Documentation
tracellm export
Export trace data for analysis and sharing.
Overview
The tracellm export command fetches recent traces from MongoDB and writes them to a file in JSON or CSV format. Exported data includes full trace objects with all steps (JSON) or a flat summary table (CSV).
Syntax
terminalCopy
bash
tracellm export [OPTIONS]
| Option | Type | Default | Description |
|---|---|---|---|
| --format | str | json | Export format: json or csv |
| --limit | int | 100 | Number of recent traces to export (1–1000) |
What Happens Internally
- Validates format — raises an error if the format is not
jsonorcsv. - Fetches recent traces from MongoDB via
fetch_recent_traces(limit). - Creates export directory —
./exports/is created if it does not exist. - Generates a timestamp in the format
YYYYMMDD-HHMMSSfor the filename. - Writes the file:
- JSON — full trace objects with all steps, serialized with model_dump(mode="json")
- CSV — flat fields only: trace_id, project_id, prompt, model, status, latency, tokens, retries, timestamps, step count. Steps are not included.
- Prints success message — shows the number of exported traces and the file path.
Example Output
Terminal outputCopy
text
Exporting traces... ╭── Export Successful ─────────────────────────────────────╮ │ │ │ 12 traces exported │ │ File: exports/traces-20260531-142210.json │ │ │ ╰──────────────────────────────────────────────────────────╯
JSON Export Format
Each exported JSON file contains an array of trace objects. Each trace includes the full execution details:
exported-traces.jsonCopy
json
[
{
"trace_id": "tr_2kf9q3m1",
"project_id": "default",
"project_name": "default",
"environment": "development",
"prompt": "Explain transformers",
"model_name": "gpt-4.1-mini",
"status": "success",
"latency": 3420.0,
"token_count": 1247,
"retry_count": 1,
"slow_request": false,
"failure_reason": null,
"created_at": "2026-05-31T14:22:10",
"updated_at": "2026-05-31T14:22:14",
"steps": [
{
"tool_name": "query.embed",
"duration": 180.0,
"success": true,
"input": {},
"output": { "embedding": [...] }
}
]
}
]CSV Export Format
The CSV format exports flat fields suitable for spreadsheet analysis. Steps are not included — only the step count per trace is recorded.
exported-traces.csvCopy
text
trace_id,project_id,project_name,environment,prompt,model_name,status,latency,token_count,retry_count,slow_request,failure_reason,created_at,updated_at,step_count tr_2kf9q3m1,default,default,development,"Explain transformers",gpt-4.1-mini,success,3420.0,1247,1,false,,2026-05-31T14:22:10,2026-05-31T14:22:14,9 tr_4h8j2p5,default,default,development,"What is attention?",gpt-4.1-mini,success,1890.0,892,0,false,,2026-05-31T14:20:05,2026-05-31T14:20:08,7
Common Errors
| Error | Cause | Fix |
|---|---|---|
| Invalid format | Format is not json or csv | Use --format json or --format csv |
| Export directory unwritable | Permissions issue with ./exports/ | Check directory permissions or create it manually |
| No traces to export | MongoDB is empty | Run some traces first with tracellm trace |
Troubleshooting
Tip
Export files are written to
./exports/ relative to the current working directory. Use absolute paths or run the command from your project root.