Skip to main content
After instrumenting your application, Braintrust captures every request as traces that you can view, filter, and analyze in real time. This observability enables you to monitor production behavior, identify issues, and gather data for improving your application.

​
Why observe in Braintrust

Observability in Braintrust creates a feedback loop between production and evaluation. Logs use the same data structure as experiments, which means:
  • Instrumentation code works for both logging and evaluation
  • Traces capture identical data in production and testing
  • Scores and feedback apply to both logs and experiments
  • Production data seamlessly becomes evaluation datasets
This unified structure lets you iterate faster and maintain consistency across your development workflow.

​
View your logs

The Logs page displays all traces from your application in a searchable, filterable table. Each row represents a complete trace with its root span. You can:
  • Browse traces and individual spans, in the UI or from the terminal with bt view logs.
  • Group related traces by metadata or tags.
  • Create custom columns to surface important values.
  • Extract prompts to iterate in playgrounds.
  • Apply tags to organize traces.
  • Download logs as CSV or JSON, or pull them locally with bt sync pull.

​
Use Loop

Loop is Braintrust’s AI agent. It understands your data structure, so you can investigate your logs through natural language instead of writing queries. Loop is available on the Logs page, on individual trace pages, and on its own full page, and it’s the engine behind both Patterns and the Debugger. See What Loop can do for the full range, including building datasets, writing scorers, and running evaluations from what it finds.

​
Choose an analysis approach

Reading every trace stops being practical quickly. Braintrust offers three ways to apply intelligence to your logs, at different scales. You can use any one of them on its own, or use them together. None requires the others.

​
Investigate with Patterns

Patterns puts Loop to work on your trace backlog on a schedule. It searches for recurring problems and trends that no scorer already measures, then records each one with the traces that support it and a suggested fix. See Get started with Patterns for details.

​
Discover insights with Topics

Topics automatically analyze and classify your logs without manual review. Each trace is analyzed by facets that extract short labels, then similar labels are clustered into named topics like user intents, sentiment, and issues. Use Topics for:
  • Blind-spot detection: Surface user request types and system gaps you didn’t know to look for.
  • Silent failure detection: Catch quality issues that don’t trip explicit checks.
  • Product roadmap signals: Cluster real user requests into themes to inform what to build.
  • Targeted evaluation datasets: Filter classified logs to build datasets for focused evals.
Built-in facets include Task (user intents), Sentiment (emotional tone), and Issues (agent problems). You can also create custom facets for domain-specific analysis. Where Topics tells you what your traces are about, Patterns tells you what deserves your attention. See Patterns and Topics for how the two compare, and Get started with Topics for details. Find specific traces using multiple approaches:
  • Filters: Use the Filter menu for point-and-click filtering, or write SQL for precise queries. Apply via the UI, CLI, or API.
  • Common filters: Paste ready-made filter patterns for typical queries.
  • Saved views: Save useful combinations as custom table views.
  • Loop: Ask natural language questions about your logs.

​
Monitor with dashboards

Custom dashboards aggregate metrics across your logs and experiments. Track request counts, latency, token usage, costs, scores, and custom metrics over time. Dashboards help you:
  • Visualize trends and anomalies
  • Compare performance across time periods
  • Drill into specific data points
  • Share insights with your team

​
Next steps