How Do I Place and Connect the Safety Node on the Dave by voolama Workflow Canvas?

Summary

The steps

1. 1

   Confirm your role and open the Workflow Editor

   Only users with the Admin or Curate role can open the Workflow Editor and place nodes on the canvas. If you do not see the Workflows section in the sidebar, your account does not hold one of those roles. The resource at Why Can't I See Agents, Workflows, or Other Sections in Dave by voolama? explains the role-to-section mapping. Ask your Admin to adjust your role under Admin, then Settings, then Users and Roles.

   Dave by voolama has six default roles: Admin, Create, Curate, Use, Reporting, and Financial. All six are editable. For building workflows, you need Admin or Curate. For completing tasks that the Safety node may create at runtime, you need Admin or Use.

   Once you have the right role, navigate to Workflows in the sidebar. Open an existing workflow or create a new one, then click into the Workflow Editor. The editor uses a three-panel layout: the Node Palette on the left, the Canvas in the center, and the Properties Panel on the right.

   Do this

   • Navigate to Workflows in the sidebar. If the section is missing, check your role assignment using the guide linked above.
   • Open an existing workflow or click New Workflow to start one.
   • Click the workflow name to open the Workflow Editor.
   • Confirm the Node Palette is visible on the left. If the canvas is blank, drag a Start node from the palette to begin.

   Example

   A hypothetical marketing agency has an office manager with the Curate role who builds workflows, and two content reviewers with the Use role who complete tasks. The office manager opens the Workflow Editor to add a Safety node to an existing blog-drafting workflow. The reviewers do not see the Workflow Editor, but they will see any tasks the Safety node creates in their Task Inbox.

   Best practice

   • Do not give the Curate role to people whose only job is completing tasks. Curate allows editing workflows; Use is sufficient for task completion.
   • Save the workflow before adding new nodes so you have a clean version to return to if needed. Each save creates an immutable snapshot.
2. 2

   Find the Safety node in the Node Palette: code, icon, and position among all ten node types

   The Workflow Editor's Node Palette on the left side of the canvas lists all ten node types available in Dave by voolama. The Safety node is the tenth entry. Here are all ten, with their code values and icon colors as they appear in the palette:

   • Start (code: start, green play icon): entry point, required once per workflow
   • End (code: end, red square icon): terminal node, marks workflow completion
   • User Interaction (code: user_interaction, blue user icon): pauses the workflow and creates a human task for input
   • Human Review (code: human_review, amber eye icon): creates a review task, reviewer clicks Approve, Reject, or Request Changes
   • Agent Interaction (code: agent_interaction, purple bot icon): invokes an AI agent automatically, requires an agent assignment
   • Agent Review (code: agent_review, pink bot icon): AI-powered quality check against defined criteria, requires an agent assignment
   • Routing (code: routing, purple fork icon): logic-only conditional branching, no user or AI involvement
   • API Call (code: api_call, cyan zap icon): calls an external REST API and maps the response back into the workflow
   • Information (code: information, indigo info icon): documentation node, no effect on execution
   • Safety (code: safety, teal shield icon): content safety and compliance check, flags or blocks policy violations, no agent required

   The Safety node is the only node type in this list that runs an automated compliance check without requiring an agent. This canvas-level checkpoint is one concrete way to implement the human oversight documentation that NIST AI RMF Govern 1.1 calls for in AI risk governance policies, and that EU AI Act Article 14(4) (Regulation (EU) 2024/1689) requires for high-risk AI systems: the checkpoint is visible in the workflow graph, recorded in the audit trail, and configurable per workflow.

   Do this

   • Scroll the Node Palette on the left until you see the Safety entry (teal shield icon, listed tenth).
   • Click and drag it onto the canvas. Drop it below your Agent Interaction node.
   • Click the Safety node to open the Properties Panel on the right.
   • Change the default label to something descriptive, such as "Screen AI Output" or "Safety Check."
   • Review the Properties Panel sections that appear for this node type to explore what is configurable in your installation.

   Example

   A hypothetical small law office uses a workflow where an AI agent drafts a first-pass summary of a client intake form. The office manager places a Safety node immediately after the Agent Interaction node and labels it "Screen Draft Summary." The Safety node runs before the draft reaches the attorney's Human Review task, so the attorney only sees output that has already passed the safety check. The node's presence in the workflow graph gives the firm a documented record of where the screening checkpoint sits, supporting the firm's AI use policy under ABA Model Rule 1.6 (confidentiality of client information).

