1  """
 2  ARGUS-AI Agentic Workflow Evaluation
 3  
 4  Demonstrates ASF, ERR, and CPCS metrics for autonomous agent monitoring.
 5  
 6  Author: Anil Prasad | Ambharii Labs
 7  """
 8  
 9  import argus_ai
10  from argus_ai.types import AgenticEvalRequest
11  
12  # Initialize with agentic profile
13  argus = argus_ai.init(profile="agentic")
14  
15  # Simulate an agentic workflow execution
16  workflow = AgenticEvalRequest(
17      prompt="Research competitor pricing and generate a comparison report",
18      response="Generated comprehensive pricing comparison across 5 competitors.",
19      steps_planned=8,
20      steps_completed=7,
21      steps_failed=2,
22      steps_recovered=1,
23      retries=3,
24      total_cost_usd=0.45,
25      tool_calls=[
26          {"tool": "web_search", "status": "success", "latency_ms": 800},
27          {"tool": "web_search", "status": "success", "latency_ms": 650},
28          {"tool": "web_scrape", "status": "failed", "latency_ms": 5000},
29          {"tool": "web_scrape", "status": "success", "latency_ms": 1200},
30          {"tool": "data_extract", "status": "success", "latency_ms": 300},
31          {"tool": "llm_analyze", "status": "success", "latency_ms": 2100},
32          {"tool": "llm_analyze", "status": "failed", "latency_ms": 30000},
33          {"tool": "report_gen", "status": "success", "latency_ms": 1800},
34      ],
35      model_name="claude-sonnet-4",
36      latency_ms=42000.0,
37      metadata={"workflow_type": "competitive_analysis"},
38  )
39  
40  # Run full evaluation (G-ARVIS + agentic metrics)
41  result, agentic_metrics = argus.evaluate_agentic(workflow)
42  
43  print("=== G-ARVIS Scores ===")
44  print(f"  Composite:      {result.garvis_composite:.3f}")
45  print(f"  Groundedness:   {result.groundedness:.3f}")
46  print(f"  Accuracy:       {result.accuracy:.3f}")
47  print(f"  Reliability:    {result.reliability:.3f}")
48  print(f"  Variance:       {result.variance:.3f}")
49  print(f"  Inference Cost: {result.inference_cost:.3f}")
50  print(f"  Safety:         {result.safety:.3f}")
51  
52  print("\n=== Agentic Metrics ===")
53  for metric in agentic_metrics:
54      print(f"  {metric.name}: {metric.score:.3f}")
55      for k, v in metric.details.items():
56          print(f"    {k}: {v}")
57  
58  print(f"\nOverall Passing: {result.passing}")
59  if result.alerts:
60      print(f"\nAlerts ({len(result.alerts)}):")
61      for alert in result.alerts:
62          print(f"  {alert}")