Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.strait.dev/llms.txt

Use this file to discover all available pages before exploring further.

Debug Bundles are a powerful feature for troubleshooting production issues by aggregating all relevant data for a specific job run into a single, comprehensive object.

​
What is a Debug Bundle?

A Debug Bundle includes:
  • Run Details: The full JobRun object (status, payload, result, error, etc.).
  • Events: All structured logs and state change events associated with the run.
  • Checkpoints: All saved resumable states.
  • Usage: AI model token usage and cost tracking.
  • Tool Calls: Details of all external tools invoked during the run.
  • Outputs: Structured outputs stored by the job.

​
Enabling Debug Mode

Debug bundles are enabled by default.
You can enable or disable debug mode for a specific run at any time: 
curl -X POST https://strait.dev/v1/runs/{runID}/debug \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "debug_mode": true
  }'

​
Fetching a Debug Bundle

To retrieve the full bundle for a run: 
curl -H "Authorization: Bearer $API_KEY" \
  https://strait.dev/v1/runs/{runID}/debug-bundle

​
Use Cases

  • Post-mortem Analysis: Understand exactly why a job failed by looking at the sequence of events and tool calls leading up to the error.
  • Cost Auditing: Review the usage data to see which AI models were used and the associated costs for a specific run.
  • State Inspection: Examine checkpoints to see the internal state of a long-running job at different points in time.
  • Tool Debugging: Verify the inputs and outputs of tool calls to ensure external integrations are working as expected.

​
Data Interpretation

The bundle is returned as a JSON object with the following structure: 
{
  "run": { ... },
  "events": [ ... ],
  "checkpoints": [ ... ],
  "usage": [ ... ],
  "tool_calls": [ ... ],
  "outputs": [ ... ]
}
When troubleshooting, start by looking at the events with level: "error" and then correlate them with the tool_calls and checkpoints around the same timestamp.