Get helicone costs (#220)

Signed-off-by: Merwane Hamadi <merwanehamadi@gmail.com>

agbenchmark/get_data_from_helicone.py

@@ -0,0 +1,53 @@
+import os
+from typing import Optional
+
+import requests
+
+from agbenchmark.start_benchmark import BENCHMARK_START_TIME
+
+
+def get_data_from_helicone(challenge: str) -> Optional[float]:
+    # Define the endpoint of your GraphQL server
+    url = "https://www.helicone.ai/api/graphql"
+
+    # Set the headers, usually you'd need to set the content type and possibly an authorization token
+    headers = {"authorization": "Bearer {os.environ.get('HELICONE_API_KEY')}"}
+
+    # Define the query, variables, and operation name
+    query = """
+query ExampleQuery {
+  aggregatedHeliconeRequest {
+    cost
+  }
+}
+"""
+
+    variables = {
+        "filters": [
+            {
+                "property": {
+                    "value": {"equals": os.environ.get("AGENT_NAME")},
+                    "name": "agent",
+                }
+            },
+            {
+                "property": {
+                    "value": {"equals": BENCHMARK_START_TIME},
+                    "name": "benchmark_start_time",
+                }
+            },
+            {"property": {"value": {"equals": challenge}, "name": "challenge"}},
+        ]
+    }
+
+    operation_name = "ExampleQuery"
+
+    # Make the request
+    response = requests.post(
+        url,
+        headers=headers,
+        json={"query": query, "variables": variables, "operationName": operation_name},
+    )
+    data = response.json()
+
+    return data.get("data", {}).get("aggregatedHeliconeRequest", {}).get("cost", None)

agbenchmark/reports/reports.py

@@ -1,10 +1,13 @@
 import json
+import os
 import sys
 from pathlib import Path
 from typing import Any, Callable
 
 import pytest
 
+from agbenchmark.agent_interface import MOCK_FLAG
+from agbenchmark.get_data_from_helicone import get_data_from_helicone
 from agbenchmark.reports.ReportManager import ReportManager
 from agbenchmark.start_benchmark import CONFIG_PATH, REGRESSION_TESTS_PATH, REPORTS_PATH
 from agbenchmark.utils.data_types import DIFFICULTY_MAP, DifficultyLevel, SuiteConfig
@@ -234,6 +237,11 @@ def finalize_reports(item: Any, challenge_data: dict[str, Any]) -> None:
 
     if info_details and test_name:
         if run_time:
+            cost = None
+            if not MOCK_FLAG and os.environ.get("HELICONE_API_KEY"):
+                cost = get_data_from_helicone(test_name)
+
+            info_details["metrics"]["cost"] = cost
             info_details["metrics"]["run_time"] = f"{str(round(run_time, 3))} seconds"
 
             info_details["reached_cutoff"] = float(run_time) > challenge_data["cutoff"]