Fix stagehand.metrics

Author: miguelg719

.changeset/dog-of-interesting-energy.md

@@ -0,0 +1,5 @@
+---
+"stagehand": patch
+---
+
+Fix stagehand.metrics on env:BROWSERBASE

pyproject.toml

@@ -9,7 +9,19 @@ description = "Python SDK for Stagehand"
 readme = "README.md"
 classifiers = [ "Programming Language :: Python :: 3", "License :: OSI Approved :: MIT License", "Operating System :: OS Independent",]
 requires-python = ">=3.9"
-dependencies = [ "httpx>=0.24.0", "python-dotenv>=1.0.0", "pydantic>=1.10.0", "playwright>=1.42.1", "requests>=2.31.0", "browserbase>=1.4.0", "rich>=13.7.0", "openai>=1.83.0", "anthropic>=0.51.0", "litellm>=1.72.0",]
+dependencies = [
+ "httpx>=0.24.0",
+ "python-dotenv>=1.0.0",
+ "pydantic>=1.10.0",
+ "playwright>=1.42.1",
+ "requests>=2.31.0",
+ "browserbase>=1.4.0",
+ "rich>=13.7.0",
+ "openai>=1.83.0",
+ "anthropic>=0.51.0",
+ "litellm>=1.72.0",
+ "nest-asyncio>=1.6.0",
+]
 [[project.authors]]
 name = "Browserbase, Inc."
 email = "[email protected]"

stagehand/api.py

@@ -5,7 +5,7 @@
 
 from .utils import convert_dict_keys_to_camel_case
 
-__all__ = ["_create_session", "_execute"]
+__all__ = ["_create_session", "_execute", "_get_replay_metrics"]
 
 
 async def _create_session(self):
@@ -177,3 +177,91 @@ async def _execute(self, method: str, payload: dict[str, Any]) -> Any:
     except Exception as e:
         self.logger.error(f"[EXCEPTION] {str(e)}")
         raise
+
+
+async def _get_replay_metrics(self):
+    """
+    Fetch replay metrics from the API endpoint /sessions/:id/replay and parse them
+    into StagehandMetrics format.
+    """
+    from .metrics import StagehandMetrics
+
+    if not self.session_id:
+        raise ValueError("session_id is required to fetch metrics.")
+
+    headers = {
+        "x-bb-api-key": self.browserbase_api_key,
+        "x-bb-project-id": self.browserbase_project_id,
+        "Content-Type": "application/json",
+    }
+
+    try:
+        response = await self._client.get(
+            f"{self.api_url}/sessions/{self.session_id}/replay",
+            headers=headers,
+        )
+
+        if response.status_code != 200:
+            error_text = (
+                await response.aread() if hasattr(response, "aread") else response.text
+            )
+            self.logger.error(
+                f"[HTTP ERROR] Failed to fetch metrics. Status {response.status_code}: {error_text}"
+            )
+            raise RuntimeError(
+                f"Failed to fetch metrics with status {response.status_code}: {error_text}"
+            )
+
+        data = response.json()
+
+        if not data.get("success"):
+            raise RuntimeError(
+                f"Failed to fetch metrics: {data.get('error', 'Unknown error')}"
+            )
+
+        # Parse the API data into StagehandMetrics format
+        api_data = data.get("data", {})
+        metrics = StagehandMetrics()
+
+        # Parse pages and their actions
+        pages = api_data.get("pages", [])
+        for page in pages:
+            actions = page.get("actions", [])
+            for action in actions:
+                # Get method name and token usage
+                method = action.get("method", "").lower()
+                token_usage = action.get("tokenUsage", {})
+
+                if token_usage:
+                    input_tokens = token_usage.get("inputTokens", 0)
+                    output_tokens = token_usage.get("outputTokens", 0)
+                    time_ms = token_usage.get("timeMs", 0)
+
+                    # Map method to metrics fields
+                    if method == "act":
+                        metrics.act_prompt_tokens += input_tokens
+                        metrics.act_completion_tokens += output_tokens
+                        metrics.act_inference_time_ms += time_ms
+                    elif method == "extract":
+                        metrics.extract_prompt_tokens += input_tokens
+                        metrics.extract_completion_tokens += output_tokens
+                        metrics.extract_inference_time_ms += time_ms
+                    elif method == "observe":
+                        metrics.observe_prompt_tokens += input_tokens
+                        metrics.observe_completion_tokens += output_tokens
+                        metrics.observe_inference_time_ms += time_ms
+                    elif method == "agent":
+                        metrics.agent_prompt_tokens += input_tokens
+                        metrics.agent_completion_tokens += output_tokens
+                        metrics.agent_inference_time_ms += time_ms
+
+                    # Always update totals for any method with token usage
+                    metrics.total_prompt_tokens += input_tokens
+                    metrics.total_completion_tokens += output_tokens
+                    metrics.total_inference_time_ms += time_ms
+
+        return metrics
+
+    except Exception as e:
+        self.logger.error(f"[EXCEPTION] Error fetching replay metrics: {str(e)}")
+        raise

stagehand/main.py

@@ -7,6 +7,7 @@
 from typing import Any, Optional
 
 import httpx
+import nest_asyncio
 from dotenv import load_dotenv
 from playwright.async_api import (
     BrowserContext,
@@ -16,7 +17,7 @@
 from playwright.async_api import Page as PlaywrightPage
 
 from .agent import Agent
-from .api import _create_session, _execute
+from .api import _create_session, _execute, _get_replay_metrics
 from .browser import (
     cleanup_browser_resources,
     connect_browserbase_browser,
@@ -206,7 +207,7 @@ def __init__(
         )
 
