Trace Explorer

Monitor and inspect your LLM traces in real time

Overview

The Trace Explorer is the central view for monitoring your LLM applications. Every interaction captured by the Python SDK or Proxy integration appears here as a structured trace with full request-response data, performance metrics, and detection results.

Open the Trace Explorer at app.avaliar.ai/traces.

Trace Types

Avaliar captures traces from two integration paths. Each path stores slightly different metadata depending on the source.

SDK Traces

Traces created by the @traceable decorator in the Python SDK. These include:

Field	Description
Trace ID	Unique identifier for the trace
Span type	llm, tool, agent, or generic
Model	The model used (e.g., gpt-4o, claude-sonnet-4-20250514)
Provider	The LLM provider (e.g., openai, anthropic)
Input	The prompt or messages sent to the model
Output	The model’s response
Latency	End-to-end execution time
Tokens	Prompt tokens, completion tokens, and total
Cost	Estimated cost based on model pricing
Detection mode	local, cloud, or none
Environment	development, staging, or production
Issues	Any safety issues detected

Proxy Traces

Traces captured by the Avaliar Proxy gateway. These include:

Field	Description
Trace ID	Unique identifier for the trace
Model	The target model
Provider	The LLM provider
Messages	The full prompt messages
Response	The model’s completion
Latency	Total round-trip time including proxy overhead
Tokens	Prompt tokens, completion tokens, and total
Cost	Estimated cost based on model pricing
Status	HTTP status code from the LLM provider
Client IP	The originating IP address
Environment	Inherited from the API key used
Issues	Any safety issues detected

Filtering and Search

The Trace Explorer provides several ways to find specific traces:

• Search — Search by trace ID or model name using the search bar
• Environment filter — Filter by development, staging, or production
• Severity filter — Show only traces with issues of a specific severity level
• Issue type filter — Filter by issue type (e.g., prompt_injection, toxicity, pii)
• Sort — Sort results by timestamp, model, latency, or cost

Combine multiple filters to narrow down to exactly the traces you need.

Live Mode

Enable Live Mode to auto-refresh the Trace Explorer every 30 seconds. New traces appear at the top of the list without a manual page reload.

Live Mode is useful when you are actively testing or deploying changes and want to watch traces arrive in real time.

Trace Detail View

Click any trace to open the detail view. The detail panel shows:

1. Prompt and Response

   The full prompt (or message array) and the model’s response. Responses are rendered as Markdown for readability.

2. Detected Issues

   A list of all issues found by the detection pipeline. Each issue shows its type, severity badge, confidence score, and the excerpt that triggered the finding.

3. Performance Metrics

   Key metrics for the trace:

   Metric	Description
   Latency	End-to-end execution time in milliseconds
   Prompt tokens	Number of tokens in the input
   Completion tokens	Number of tokens in the output
   Total tokens	Combined token count
   Cost	Estimated cost in USD

4. Trace Tree

   For SDK traces with nested spans, the trace tree shows the full execution flow as a parent-child hierarchy. Each node displays its span type, latency, and status.

Use the trace tree view to see the full execution flow of nested @traceable calls. This is especially useful for debugging multi-step pipelines like RAG workflows or agent loops.

Trace States

Every trace moves through one of three states after ingestion:

State	Badge	Description
Pending	Pending	The detection pipeline is still processing this trace.
Issues Found	Issues	One or more safety issues were detected. The severity badge reflects the highest severity found.
Clean	Clean	Detection completed and no issues were found.

Traces in the Pending state typically resolve within a few seconds. If a trace stays pending for longer than 30 seconds, check the Architecture overview for details on async processing.

Next Steps

Analytics

View aggregated metrics and trends across all your traces.

Alerts

Configure notifications for safety issues detected in your traces.