Skip to main content

​
Task Execution

Learn how Xagent plans, executes, and manages your tasks from start to finish.

​
Execution Phases

​
1. Planning Phase

When you submit a task, Xagent first analyzes your request and creates an execution plan. What happens:
  • Xagent understands your goal
  • Breaks down the task into steps
  • Identifies required tools
  • Generates a DAG (Directed Acyclic Graph)
The DAG (Execution Plan) The DAG is a visual representation of your task workflow: Execution Plan Understanding the DAG:
  • Nodes (circles/boxes) - Individual execution steps
  • Edges (arrows) - Dependencies between steps
  • Flow - Execution order from start to finish

​
2. Execution Phase

Once the plan is ready, Xagent begins executing each step. What happens:
  • Steps execute in dependency order
  • Tools are called automatically
  • Results are passed between steps
  • Progress updates in real-time
Execution features:
  • Real-time progress tracking
  • Step-by-step logging
  • Error handling and retry
  • Dynamic adaptation

​
3. Completion Phase

After all steps complete, Xagent compiles the final result. What happens:
  • Results from all steps are aggregated
  • Final output is generated
  • Execution summary is displayed
  • Task is saved for reference

​
The Execution Interface

When your task runs, you’ll see the Agent Interface with a vertical step layout: Task Execution Interface

​
Chat Area (Top)

  • Your original request
  • Real-time status updates
  • Final results
  • Follow-up conversation
  • File attachments

​
Execution Steps (Main Area)

  • Vertical flow of execution steps
  • Each step displays:
    • Step number and title
    • Agent reasoning and output
    • Tool calls (shown inline with step content)
    • Tool responses and results
    • Execution status indicators
  • Expandable for detailed view

​
Understanding Tool Calls

When a step uses tools, you’ll see: 
Step 1: Planning
[Agent reasoning about the task...]

🔧 web_search
  Query: "AI developments 2024"
  Results: [search results...]

🔧 browse_results
  URL: [specific page]
  Content: [page content...]
Common tools you’ll see:
  • web_search - Search the web for information
  • browse_results - Read specific web pages
  • file_read - Read uploaded files
  • code_execution - Run code snippets
  • Various API integrations

​
Monitoring Execution

​
Step Status Indicators

  • 🔵 Pending - Step waiting to run
  • 🟡 Running - Step currently executing
  • 🟢 Completed - Step finished successfully
  • 🔴 Failed - Step encountered an error

​
Real-Time Updates

Watch your task execute:
  • Steps change color as they progress
  • New steps appear as dependencies resolve
  • Results populate in real-time
  • Tokens and time tracked

​
Interacting During Execution

​
Send Follow-up Messages

Add context or modify the task:
  • “Focus on the European market”
  • “Add more visual elements”
  • “Include competitor analysis”

​
Upload Files

Add more resources mid-execution:
  • Click attachment icon
  • Select files to upload
  • Reference them in follow-up

​
Cancel Execution

Stop the task if needed:
  • Click cancel/stop button
  • Current step completes
  • Partial results saved

​
Post-Execution Actions

​
View Full Execution

After completion:
  • Review all steps
  • Check tool usage
  • Analyze performance
  • Export execution log

​
Start New Task

Create additional tasks:
  • Return to Task page
  • Submit a new request
  • Build on previous results
  • Refine with follow-ups

​
Performance Metrics

Each execution shows: Token Usage
  • Input tokens consumed
  • Output tokens generated
  • Total across all steps
  • Cost estimate
Execution Time
  • Start and end timestamps
  • Duration per step
  • Total execution time
Resource Usage
  • Tools called
  • Files accessed
  • APIs used
  • Memory consumed

​
Troubleshooting

​
Task Not Starting

Check:
  • LLM provider configured
  • API key valid
  • Request is clear and specific

​
Execution Stuck

Solutions:
  • Wait longer (complex tasks take time)
  • Check for infinite loops
  • Cancel and retry
  • Simplify the request

​
Step Failed

What to do:
  1. Check error message in right panel
  2. Verify tool availability
  3. Review input data
  4. Request retry or modify approach

​
DAG Not Loading

Try:
  • Refresh the page
  • Check browser console
  • Clear cache
  • Try different browser

​
Best Practices

​
1. Be Specific

Clear requests lead to better plans:
  • “Analyze Q4 sales by region and create charts”
  • Not: “Analyze sales”

​
2. Break Down Complex Tasks

For very complex goals:
  • Start with a high-level request
  • Let Xagent create the plan
  • Review the DAG
  • Send follow-ups to refine

​
3. Provide Context

Give Xagent what it needs:
  • Upload relevant files
  • Reference specific data sources
  • Set constraints and preferences
  • Define output format

​
4. Monitor Progress

Watch the execution:
  • Check DAG for bottlenecks
  • Review step results
  • Catch errors early
  • Adjust as needed

​
Next Steps