Scrubbing sensitive content

[adtyavrdhn]

Fixes #1571

Adds include_content flag in InstrumentationSettings, default set to True

Adds include_tool_args to remove the arguments from the running tool spans when incude_content is set to False

Adds example in logfire docs

[hyperlint-ai]

PR Change Summary

Introduced a new flag to manage sensitive content in instrumentation settings and updated documentation accordingly.

• Added include_sensitive_content flag in InstrumentationSettings, default set to True
• Implemented guard clauses to scrub sensitive content while retaining the event
• Updated logfire documentation with an example for excluding sensitive content

Modified Files

• docs/logfire.md

---

How can I customize these reviews?

Check out the Hyperlint AI Reviewer docs for more information on how to customize the review.

If you just want to ignore it on this PR, you can add the hyperlint-ignore label to the PR. Future changes won't trigger a Hyperlint review.

Note specifically for link checks, we only check the first 30 links in a file and we cache the results for several hours (for instance, if you just added a page, you might experience this). Our recommendation is to add hyperlint-ignore to the PR to ignore the link check for this PR.

[alexmojaki]

Tool call arguments also need to be omitted.

[alexmojaki]

The tool call arguments also need to be excluded from the 'running tool' span.

[alexmojaki]

Reminder: The tool call arguments also need to be excluded from the 'running tool' span.

[alexmojaki]

This implies that the number of events produced will depend on whether content is included if multiple tool calls are made in a single message.

[adtyavrdhn]

Trying to understand what you mean exactly

Would you say the below test captures what you are trying to imply:

def test_otel_events_consistency_with_include_content():
    """Test that the number of OpenTelemetry events is consistent regardless of include_content setting."""

    # Create a response with multiple tool calls followed by text
    response = ModelResponse(parts=[
        ToolCallPart('tool1', {'arg1': 'value1'}, 'call_1'),
        ToolCallPart('tool2', {'arg2': 'value2'}, 'call_2'),
        TextPart('Some text response')
    ])

    settings_with_content = InstrumentationSettings(include_content=True)
    events_with_content = response.otel_events(settings_with_content)

    settings_without_content = InstrumentationSettings(include_content=False)
    events_without_content = response.otel_events(settings_without_content)

    assert len(events_with_content) == len(events_without_content), (
        f"Event count differs: with_content={len(events_with_content)}, "
        f"without_content={len(events_without_content)}"
    )

[alexmojaki]

sorry, i didn't think this through properly. the difference would be if a message had multiple text parts. but i don't think this happens in practice, so it isn't worth worrying about.

[dhimmel]

the difference would be if a message had multiple text parts. but i don't think this happens in practice

I'm not caught up on this whole conversation, but I do think a message with multiple text parts could be realistic. For example, multiple text parts could be used to separate different inputs:

• text part: compare these two paragraphs, which is better?
• text part: paragraph 1
• text part: paragraph 2

Of course this could be done by concatenating all text into a single part.

[adtyavrdhn]

Yep, I get it. Something like this would cause this issue.

    response = ModelResponse(parts=[
        TextPart('Some text response'),
        ToolCallPart('tool2', {'arg2': 'value2'}, 'call_2'),
        TextPart('Some more text response'),
        ToolCallPart('tool1', {'arg1': 'value1'}, 'call_1'),
        TextPart('Even more text response'),
    ])

[alexmojaki]

Thanks!