> ## Documentation Index
> Fetch the complete documentation index at: https://docs.veadk.xyz/llms.txt
> Use this file to discover all available pages before exploring further.

# Application logging

VeADK 1.0.5 uses Python's standard `logging` package for runtime logs and no longer depends on Loguru. VeADK loggers live under the `veadk` namespace and write to standard output by default, which works with container platforms and log collectors.

## Set the log level

Set `LOGGING_LEVEL` before importing VeADK. Its default is `DEBUG`.

```bash lines theme={null}
export LOGGING_LEVEL=INFO
python app.py
```

Use standard Python logging levels, including `DEBUG`, `INFO`, `WARNING`, `ERROR`, and `CRITICAL`.

## Log format

Each record contains the time, level, call location, and message:

```text lines theme={null}
2026-07-14 10:30:00 | INFO | runner.py:180 - Event output: request completed
```

When code runs inside a recording OpenTelemetry span, VeADK adds `trace_id` so you can correlate a request between logs and traces:

```text lines theme={null}
2026-07-14 10:30:00 | INFO | runner.py:180 | trace_id=0123456789abcdef - Event output: request completed
```

Interactive terminals use colors for the time, level, and location. Redirected or collected standard output does not contain ANSI color escape sequences.

## Integrate application logs

Your application can use standard-library loggers and control the `veadk` namespace independently:

```python lines theme={null}
import logging

from veadk import Agent

logging.getLogger("veadk").setLevel(logging.INFO)
app_logger = logging.getLogger("company_assistant")

agent = Agent(name="company_assistant")
app_logger.info("Agent initialized")
```

The VeADK handler does not propagate to the Python root logger, which prevents duplicate output when the root logger is configured separately. To send VeADK records to a custom backend, attach a standard `logging` handler to `logging.getLogger("veadk")`.

## Streaming output logs

During streaming execution, `Runner` does not log each partial chunk. Reasoning is logged once when a complete section finishes, while final answers, tool calls, and tool results are recorded as complete events. Events are still streamed unchanged to the caller; this behavior only reduces token-by-token log noise.

<Warning>
  `DEBUG` records can contain model output, reasoning, tool arguments, and tool results. When production traffic contains sensitive data, set `LOGGING_LEVEL` to `INFO` or higher and apply suitable access and retention controls to collected logs.
</Warning>