   Best practice

   • Name the Safety node to reflect what it is checking, not just that it is a safety check. "Screen for Confidential Data" is more useful than "Safety" when a future editor reads the canvas.
   • The Properties Panel silently ignores invalid JSON while you type, but saves only the last valid state. Double-check any JSON config before clicking Save Version.
3. 3

   Connect the Safety node: source handle to target handle, step by step

   In the Workflow Editor, connections flow from the source handle (the small circle at the bottom of a node) to the target handle (the small circle at the top of a node). You cannot connect source-to-source or target-to-target. This direction is fixed: upstream node bottom to downstream node top.

   The standard pattern for a safety-screened workflow is:

   1. Start node: entry point.
   2. User Interaction node (optional): collects input from a person before the agent runs.
   3. Agent Interaction node: invokes the AI agent to produce output.
   4. Safety node: screens the agent's output against configured policies.
   5. Human Review node: a named person reviews the screened output and clicks Approve, Reject, or Request Changes.
   6. End node: marks workflow completion.

   To connect the Agent Interaction node to the Safety node: hover over the bottom handle of the Agent Interaction node, drag to the top handle of the Safety node, and release. Then connect the Safety node's bottom handle to the top handle of the Human Review node. Ensure all paths lead to an End node. A node with no outgoing edge and no End node will cause the workflow instance to hang at runtime.

   The RFP response workflow at Building an RFP Response Workflow in Dave by voolama uses a multi-node pattern with Agent Review before Human Review. The Safety node fits into that pattern in the same position, immediately after the last agent step and before the human decision.

   Do this

   • Hover over the bottom circle (source handle) of the Agent Interaction node.
   • Click and drag to the top circle (target handle) of the Safety node. Release to create the edge.
   • Hover over the bottom circle of the Safety node and drag to the top circle of the Human Review node.
   • Connect the Human Review node to an End node.
   • If the Safety node has multiple output ports (for example, a flagged path and a passed path), connect each port to the appropriate next node. Check the Properties Panel for the port names available in your installation.
   • Click Save Version in the top-right corner of the editor.

   Example

   A hypothetical content agency builds a five-node workflow: Start, Collect Brief (User Interaction), Draft Post (Agent Interaction), Screen Draft (Safety), Review and Approve (Human Review), End. The Safety node sits between the AI output and the human reviewer. If the Safety node flags the draft, the workflow may pause and create a task before the reviewer sees it. If it passes, the draft moves directly to the reviewer's Task Inbox.

   Best practice

   • Save a version after connecting each major section of the graph. If you make a connection mistake, you can load the previous version rather than undoing multiple steps.
   • Use the mini-map in the bottom-right corner of the canvas to navigate large graphs. It shows the full layout and lets you click to jump to a section.
4. 4

   What happens at runtime when the Safety node fires

   When a workflow instance reaches the Safety node at runtime, the node runs its configured content safety and compliance checks on the current workflow output. According to the Workflow Editor documentation, the Safety node can flag or block content that violates configured policies, and it may create a task for a human reviewer.

   The Safety node is one layer in a broader set of AI governance controls in Dave by voolama. The platform also applies company-wide and per-workspace AI guidelines to every agent call before output is produced. Prompt injection detection runs before the agent call is sent to the AI provider, and when SOC 2 compliance mode is enabled this detection is logged as a security event supporting SOC 2 CC7.2 (anomaly monitoring). These platform-level controls are configured by your Admin under Admin, then Settings. The Safety node on the canvas is an additional, explicit checkpoint that you place and control as part of your workflow design.

   For a broader look at how to structure human review so it does not become a bottleneck, see How to Review AI Output Without Becoming the Bottleneck. That piece covers how the Task Inbox turns each review step into a discrete, trackable work item, which applies to tasks the Safety node may create as well as Human Review tasks.

   What the help documentation does not describe: The Workflow Editor help doc does not specify the exact configuration fields for the Safety node's policies, the exact names of its output ports, or the precise content of the task payload when it fires. These details depend on your installation's configuration. Use the Properties Panel when the Safety node is selected to explore what is configurable, and contact your Admin or voolama support if you need clarification on a specific setting.

