browser-use · MagMueller · Feb 23, 2025 · Feb 23, 2025 · Feb 23, 2025 · Feb 23, 2025
diff --git a/browser_use/agent/message_manager/service.py b/browser_use/agent/message_manager/service.py
@@ -238,7 +238,7 @@ def _count_text_tokens(self, text: str) -> int:
 
 	def cut_messages(self):
 		"""Get current message list, potentially trimmed to max tokens"""
-		diff = self.state.history.total_tokens - self.settings.max_input_tokens
+		diff = self.state.history.current_tokens - self.settings.max_input_tokens
 		if diff <= 0:
 			return None
 
@@ -252,9 +252,9 @@ def cut_messages(self):
 					msg.message.content.remove(item)
 					diff -= self.settings.image_tokens
 					msg.metadata.tokens -= self.settings.image_tokens
-					self.state.history.total_tokens -= self.settings.image_tokens
+					self.state.history.current_tokens -= self.settings.image_tokens
 					logger.debug(
-						f'Removed image with {self.settings.image_tokens} tokens - total tokens now: {self.state.history.total_tokens}/{self.settings.max_input_tokens}'
+						f'Removed image with {self.settings.image_tokens} tokens - total tokens now: {self.state.history.current_tokens}/{self.settings.max_input_tokens}'
 					)
 				elif 'text' in item and isinstance(item, dict):
 					text += item['text']
@@ -290,7 +290,7 @@ def cut_messages(self):
 		last_msg = self.state.history.messages[-1]
 
 		logger.debug(
-			f'Added message with {last_msg.metadata.tokens} tokens - total tokens now: {self.state.history.total_tokens}/{self.settings.max_input_tokens} - total messages: {len(self.state.history.messages)}'
+			f'Added message with {last_msg.metadata.tokens} tokens - total tokens now: {self.state.history.current_tokens}/{self.settings.max_input_tokens} - total messages: {len(self.state.history.messages)}'
 		)
 
 	def _remove_last_state_message(self) -> None:

diff --git a/browser_use/agent/message_manager/tests.py b/browser_use/agent/message_manager/tests.py
@@ -174,7 +174,7 @@ def test_token_overflow_handling_with_real_flow(message_manager: MessageManager,
 			else:
 				raise e
 
-		assert message_manager.state.history.total_tokens <= message_manager.settings.max_input_tokens + 100
+		assert message_manager.state.history.current_tokens <= message_manager.settings.max_input_tokens + 100
 
 		last_msg = messages[-1]
 		assert isinstance(last_msg, HumanMessage)
@@ -217,7 +217,7 @@ def test_token_overflow_handling_with_real_flow(message_manager: MessageManager,
 		assert f'step {i}' in messages[-1].content  # Should contain current step info
 
 		# Log token usage for debugging
-		token_usage = message_manager.state.history.total_tokens
+		token_usage = message_manager.state.history.current_tokens
 		token_limit = message_manager.settings.max_input_tokens
 		# print(f'Step {i}: Using {token_usage}/{token_limit} tokens')
 
@@ -231,7 +231,7 @@ def test_token_overflow_handling_with_real_flow(message_manager: MessageManager,
 			real_tokens.append(message_manager._count_tokens(msg.message))
 		assert total_tokens == sum(real_tokens)
 		assert stored_tokens == real_tokens
-		assert message_manager.state.history.total_tokens == total_tokens
+		assert message_manager.state.history.current_tokens == total_tokens
 
 
 # pytest -s browser_use/agent/message_manager/tests.py
diff --git a/browser_use/agent/message_manager/views.py b/browser_use/agent/message_manager/views.py
@@ -64,7 +64,7 @@ class MessageHistory(BaseModel):
 	"""History of messages with metadata"""
 
 	messages: list[ManagedMessage] = Field(default_factory=list)
-	total_tokens: int = 0
+	current_tokens: int = 0
 
 	model_config = ConfigDict(arbitrary_types_allowed=True)
 
@@ -74,7 +74,7 @@ def add_message(self, message: BaseMessage, metadata: MessageMetadata, position:
 			self.messages.append(ManagedMessage(message=message, metadata=metadata))
 		else:
 			self.messages.insert(position, ManagedMessage(message=message, metadata=metadata))
-		self.total_tokens += metadata.tokens
+		self.current_tokens += metadata.tokens
 
 	def add_model_output(self, output: 'AgentOutput') -> None:
 		"""Add model output as AI message"""
@@ -103,20 +103,20 @@ def get_messages(self) -> list[BaseMessage]:
 
 	def get_total_tokens(self) -> int:
 		"""Get total tokens in history"""
-		return self.total_tokens
+		return self.current_tokens
 
 	def remove_oldest_message(self) -> None:
 		"""Remove oldest non-system message"""
 		for i, msg in enumerate(self.messages):
 			if not isinstance(msg.message, SystemMessage):
-				self.total_tokens -= msg.metadata.tokens
+				self.current_tokens -= msg.metadata.tokens
 				self.messages.pop(i)
 				break
 
 	def remove_last_state_message(self) -> None:
 		"""Remove last state message from history"""
 		if len(self.messages) > 2 and isinstance(self.messages[-1].message, HumanMessage):
-			self.total_tokens -= self.messages[-1].metadata.tokens
+			self.current_tokens -= self.messages[-1].metadata.tokens
 			self.messages.pop()
 
