AutoGPT is the vision of accessible AI for everyone, to use and to build on. Our mission is to provide the tools, so that you can focus on what matters.
fix(backend/copilot): null-safe token accumulation for OpenRouter null cache fields (#12789)
Zamil Majdyd82ecac363c6
diff --git a/autogpt_platform/backend/backend/copilot/sdk/service.py b/autogpt_platform/backend/backend/copilot/sdk/service.pyindex f4aa019b08..c7d166adba 100644--- a/autogpt_platform/backend/backend/copilot/sdk/service.py+++ b/autogpt_platform/backend/backend/copilot/sdk/service.py@@ -1865,15 +1865,20 @@ async def _run_stream_attempt(# cache_read_input_tokens = served from cache# cache_creation_input_tokens = written to cacheif sdk_msg.usage:- state.usage.prompt_tokens += sdk_msg.usage.get("input_tokens", 0)- state.usage.cache_read_tokens += sdk_msg.usage.get(- "cache_read_input_tokens", 0+ # Use `or 0` instead of a default in .get() because+ # OpenRouter may include the key with a null value (e.g.+ # {"cache_read_input_tokens": null}) for models that don't+ # yet report cache tokens, making .get("key", 0) return+ # None rather than the fallback 0.+ state.usage.prompt_tokens += sdk_msg.usage.get("input_tokens") or 0+ state.usage.cache_read_tokens += (+ sdk_msg.usage.get("cache_read_input_tokens") or 0)- state.usage.cache_creation_tokens += sdk_msg.usage.get(- "cache_creation_input_tokens", 0+ state.usage.cache_creation_tokens += (+ sdk_msg.usage.get("cache_creation_input_tokens") or 0)- state.usage.completion_tokens += sdk_msg.usage.get(- "output_tokens", 0+ state.usage.completion_tokens += (+ sdk_msg.usage.get("output_tokens") or 0)logger.info("%s Token usage: uncached=%d, cache_read=%d, "diff --git a/autogpt_platform/backend/backend/copilot/sdk/service_helpers_test.py b/autogpt_platform/backend/backend/copilot/sdk/service_helpers_test.pyindex be2c46bdbb..5f1487c43b 100644--- a/autogpt_platform/backend/backend/copilot/sdk/service_helpers_test.py+++ b/autogpt_platform/backend/backend/copilot/sdk/service_helpers_test.py@@ -21,6 +21,7 @@ from .service import (_is_tool_only_message,_iter_sdk_messages,_reduce_context,+ _TokenUsage,)# ---------------------------------------------------------------------------@@ -350,3 +351,87 @@ class TestIsParallelContinuation:msg = MagicMock(spec=AssistantMessage)msg.content = [self._make_tool_block()]assert _is_tool_only_message(msg) is True+++# ---------------------------------------------------------------------------+# _TokenUsage — null-safe accumulation (OpenRouter initial-stream-event bug)+# ---------------------------------------------------------------------------+++class TestTokenUsageNullSafety:+ """Verify that ResultMessage.usage dicts with null-valued cache fields+ (as emitted by OpenRouter for the initial streaming event before real+ token counts are available) do not crash the accumulator.++ Before the fix, dict.get("cache_read_input_tokens", 0) returned None+ when the key existed with a null value, causing 'int += None' TypeError.+ """++ def _apply_usage(self, usage: dict, acc: _TokenUsage) -> None:+ """Mirror the production accumulation in sdk/service.py."""+ acc.prompt_tokens += usage.get("input_tokens") or 0+ acc.cache_read_tokens += usage.get("cache_read_input_tokens") or 0+ acc.cache_creation_tokens += usage.get("cache_creation_input_tokens") or 0+ acc.completion_tokens += usage.get("output_tokens") or 0++ def test_null_cache_tokens_do_not_crash(self):+ """OpenRouter initial event: cache keys present with null value."""+ usage = {+ "input_tokens": 0,+ "output_tokens": 0,+ "cache_read_input_tokens": None,+ "cache_creation_input_tokens": None,+ }+ acc = _TokenUsage()+ self._apply_usage(usage, acc) # must not raise TypeError+ assert acc.prompt_tokens == 0+ assert acc.cache_read_tokens == 0+ assert acc.cache_creation_tokens == 0+ assert acc.completion_tokens == 0++ def test_real_cache_tokens_are_accumulated(self):+ """OpenRouter final event: real cache token counts are captured."""+ usage = {+ "input_tokens": 10,+ "output_tokens": 349,+ "cache_read_input_tokens": 16600,+ "cache_creation_input_tokens": 512,+ }+ acc = _TokenUsage()+ self._apply_usage(usage, acc)+ assert acc.prompt_tokens == 10+ assert acc.cache_read_tokens == 16600+ assert acc.cache_creation_tokens == 512+ assert acc.completion_tokens == 349++ def test_absent_cache_keys_default_to_zero(self):+ """Minimal usage dict without cache keys defaults correctly."""+ usage = {"input_tokens": 5, "output_tokens": 20}+ acc = _TokenUsage()+ self._apply_usage(usage, acc)+ assert acc.prompt_tokens == 5+ assert acc.cache_read_tokens == 0+ assert acc.cache_creation_tokens == 0+ assert acc.completion_tokens == 20++ def test_multi_turn_accumulation(self):+ """Null event followed by real event: only real tokens counted."""+ null_event = {+ "input_tokens": 0,+ "output_tokens": 0,+ "cache_read_input_tokens": None,+ "cache_creation_input_tokens": None,+ }+ real_event = {+ "input_tokens": 10,+ "output_tokens": 349,+ "cache_read_input_tokens": 16600,+ "cache_creation_input_tokens": 512,+ }+ acc = _TokenUsage()+ self._apply_usage(null_event, acc)+ self._apply_usage(real_event, acc)+ assert acc.prompt_tokens == 10+ assert acc.cache_read_tokens == 16600+ assert acc.cache_creation_tokens == 512+ assert acc.completion_tokens == 349