Running and Interacting with Agent Workflows

This guide explains how to execute tasks and interact with your configured Agent Workflows (agent teams) in the Bourne AI console.

Accessing the Workflow

  1. Navigate to the Agent Workflow List: Go to the main "Agent" section from the dashboard.
  2. Select the Workflow: Locate the workflow you wish to run in the list.
  3. Click the Workflow Title: Click on the title to open the workflow's detail page.

This page typically displays the chat interface and task execution details.

Understanding the Interface

The workflow detail page usually consists of two main panels:

  • Task/Timeline Panel (Left): Shows the sequence of messages, agent actions, and thoughts as the workflow progresses.
  • Terminal/Results Panel (Right): Displays the detailed output of the currently selected step from the left panel (e.g., agent's reasoning, tool usage, final output, search results).
  • Input Bar (Bottom): Where you type your messages and control the interaction mode.

Initiating a Task

  1. Locate the Input Bar: Find the message input field at the bottom of the screen.
  2. Select Interaction Mode (Optional but Recommended):
    • Agent Mode: (Usually the default or toggled on) Select this mode when you want the entire workflow to process your input based on its configuration (Sequential or Supervisor). Use the toggle switch [ToggleSwitch]
    • Chat Mode: Select this mode to have a direct conversation based on the existing message history. The system will review the context of the previous messages and provide a response, without executing the full agent workflow again. Use the toggle switch [ToggleSwitch]
    • For most tasks where you want the agent team to collaborate, use Agent Mode.
  3. Type Your Task or Prompt: Enter the initial request, question, or instruction for the agent team in the input field.
  4. Send the Message:
    • Press the Enter key.
    • Click the Send button

Tool Execution Control: Ask vs Auto-Accept Mode

When your workflow is in Agent Mode, you can control how tool executions are handled using the Tool Approval toggle (visible only when Agent Mode is active).

Auto Mode (Default)

  • What It Does: Tools execute automatically without requiring your approval
  • When to Use: For read-only operations, research tasks, or when you trust the agent's tool usage decisions
  • Toggle Position: Off (left side of toggle)
  • Indicator: Toggle shows "Auto" label

Ask Mode (Require Approval)

  • What It Does: System pauses and requests your approval before executing each tool
  • When to Use: For sensitive operations (posting to social media, sending emails, making API changes), testing new workflows, or when you want manual oversight
  • Toggle Position: On (right side of toggle)
  • Indicator: Toggle shows "Require Approval" label

How Approval Requests Work

  1. Agent decides to use a tool (e.g., "Post to LinkedIn")
  2. Execution pauses and an approval request appears in the timeline
  3. You see:
    • Which tool the agent wants to use
    • The parameters/content for the tool
    • Approve or Deny buttons
  4. Click Approve to proceed or Deny to skip that tool execution
  5. Workflow continues based on your decision

Example Approval Request:

Agent wants to use: Post to LinkedIn
Content: "Excited to announce our new product launch..."
[Approve] [Deny]

Tip: Use Ask Mode when testing new workflows or when agents are performing actions that affect external systems.

Deep Research Mode

For supervisor-mode workflows configured with Deep Research enabled, the agent orchestrates a comprehensive multi-phase research process.

What is Deep Research Mode?

Deep Research Mode activates an enhanced research workflow designed for thorough investigation of complex topics. Instead of a single query-response cycle, the agent:

  1. Creates a Research Plan: Breaks down the topic into specific research questions
  2. Executes Parallel Research: Delegates research tasks to gather comprehensive information
  3. Synthesizes Findings: Compiles and analyzes all gathered data into a cohesive report

How It Works

Phase 1: Planning

  • Agent analyzes your research request
  • Creates a detailed research plan with specific subtopics
  • Presents the plan in the timeline (you'll see "Research Plan Presented")

Phase 2: Execution

  • Agent executes research steps (web search, document analysis, data gathering)
  • Uses multiple tools and sources to ensure comprehensive coverage
  • Collects facts, statistics, quotes, and examples

Phase 3: Synthesis

  • Agent compiles all research findings
  • Synthesizes information into a structured report
  • Provides citations and source links
  • Presents final comprehensive answer

When to Use Deep Research

  • Complex Topics: Multi-faceted questions requiring investigation of several angles
  • Comprehensive Reports: When you need thorough analysis, not just quick answers
  • Data-Heavy Research: Gathering statistics, trends, and evidence across multiple sources
  • Comparative Analysis: Researching multiple options or competitors
  • Due Diligence: In-depth investigation before decisions

Example Use Cases

  • "Research the top 10 AI automation platforms, comparing features, pricing, and customer reviews"
  • "Investigate best practices for remote team management across different industries"
  • "Analyze market trends for electric vehicles in Europe over the past 3 years"
  • "Compare regulatory requirements for data privacy across US, EU, and Asia"

Note: Deep Research Mode is only available in Supervisor-mode workflows where the supervisor agent has been configured with the "Deep Research" setting enabled.

Monitoring Execution

Once you send the message in Agent Mode:

  • Timeline Updates (Left Panel): You will see new entries appearing in the timeline panel. These represent:
    • Your initial message (User).
    • Agents starting their tasks (Agent Name badge).
    • Tool executions with specific tool badges (e.g., Search Web, Post to LinkedIn, Get Instagram Comments)
    • Tool approval requests (if in Ask Mode)
    • Intermediate thoughts or outputs from individual agents.
    • Supervisor decisions (if in Supervisor mode).
    • Research plan presentations (if in Deep Research mode)
    • A loading indicator appears while the workflow is actively processing.
  • Terminal View Updates (Right Panel): This panel shows the detailed output corresponding to the currently selected step in the left timeline. You can click on different timeline entries to see their specific details:
    • Agent's full reasoning and decision-making process
    • Complete tool outputs (search results, API responses, generated content)
    • Generated documents or content
    • Error messages or debugging information (if tool execution fails)
    • Research findings and synthesis (in Deep Research mode)
    • Use the Back
    • Use the Copy button

Viewing the Final Result

The final output of the workflow will typically appear as the last message(s) in the timeline panel from the designated final agent or the supervisor.

Continuing the Conversation

After a workflow run completes, you have options for follow-up:

  • Ask a follow-up question based on the conversation: Switch to Chat Mode (using the toggle or shortcut ⌘K / Alt+K). Type your follow-up question in the input bar and send it. The system reviews the existing conversation context and provides a relevant response, without re-running the entire workflow.
  • Start a new task with the same workflow: Keep Agent Mode selected (or toggle back to it using ⌘L / Alt+L). Typing a new message and sending it in this mode allows the supervisor logic to intelligently continue the workflow based on the previous conversation history and the last state, targeting specific tasks as needed. To start a truly fresh run from the beginning, use the Clear Messages button first.

Clearing the Conversation

To start a fresh interaction with no prior context:

  1. Click the Clear Messages button (eraser icon) located near the input bar
  2. Or use keyboard shortcut: ⌘⇧D (Mac) or Ctrl+Shift+D (Windows/Linux)

What This Does:

  • Removes all messages from the current session view in the timeline
  • Clears the conversation history that agents use for context
  • Resets the workflow state to start fresh
  • Does NOT delete the workflow configuration or settings

When to Use Clear Messages:

  • Starting a completely new, unrelated task
  • Troubleshooting workflow issues by removing context conflicts
  • Resetting after a long conversation to improve agent focus
  • Testing workflow behavior from a clean state

Note: After clearing, the next message sent in Agent Mode will start the workflow from the beginning with no prior context.