 

diff --git a/browser_use/agent/service.py b/browser_use/agent/service.py
@@ -4,6 +4,7 @@
 import json
 import logging
 import re
+import time
 from pathlib import Path
 from typing import Any, Awaitable, Callable, Dict, Generic, List, Optional, Type, TypeVar
 
@@ -31,6 +32,7 @@
 	AgentSettings,
 	AgentState,
 	AgentStepInfo,
+	StepMetadata,
 	ToolCallingMethod,
 )
 from browser_use.browser.browser import Browser
@@ -301,6 +303,7 @@ async def step(self, step_info: Optional[AgentStepInfo] = None) -> None:
 		state = None
 		model_output = None
 		result: list[ActionResult] = []
+		step_start_time = time.time()
 
 		try:
 			state = await self.browser_context.get_state()
@@ -361,6 +364,7 @@ async def step(self, step_info: Optional[AgentStepInfo] = None) -> None:
 			self.state.last_result = result
 
 		finally:
+			step_end_time = time.time()
 			actions = [a.model_dump(exclude_unset=True) for a in model_output.action] if model_output else []
 			self.telemetry.capture(
 				AgentStepTelemetryEvent(
@@ -375,7 +379,13 @@ async def step(self, step_info: Optional[AgentStepInfo] = None) -> None:
 				return
 
 			if state:
-				self._make_history_item(model_output, state, result)
+				metadata = StepMetadata(
+					step_number=self.state.n_steps,
+					step_start_time=step_start_time,
+					step_end_time=step_end_time,
+					input_tokens=self._message_manager.state.history.current_tokens,
+				)
+				self._make_history_item(model_output, state, result, metadata)
 
 	@time_execution_async('--handle_step_error (agent)')
 	async def _handle_step_error(self, error: Exception) -> list[ActionResult]:
@@ -417,6 +427,7 @@ def _make_history_item(
 		model_output: AgentOutput | None,
 		state: BrowserState,
 		result: list[ActionResult],
+		metadata: Optional[StepMetadata] = None,
 	) -> None:
 		"""Create and store history item"""
 
@@ -433,7 +444,7 @@ def _make_history_item(
 			screenshot=state.screenshot,
 		)
 
-		history_item = AgentHistory(model_output=model_output, result=result, state=state_history)
+		history_item = AgentHistory(model_output=model_output, result=result, state=state_history, metadata=metadata)
 
 		self.state.history.history.append(history_item)
 

diff --git a/browser_use/agent/views.py b/browser_use/agent/views.py
@@ -93,6 +93,20 @@ class ActionResult(BaseModel):
 	include_in_memory: bool = False  # whether to include in past messages as context or not
 
 
+class StepMetadata(BaseModel):
+	"""Metadata for a single step including timing and token information"""
+
+	step_start_time: float
+	step_end_time: float
+	input_tokens: int  # Approximate tokens from message manager for this step
+	step_number: int
+
+	@property
+	def duration_seconds(self) -> float:
+		"""Calculate step duration in seconds"""
+		return self.step_end_time - self.step_start_time
+
+
 class AgentBrain(BaseModel):
 	"""Current state of the agent"""
 
@@ -135,6 +149,7 @@ class AgentHistory(BaseModel):
 	model_output: AgentOutput | None
 	result: list[ActionResult]
 	state: BrowserStateHistory
+	metadata: Optional[StepMetadata] = None
 
 	model_config = ConfigDict(arbitrary_types_allowed=True, protected_namespaces=())
 
@@ -166,6 +181,7 @@ def model_dump(self, **kwargs) -> Dict[str, Any]:
 			'model_output': model_output_dump,
 			'result': [r.model_dump(exclude_none=True) for r in self.result],
 			'state': self.state.to_dict(),
+			'metadata': self.metadata.model_dump() if self.metadata else None,
 		}
 
 
@@ -174,6 +190,30 @@ class AgentHistoryList(BaseModel):
 
 	history: list[AgentHistory]
 
+	def total_duration_seconds(self) -> float:
+		"""Get total duration of all steps in seconds"""
+		total = 0.0
+		for h in self.history:
+			if h.metadata:
+				total += h.metadata.duration_seconds
+		return total
+
+	def total_input_tokens(self) -> int:
+		"""
+		Get total tokens used across all steps.
+		Note: These are from the approximate token counting of the message manager.
+		For accurate token counting, use tools like LangChain Smith or OpenAI's token counters.
+		"""
+		total = 0
+		for h in self.history:
+			if h.metadata:
+				total += h.metadata.input_tokens
+		return total
+
+	def input_token_usage(self) -> list[int]:
+		"""Get token usage for each step"""
+		return [h.metadata.input_tokens for h in self.history if h.metadata]
+
 	def __str__(self) -> str:
 		"""Representation of the AgentHistoryList object"""
 		return f'AgentHistoryList(all_results={self.action_results()}, all_model_outputs={self.model_actions()})'

diff --git a/browser_use/telemetry/views.py b/browser_use/telemetry/views.py
@@ -1,8 +1,6 @@
 from abc import ABC, abstractmethod
 from dataclasses import asdict, dataclass
-from typing import Any, Dict, Optional
-
-from browser_use.controller.registry.views import ActionModel
+from typing import Any, Dict, Sequence
 
 
 @dataclass
@@ -57,5 +55,5 @@ class AgentEndTelemetryEvent(BaseTelemetryEvent):
 	steps: int
 	max_steps_reached: bool
 	success: bool
-	errors: list[str]
+	errors: Sequence[str | None]
 	name: str = 'agent_end'