Offline Trace Export

If you cannot send traces to Deepchecks in real time - for example, in isolated environments without internet access, or for batch processing workflows - you can export traces to a JSON file locally and upload them in a separate step.

This two-step process works with all framework integrations (LanggraphIntegration, CrewaiIntegration, GoogleAdkIntegration).

Step 1: Export traces to JSON

Instead of calling register_dc_exporter, call register_json_exporter. Traces are saved locally as JSON files as your pipeline runs.

from deepchecks_llm_client.otel import LanggraphIntegration

LanggraphIntegration().register_json_exporter(
    output_folder="./traces",
    log_to_console=True,
)

# Run your pipeline as normal - traces are saved to ./traces/ as JSON files

Step 2: Upload the JSON to Deepchecks

Once you are ready to upload, use the log_spans_file function:

from deepchecks_llm_client.client import DeepchecksLLMClient

dc_client = DeepchecksLLMClient(api_token="your-api-key")

dc_client.log_spans_file(
    app_name="Your App Name",
    version_name="v1",
    env_type="EVAL",
    json_path="./traces/your-trace-file.json",
)

After upload, Deepchecks processes the traces exactly as it would for real-time exports - parsing attributes, computing metrics, calculating properties, and running annotations.