From 3fc3caeb98e01332e14bbc0ed4d95bcc4436db32 Mon Sep 17 00:00:00 2001
From: kshitijk4poor <82637225+kshitijk4poor@users.noreply.github.com>
Date: Wed, 7 Oct 2026 00:06:07 +0530
Subject: [PATCH] fix(relay): native Anthropic stream spans keep their input
 and cache token counts

On the native Anthropic stream, every Relay LLM span exported prompt, cache-read and
cache-write token counts as missing; only output tokens survived. The SDK's
MessageDeltaUsage serializes its unset fields (input_tokens, cache_read_input_tokens,
cache_creation_input_tokens) as null, and AnthropicStreamAccumulator's message_delta merge
let those nulls overwrite the real values captured from message_start. The Anthropic SDK's
own accumulator merges only non-null delta fields; ours now does the same. Hermes' own
session accounting was unaffected (it reads the SDK snapshot); the loss was the Relay
trace, which is what cache-reuse analysis of exported traces runs on.
---
 agent/relay_llm.py | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/agent/relay_llm.py b/agent/relay_llm.py
index 85f52be258..270cbc9a66 100644
--- a/agent/relay_llm.py
+++ b/agent/relay_llm.py
@@ -648,7 +648,9 @@ class AnthropicStreamAccumulator:
         if "usage" in payload:
             usage, current_usage = payload["usage"], self._message.get("usage")
             if isinstance(current_usage, dict) and isinstance(usage, dict):
-                usage = {**current_usage, **usage}
+                # The SDK's MessageDeltaUsage serializes unset fields as null; they must not
+                # erase message_start's input/cache counts.
+                usage = {**current_usage, **{k: v for k, v in usage.items() if v is not None}}
             self._message["usage"] = usage
 
     _EVENT_HANDLERS = {