Trace Explorer

The Trace Explorer is your primary tool for investigating AI agent activity. Search across all traces, apply filters, and drill into details.

Overview

Access the Trace Explorer from the main navigation sidebar. It provides:

• Table View - List all traces with sortable columns
• Advanced Filters - Narrow results by multiple criteria
• Detail Drawer - Full trace information without leaving the page
• Bulk Actions - Select and operate on multiple traces

Table View

Columns

The trace table displays key information:

Column	Description
Name	Trace/operation name
Agent	Agent that executed the trace
Status	Success, Warning, or Error
Duration	Execution time
Tokens	Input + Output tokens
Cost	Estimated cost
Time	When the trace occurred
Source	Integration source (Langfuse, Databricks, etc.)

Sorting

Click any column header to sort:

• Click once: Sort ascending
• Click twice: Sort descending
• Click third: Remove sort

Pagination

Navigate through results:

• Page Size: 25, 50, 100 rows per page
• Navigation: First, Previous, Next, Last buttons
• Jump to Page: Enter specific page number

Filtering

Quick Filters

Use the filter bar above the table:

┌─────────────────────────────────────────────────────────────────┐
│  🔍 Search...    Time: [Last 24h ▼]    Status: [All ▼]    ...  │
└─────────────────────────────────────────────────────────────────┘

Available Filters

Filter	Options
Search	Full-text search across all fields
Time Range	Last hour, 24h, 7d, 30d, Custom
Status	Success, Warning, Error, All
Agent	Select from discovered agents
Source	Filter by integration
Has Violations	Only traces with policy violations

Advanced Filters

Click "Advanced" to access more filter options:

• Duration Range - Min/max execution time
• Token Range - Min/max token count
• Cost Range - Min/max cost
• Custom Attributes - Filter by metadata fields

Combining Filters

Filters combine with AND logic:

Time: Last 24 hours
AND Status: Error
AND Agent: CustomerService

Saving Filter Presets

Save frequently used filter combinations:

1. Configure your filters
2. Click "Save Preset"
3. Name your preset
4. Access from the preset dropdown

Search

Full-Text Search

The search box searches across:

• Trace name/operation
• Agent name
• Input/output content
• Metadata values
• Error messages

Search Syntax

Use operators for precise searches:

Operator	Example	Description
Exact phrase	"customer query"	Match exact phrase
OR	error OR failed	Match either term
NOT	NOT test	Exclude term
Field	agent:ChatBot	Search specific field

Examples

# Find traces with errors from ChatBot
agent:ChatBot status:error

# Find traces mentioning "payment" but not "test"
payment NOT test

# Find exact operation name
"process_order"

Trace Details

Opening Details

Click any trace row to open the detail drawer:

┌──────────────────────────────────────────────────────────────┐
│  Trace Details                                         [×]    │
├──────────────────────────────────────────────────────────────┤
│  CustomerService Agent                                        │
│  Operation: customer_support_query                            │
│  Status: ● Success                                            │
│                                                               │
│  [Overview] [Timeline] [Input/Output] [Metadata] [Violations] │
│                                                               │
│  ... (tab content) ...                                        │
└──────────────────────────────────────────────────────────────┘

Overview Tab

Key metrics at a glance:

• Trace ID - Unique identifier
• Started - Start timestamp
• Duration - Total execution time
• Status - Success/Warning/Error
• Agent - Agent name and version
• Model - LLM model used
• Tokens - Input/Output counts
• Cost - Estimated cost

Timeline Tab

Span breakdown showing execution flow:

customer_support_query [1.23s]
├── fetch_user_context [0.05s]
├── call_llm [0.85s] ─── OpenAI GPT-4
│   ├── input_tokens: 850
│   └── output_tokens: 400
├── query_knowledge_base [0.25s]
└── format_response [0.08s]

Each span shows:

• Name and duration
• Start/end times
• Attributes and metadata
• Nested child spans

Input/Output Tab

View the actual data:

Input:

{
  "user_message": "How do I reset my password?",
  "session_id": "sess_abc123",
  "context": { ... }
}

Output:

{
  "response": "To reset your password, follow these steps...",
  "confidence": 0.95,
  "sources": [ ... ]
}

Metadata Tab

Custom attributes attached to the trace:

agent:
  name: CustomerService
  version: 2.1.0
  type: RAG

model:
  name: gpt-4-turbo
  provider: openai
  temperature: 0.7

user:
  id: user_12345
  type: human
  tier: premium

session:
  id: sess_abc123
  channel: web

Violations Tab

Policy violations for this trace:

┌─────────────────────────────────────────────────────────────┐
│  Policy Violations (2)                                       │
├─────────────────────────────────────────────────────────────┤
│  ⚠ MEDIUM: Token Limits Exceeded                            │
│  Policy: Max 2000 tokens, Actual: 2150 tokens               │
├─────────────────────────────────────────────────────────────┤
│  🔴 HIGH: PII Detected                                       │
│  Pattern: Email address in output                            │
│  Evidence: "user@example.com"                                │
└─────────────────────────────────────────────────────────────┘

Actions

Single Trace Actions

Right-click a trace or use the action menu:

• View Details - Open detail drawer
• Copy Trace ID - Copy to clipboard
• View in Langfuse - Open in source platform
• Compare - Add to comparison selection
• Export - Download trace data

Bulk Actions

Select multiple traces using checkboxes:

1. Check individual traces or "Select All"
2. Use the bulk action menu:
   • Export Selected - Download as JSON/CSV
   • Compare - Open comparison view
   • Acknowledge Violations - Mark violations as reviewed

Comparison Mode

Compare multiple traces side-by-side:

1. Select 2-4 traces
2. Click "Compare"
3. View comparison modal:

┌──────────────────────────────────────────────────────────────┐
│  Compare Traces                                        [×]    │
├──────────────────┬─────────────┬─────────────┬──────────────┤
│  Metric          │  Trace 1    │  Trace 2    │  Trace 3     │
├──────────────────┼─────────────┼─────────────┼──────────────┤
│  Duration        │  1.23s      │  2.15s      │  0.89s       │
│  Tokens          │  1,250      │  2,100      │  950         │
│  Cost            │  $0.042     │  $0.068     │  $0.031      │
│  Status          │  Success    │  Warning    │  Success     │
│  Model           │  gpt-4      │  gpt-4      │  gpt-4-turbo │
└──────────────────┴─────────────┴─────────────┴──────────────┘

Keyboard Shortcuts

Shortcut	Action
/	Focus search
Esc	Clear selection / Close drawer
↑ ↓	Navigate rows
Enter	Open selected trace
c	Compare selected traces
e	Export selected
r	Refresh data

Tips & Best Practices

1. Start Broad, Then Narrow

Begin with a wide time range, then add filters to narrow down.

2. Use Saved Presets

Save filter combinations you use regularly:

• "Production Errors (Last 24h)"
• "My Agent (Last Week)"
• "High-Cost Traces"

3. Leverage Comparison

Compare successful vs failed traces to identify issues:

1. Filter to errors
2. Find a similar successful trace
3. Compare to spot differences

4. Check Violations First

When investigating issues, start with the Violations tab.

---

Next Steps

• Agent Activity - Visualize agent behavior
• Policy Violations - Understand violation details
• Troubleshooting - Common issues and solutions