         # Initialize metrics tracking
-        self.metrics = StagehandMetrics()
+        self._local_metrics = StagehandMetrics()  # Internal storage for local metrics
         self._inference_start_time = 0  # To track inference time
 
         # Validate env
@@ -372,26 +373,26 @@ def update_metrics(
             inference_time_ms: Time taken for inference in milliseconds
         """
         if function_name == StagehandFunctionName.ACT:
-            self.metrics.act_prompt_tokens += prompt_tokens
-            self.metrics.act_completion_tokens += completion_tokens
-            self.metrics.act_inference_time_ms += inference_time_ms
+            self._local_metrics.act_prompt_tokens += prompt_tokens
+            self._local_metrics.act_completion_tokens += completion_tokens
+            self._local_metrics.act_inference_time_ms += inference_time_ms
         elif function_name == StagehandFunctionName.EXTRACT:
-            self.metrics.extract_prompt_tokens += prompt_tokens
-            self.metrics.extract_completion_tokens += completion_tokens
-            self.metrics.extract_inference_time_ms += inference_time_ms
+            self._local_metrics.extract_prompt_tokens += prompt_tokens
+            self._local_metrics.extract_completion_tokens += completion_tokens
+            self._local_metrics.extract_inference_time_ms += inference_time_ms
         elif function_name == StagehandFunctionName.OBSERVE:
-            self.metrics.observe_prompt_tokens += prompt_tokens
-            self.metrics.observe_completion_tokens += completion_tokens
-            self.metrics.observe_inference_time_ms += inference_time_ms
+            self._local_metrics.observe_prompt_tokens += prompt_tokens
+            self._local_metrics.observe_completion_tokens += completion_tokens
+            self._local_metrics.observe_inference_time_ms += inference_time_ms
         elif function_name == StagehandFunctionName.AGENT:
-            self.metrics.agent_prompt_tokens += prompt_tokens
-            self.metrics.agent_completion_tokens += completion_tokens
-            self.metrics.agent_inference_time_ms += inference_time_ms
+            self._local_metrics.agent_prompt_tokens += prompt_tokens
+            self._local_metrics.agent_completion_tokens += completion_tokens
+            self._local_metrics.agent_inference_time_ms += inference_time_ms
 
         # Always update totals
-        self.metrics.total_prompt_tokens += prompt_tokens
-        self.metrics.total_completion_tokens += completion_tokens
-        self.metrics.total_inference_time_ms += inference_time_ms
+        self._local_metrics.total_prompt_tokens += prompt_tokens
+        self._local_metrics.total_completion_tokens += completion_tokens
+        self._local_metrics.total_inference_time_ms += inference_time_ms
 
     def update_metrics_from_response(
         self,
@@ -426,9 +427,9 @@ def update_metrics_from_response(
                     f"{completion_tokens} completion tokens, {time_ms}ms"
                 )
                 self.logger.debug(
-                    f"Total metrics: {self.metrics.total_prompt_tokens} prompt tokens, "
-                    f"{self.metrics.total_completion_tokens} completion tokens, "
-                    f"{self.metrics.total_inference_time_ms}ms"
+                    f"Total metrics: {self._local_metrics.total_prompt_tokens} prompt tokens, "
+                    f"{self._local_metrics.total_completion_tokens} completion tokens, "
+                    f"{self._local_metrics.total_inference_time_ms}ms"
                 )
             else:
                 # Try to extract from _hidden_params or other locations
@@ -736,7 +737,50 @@ def page(self) -> Optional[StagehandPage]:
 
         return self._live_page_proxy
 
+    def __getattribute__(self, name):
+        """
+        Intercept access to 'metrics' to fetch from API when use_api=True.
+        """
+        if name == "metrics":
+            use_api = (
+                object.__getattribute__(self, "use_api")
+                if hasattr(self, "use_api")
+                else False
+            )
+
+            if use_api:
+                # Need to fetch from API
+                try:
+                    # Get the _get_replay_metrics method
+                    get_replay_metrics = object.__getattribute__(
+                        self, "_get_replay_metrics"
+                    )
+
+                    # Try to get current event loop
+                    try:
+                        asyncio.get_running_loop()
+                        # We're in an async context, need to handle this carefully
+                        # Create a new task and wait for it
+                        nest_asyncio.apply()
+                        return asyncio.run(get_replay_metrics())
+                    except RuntimeError:
+                        # No event loop running, we can use asyncio.run directly
+                        return asyncio.run(get_replay_metrics())
+                except Exception as e:
+                    # Log error and return empty metrics
+                    logger = object.__getattribute__(self, "logger")
+                    if logger:
+                        logger.error(f"Failed to fetch metrics from API: {str(e)}")
+                    return StagehandMetrics()
+            else:
+                # Return local metrics
+                return object.__getattribute__(self, "_local_metrics")
+
+        # For all other attributes, use normal behavior
+        return object.__getattribute__(self, name)
+
 
 # Bind the imported API methods to the Stagehand class
 Stagehand._create_session = _create_session
 Stagehand._execute = _execute
+Stagehand._get_replay_metrics = _get_replay_metrics