Unable to disable verbose status streaming to STDOUT

[aaronsteers]

I'm unable to disable verbose status streaming.

Can I disable the streaming of LLM output to STDOUT/STDERR?

Specifically, this output:

I checked docs but could not find any mention of it.

[zzstoatzz]

hi @aaronsteers - you should be able to control this via MARVIN_ENABLE_DEFAULT_PRINT_HANDLER

» uv run python -c "import marvin; result = print(marvin.classify('elmo', ['red', 'blue', 'green']))"
╭─ Agent "Marvin" (cd0968dc) ──────────────────────────────────────────────────────────────────────╮
│                                                                                                  │
│  Tool:    Mark Task f9654b92 ("Classification Task") successful                                  │
│  Status:  ✅                                                                                     │
│  Result   0                                                                                      │
│                                                                                                  │
╰──────────────────────────────────────────────────────────────────────────────────── 10:58:37 PM ─╯
red
» MARVIN_ENABLE_DEFAULT_PRINT_HANDLER=false uv run python -c "import marvin; result = print(marvin.classify('elmo', ['red', 'blue', 'green']))"
red

is there a way that you'd prefer to set this? we will add this to the docs

[aaronsteers]

Thanks, @zzstoatzz! I tested and this works if I set the env var before invoking my program.

It does not appear to work if I try to set os.environ["MARVIN_ENABLE_DEFAULT_PRINT_HANDLER"] = "false" after my code is already running.

In terms of optimal configuration path, I think it could make sense to add into marving.config.settings, or similar to set globally.

If it could be set locally, per command, then an option within the @fn decorator would be great - or configuring within some other object I'd attach to the decorator.

My current implementation looks like this:

"""AI functions for Morph."""

import marvin

...

@marvin.fn(
    instructions=...,
)
def evaluate_mapping_confidence(
    mappings: list[models.FieldMapping],
) -> models.TableMappingEval:
    """Evaluate the confidence of a field mapping configuration.

    ...

    """
    # This function will be implemented by Marvin AI
    ...

...

[zzstoatzz]

hi @aaronsteers - we use pydantic-settings, so this should already be supported

» ipython

#[1]
import marvin

#[2]
marvin.classify('elmo', ['red', 'green', 'blue'])
╭─ Agent "Marvin" (060d5ba4) ──────────────────────────────────────────────────────────────────────╮
│                                                                                                  │
│  Tool:    Mark Task 38eee4bd ("Classification Task") successful                                  │
│  Status:  ✅                                                                                     │
│  Result   0                                                                                      │
│                                                                                                  │
╰──────────────────────────────────────────────────────────────────────────────────── 11:12:04 PM ─╯
Out[2]: 'red'

#[3]
marvin.settings.enable_default_print_handler = False

#[4]
marvin.classify('elmo', ['red', 'green', 'blue'])
Out[4]: 'red'

[aaronsteers]

Perfect! It worked like a charm! ✅

[zzstoatzz]

🎉