"
* Conditional sections: "Only include the Incidents section if incidents occurred."
3. Click **Done**.
Every agent requires at least one Artifact Builder node and one delivery node before it can be published.
## Step 5: Configure delivery
The delivery node sends your artifact to its destination.
1. Click **+** after the Artifact Builder node and select **Email**, **Slack**, or **Webhook**.
2. Enter the destination — an email address, Slack channel name, or webhook URL.
3. Click **Done**.
To deliver to Slack, a workspace administrator must first connect your Slack workspace to WisdomAI. See [Connect to Slack](/integrations/connect-to-slack).
## Step 6: Preview and test
At any point before publishing, you can run the agent manually to verify that the output is correct. To do this:
1. Click **Preview** in the top right corner.
2. You don't have to wait for the run to finish. As the agent executes, watch the canvas for progress. A green check confirms a node succeeded, a red X indicates a failure. Click the node to see the error.
3. At any point during or after the run, click on a node to review the data generated, the analysis performed, and the underlying SQL or Python code used.
4. Click **End Preview** to return to the edit mode and adjust node instructions, trigger criteria, or formatting as needed, then preview again.
### Common fixes
| Problem | Fix |
| -------------------------- | ---------------------------------------------------------------------------------- |
| Output is too broad | Add filtering criteria and exclusions to the instruction |
| Data is missing | Name the specific fields or columns you need |
| Formatting is inconsistent | Add explicit section headings, column names, and number formats to the instruction |
| Trigger fires too often | Tighten the threshold value |
| Trigger never fires | Loosen the threshold value |
## Step 7: Publish your Agent
Once the preview output looks correct, make the agent live. To do that, click the **Publish** button. Your agent will become active instantly.
The agent will run on the schedule you configured (unless you have chosen the manual trigger) and deliver output to the method you set. You can monitor past runs, edit the agent, or adjust the schedule at any time from the **Agents** tab.
## Next steps
Learn how to edit, share, and view the run history of your agents.
Explore the full list of node types, inputs, and outputs.