Traffic logs used to truncate request and response bodies at 50KB. Anything bigger was silently cut off — exactly when you needed it most, for the long-context calls and agent runs that are the hardest to debug.
That cap is gone. Bodies are now stored in full, regardless of size.
What changed — typical small requests stay where they were: inline in the log row, instantly visible. Large bodies are transparently offloaded to dedicated cold storage purpose-built for this kind of payload, then streamed back into the detail sheet on demand. You see the full request and full response, every time.
Why it matters — agent traces with 200KB of tool-call history, RAG calls stuffed with documents, long-context prompts that push past 100KB — all now fully inspectable, replayable, and addable to datasets. No more "looks like this got cut off" guessing.
Performance — opening the detail sheet stays snappy. The first paint uses the inline metadata and a preview; the full body slides in moments later. Proxy latency is completely unchanged — the offload happens after we've already responded to your request.
Nothing for you to configure. Just send large bodies and they'll be there when you come looking.