Langfuse observability

Langfuse integration

from langfuse import Langfuse
from supercompress import Compressor

langfuse = Langfuse()
comp = Compressor()

def traced_compress(context, query, trace_id):
    result = comp.compress(context, query)
    langfuse.trace(id=trace_id, name="compress",
        input={"context": context, "query": query},
        output={"compressed": result.compressed_text},
        metadata={
            "original_tokens": result.original_tokens,
            "kept_tokens": result.kept_tokens,
            "savings": f"{result.tokens_removed} tokens"
        })
    return result

Frequently asked questions

Does Langfuse support cost tracking with compression?

Yes. Log tokens saved and calculate cost savings in your Langfuse dashboard.

Can I see per-user compression metrics?

Yes. Add user_id to the trace metadata for per-user analysis.