Skip to main content

Artifacts & Debugging

Dhenara can capture debug artifacts per model call:

  • The normalized Dhenara request (prompt/messages/instructions + call config)
  • The provider-formatted request
  • The raw provider response
  • The parsed Dhenara response
  • Optional Python logs (captured per call)

This makes it much easier to debug provider quirks, reproduce failures, and compare behavior across models.

Enable artifacts for a call​

from dhenara.ai import AIModelClient
from dhenara.ai.types import AIModelCallConfig
from dhenara.ai.types.genai.dhenara.request import ArtifactConfig

client = AIModelClient(
    model_endpoint=endpoint,
    config=AIModelCallConfig(
        streaming=False,
        artifact_config=ArtifactConfig(
            enabled=True,
            artifact_root="./runs/my_debug_run/turn_01",
            prefix="call_001",
            capture_dhenara_request=True,
            capture_provider_request=True,
            capture_provider_response=True,
            capture_dhenara_response=True,
        ),
    ),
)

resp = client.generate(prompt="Write a short spec for a CLI tool")

Where artifacts are written​

Artifacts are written under:

  • <artifact_root>/<prefix>/dai/

Each stage is captured as JSON (and optional JSONL for Python logs).

Capture Python logs (optional)​

If you want a call-scoped log capture (useful when debugging retries, request translation, parsing, etc.):

from dhenara.ai.types.genai.dhenara.request import ArtifactConfig

ArtifactConfig(
    enabled=True,
    artifact_root="./runs/my_debug_run/turn_01",
    enable_python_logs=True,
    python_log_level="INFO",
    python_logger_levels={
        # Example: reduce noise from a chatty logger
        "httpx": "WARNING",
    },
)
  • Keep artifacts off by default in production
  • Enable them per request when diagnosing issues
  • Use a structured directory per workflow/turn (makes diffing easier)

See runnable examples​

  • packages/dhenara_ai/examples/16_multi_turn_with_tools_and_messages_api.py
  • packages/dhenara_ai/examples/19_streaming_multi_turn_structured_thinking.py