""" CascadeFlow Configuration – Agent Governance Monitors per-session call count, estimated token cost, and latency budgets. Raises RuntimeError if constraints are exceeded so the agent loop can surface a graceful error rather than silently over-spending. """ from __future__ import annotations import time from dataclasses import dataclass, field try: from cascadeflow import CascadeMonitor _monitor = CascadeMonitor( max_calls_per_session=50, max_latency_ms=8000, cost_per_call_usd=0.0, # Gemini 1.5 Flash free-tier ) def check_constraints() -> None: """Raise RuntimeError if any governance constraint is violated.""" _monitor.check() def record_call() -> None: """Notify cascadeflow that one LLM call has completed.""" _monitor.record() except ImportError: # Graceful fallback: lightweight manual tracker @dataclass class _FallbackTracker: max_calls: int = 50 calls: int = 0 _start: float = field(default_factory=time.monotonic) _tracker = _FallbackTracker() def check_constraints() -> None: # type: ignore[misc] if _tracker.calls >= _tracker.max_calls: raise RuntimeError( f"Session limit reached: {_tracker.max_calls} calls. " "Please start a new session." ) def record_call() -> None: # type: ignore[misc] _tracker.calls += 1