   Do this

   • Start a test instance of the workflow using a draft workflow version before activating it for production use.
   • Navigate to Tasks in the sidebar to see whether the Safety node creates a task during the test run. Tasks are visible to Admin and Use roles in the Task Inbox.
   • Check the instance status in the Instances section to confirm the workflow progresses past the Safety node as expected.
   • If the Safety node creates a task, open it and review the Payload section to understand what information it surfaces to the reviewer.
   • Adjust the Safety node's configuration in the Properties Panel if the behavior does not match your intent, then save a new version and test again.

   Example

   A hypothetical bookkeeping practice uses a workflow where an AI agent drafts a client-facing summary of monthly transactions. The Safety node is configured to flag output that contains patterns resembling account numbers or personal identifiers. During a test run, the Safety node fires and creates a task. The Admin opens the task in the Task Inbox, sees the flagged content, and adjusts the agent's system prompt to avoid including those patterns in the output. After updating the agent version and saving a new workflow version, the next test run passes the Safety node without creating a task. For accounting practices, IRS Publication 1075 (Tax Information Security Guidelines) requires that systems handling federal tax information include technical controls to detect unauthorized disclosure; a Safety node configured to flag tax identifier patterns is one technical control that supports that requirement.

   Best practice

   • Always test with a draft workflow version before setting the workflow to Active. The Workflow Editor saves versions non-destructively, so you can return to any previous version if a test reveals a problem.
   • Assign the Safety node's task to a specific user or role in the Properties Panel so the right person sees it when it fires, not just any Admin.
5. 5

   When to use Safety versus Agent Review versus Human Review

   Dave by voolama has three node types that evaluate or screen content: Safety, Agent Review, and Human Review. They serve different purposes and can be combined in the same workflow.

   • Safety (teal shield, code: safety). Runs content safety and compliance checks automatically. Does not require an agent. Use it to screen for policy violations, prohibited content, or patterns your organization has defined as off-limits. May create a task when it fires.
   • Agent Review (pink bot icon, code: agent_review). Uses an AI agent to evaluate content quality against defined criteria. Returns a pass or fail result. Use it to check whether a draft meets a quality bar before it reaches a human. Requires an agent with a configured system prompt.
   • Human Review (amber eye icon, code: human_review). Creates a task for a named person who clicks Approve, Reject, or Request Changes. Every decision is recorded in the audit trail with the reviewer's name and a timestamp. Use it for any decision that must be made by a person, not an automated check. EU AI Act Article 14(4) (Regulation (EU) 2024/1689) requires that high-risk AI systems allow humans to effectively oversee them and intervene; the Human Review node is where that intervention is formally recorded in the Dave by voolama audit trail.

   A typical production workflow for AI-generated content uses all three in sequence: Safety screens for policy violations, Agent Review checks quality, and Human Review makes the final decision. This pattern keeps the human reviewer's attention on judgment calls rather than screening tasks. The blog-draft-and-review example at Example Workflow: Blog Draft and Review illustrates a seven-node version of this pattern.

   Do this

   • Map out your workflow before building it in the editor. Mark each step as: automated safety check (Safety node), automated quality check (Agent Review node), or human decision (Human Review node).
   • Place Safety before Agent Review, and Agent Review before Human Review, so automated checks filter out problems before a person's time is spent on them.
   • If your workflow does not need an automated quality check, connect Safety directly to Human Review and omit Agent Review.
   • If your workflow does not need a safety check, connect Agent Interaction directly to Agent Review or Human Review. The Safety node is optional, not required.

   Example

   A hypothetical insurance agency builds a workflow for drafting client communication letters. The workflow uses: Agent Interaction (draft the letter), Safety (screen for prohibited language or personal data patterns), Agent Review (check that the letter addresses all required disclosure points), Human Review (a licensed producer approves before sending). Each automated layer reduces what the producer needs to catch manually. For insurance agencies, the NAIC Model Bulletin on the Use of Artificial Intelligence Systems by Insurers (adopted by several states as of 2024) calls for governance frameworks that include human oversight of AI-generated consumer communications; the Human Review node is where that oversight is recorded in the audit trail.

   Best practice

   • Do not use the Safety node as a substitute for Human Review on decisions that require human judgment. The Safety node checks for policy violations; it does not make business decisions.
   • Review the audit trail after each production run to confirm the Safety node is firing as expected. The audit trail records state transitions and step execution for every instance.

The Safety node on the canvas: code, icon, and what makes it different from Agent Review

All ten node types, and where Safety fits among them

Where this guide stops