Magentic-One compression

Magentic-One integration

from autogen_ext.models import SuperCompressMixin
from supercompress import Compressor

class CompressedOrchestrator(MagenticOneOrchestrator):
    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)
        self.comp = Compressor()

    async def _process_agent_output(self, agent, output, context):
        compressed = self.comp.compress(
            context.get_full_context(),
            output.get_content()
        )
        context.update(output, compressed.compressed_text)
        return await super()._process_agent_output(
            agent, output, context
        )

Frequently asked questions

Does compression break agent coordination?

No. Only irrelevant context is removed. Coordination messages are preserved.

Can I use it with the Magentic-One CLI?

Yes. The compression happens at the agent context level, not the CLI level.