AI Agent Setup and Workflow Access

Learn how to enable the AI Agent, give it access to your workflows, and have conversational interactions that execute your automations.

AI Agent Setup and Workflow Access

The FlowDot AI Agent is a conversational interface that can run your workflows through natural language. Instead of filling out forms, just ask!

What the Agent Can Do

Step 1: Enable Agent Access for Workflows

By default, workflows are not available to the Agent. You must explicitly enable them.

Method 1: From the Open Modal

  1. Click the hamburger menu (top right)
  2. Click Open to see your workflows
  3. Find your workflow in the list
  4. Click the star icon next to it
  5. The workflow is now Agent-enabled

Method 2: From the Editor

  1. Open your workflow in the Editor
  2. Click Settings (gear icon)
  3. Enable Agent Access
  4. Save the workflow

What Happens When Enabled

Step 2: Using the Agent

Accessing the Agent

Click the Agent button in the main navigation (top left) or go to /agent.

The Agent Interface

Element Purpose
Chat Input Type your requests
Model Selector FlowDot, Simple, Capable, Complex
Favorite Workflows Shows "X of Y workflows enabled"
Conversation History Past messages

Your First Agent Interaction

  1. Enable your Goat Simulator workflow (from earlier tutorials)
  2. Go to the Agent
  3. Type: "What can you do?"

The Agent will respond listing:

Step 3: Running Workflows Through the Agent

Direct Invocation

Ask the Agent to run a specific workflow:

You: Run the goat simulator with the question "Do you prefer hay or grain?"

The Agent:

  1. Recognizes your intent
  2. Identifies the correct workflow
  3. Maps your input to the workflow's input schema
  4. Executes the workflow
  5. Returns the results

Conversational Invocation

Or just ask naturally:

You: Ask the goat what its favorite food is.

The Agent figures out:

Step 4: Switching Models Mid-Conversation

You can change the Agent's model anytime:

  1. FlowDot: Free, good for basic conversations
  2. Simple: Fast responses, simple tasks
  3. Capable: Most use cases
  4. Complex: Advanced reasoning, tool use

Tip: Start with Simple or FlowDot, then switch to Complex when you need sophisticated tool selection.

Example Conversation

[FlowDot model]
You: What can you do?
Agent: I can help you with... [lists capabilities]

[Switch to Complex]
You: Get the weather in Buffalo, NY, then tell me what activities a goat would enjoy.

Agent: [Runs weather workflow, interprets results, then runs goat simulator with weather context]

Step 5: Chaining Workflows

The Agent excels at combining multiple workflows.

Example: Weather + Goat Advisor

Setup:

Conversation:

You: What's the weather in Cincinnati?
Agent: [Runs weather workflow] It's currently 72°F with scattered clouds...

You: Based on that weather, what would a goat want to do?
Agent: [Runs goat simulator with weather context] Baa! With this lovely 72°F weather, I'd definitely want to graze in the meadow...

Explicit Chaining

You can also be explicit:

You: First get the weather in NYC, then post that to my Discord channel.

The Agent:

  1. Runs weather workflow
  2. Takes the output
  3. Runs Discord posting workflow with that data

Step 6: Managing Workflow Access

Viewing Enabled Workflows

In the Agent interface:

  1. Click the workflow count (e.g., "2 of 5 workflows")
  2. See all your workflows
  3. Toggle individual workflows on/off
  4. Temporarily disable without unfavoriting

Best Practices

Do Don't
Enable focused, well-named workflows Enable everything
Use clear input/output descriptions Leave default names
Test workflows before enabling Enable untested workflows
Group related workflows Mix unrelated tools

Step 7: Debugging Agent Interactions

"The Agent didn't use my workflow"

  1. Check workflow is favorited/enabled
  2. Verify your request clearly relates to the workflow
  3. Use a more capable model (Complex is best for tool use)
  4. Be more explicit: "Use my [Workflow Name] workflow"

"The Agent used wrong inputs"

  1. Check your workflow's input names are descriptive
  2. Add descriptions to input nodes
  3. Use the Text Output node's Description field
  4. Be more specific in your request

"Results aren't helpful"

  1. Workflow may return raw data (expected)
  2. Ask: "Summarize those results" or "What does that mean?"
  3. The Agent can interpret any workflow output

Pro Tips

Designing Agent-Friendly Workflows

  1. Clear names: "Weather Lookup" not "wf_001"
  2. Descriptive inputs: "City Name" not "text_input"
  3. Output context: Use Description fields on outputs
  4. Single purpose: One workflow = one task

Workflow Output for Agent Use

The MCP Output setting controls what the Agent sees:

  1. Open a node's Properties
  2. Find MCP Output
  3. Enable for data you want the Agent to access

This is separate from Frontend Mode (what users see in dashboards).

Summary

The AI Agent transforms how you interact with workflows:

Next, learn about publishing workflows and community features!

Related Tutorials

Back to Tutorials