feat(CG3e): add llm_token_usage BigQuery logger + /billing/tokens/summary endpoint

agents/core-logic/agent.py

@@ -17,6 +17,7 @@ Gjeldende modell-tilgjengelighet (mai 2026):
 import asyncio
 import os
 import logging
+import uuid
 from typing import Literal
 
 from google.adk.agents import Agent
@@ -48,6 +49,13 @@ HEAVY_MODE_ALLOWED_USERS = ["opax", "admin"]
 Mode = Literal["light", "heavy"]
 APP_NAME = "opax"
 
+# Token logger — feiler stille, stopper aldri agent
+try:
+    from token_logger import log_token_usage
+except ImportError:
+    def log_token_usage(*args, **kwargs):
+        pass
+
 
 def _normalize_mode(mode: str) -> str:
     mapping = {"A": "light", "A+": "heavy", "light": "light", "heavy": "heavy"}
@@ -138,6 +146,7 @@ root_agent = build_agent(mode="light")
 async def _run_async(message: str, user_id: str, session_id: str, mode: str) -> str:
     mode = _normalize_mode(mode)
     agent = build_agent(mode=mode)
+    models = get_models_for_mode(mode)
     session_service = InMemorySessionService()
     session = await session_service.create_session(
         app_name=APP_NAME, user_id=user_id, session_id=session_id,
@@ -145,11 +154,31 @@ async def _run_async(message: str, user_id: str, session_id: str, mode: str) ->
     runner = Runner(agent=agent, app_name=APP_NAME, session_service=session_service)
     new_message = types.Content(role="user", parts=[types.Part(text=message)])
     final_text = ""
+    input_tokens = 0
+    output_tokens = 0
+    request_id = str(uuid.uuid4())
+
     async for event in runner.run_async(
         user_id=user_id, session_id=session.id, new_message=new_message,
     ):
         if event.is_final_response() and event.content and event.content.parts:
             final_text = event.content.parts[0].text or ""
+        # Hent token-metadata fra event hvis tilgjengelig
+        if hasattr(event, "usage_metadata") and event.usage_metadata:
+            um = event.usage_metadata
+            input_tokens  += getattr(um, "prompt_token_count", 0) or 0
+            output_tokens += getattr(um, "candidates_token_count", 0) or 0
+
+    # Logg token-bruk — feiler stille
+    if input_tokens > 0 or output_tokens > 0:
+        log_token_usage(
+            agent_name=user_id,
+            model_name=models["orchestrator"],
+            input_tokens=input_tokens,
+            output_tokens=output_tokens,
+            request_id=request_id,
+        )
+
     return final_text
 
 

agents/core-logic/app.py

@@ -23,6 +23,8 @@ except ImportError as e:
 
 session_service = InMemorySessionService()
 APP_NAME = os.environ.get("CLOUD_RUN_SERVICE", "gcp-orchestrator")
+PROJECT_ID = os.environ.get("GOOGLE_CLOUD_PROJECT", "propane-will-491900-m5")
+BQ_BILLING_DATASET = os.environ.get("BQ_BILLING_DATASET", "billing_data")
 
 
 @asynccontextmanager
@@ -116,3 +118,50 @@ async def run(req: RunRequest):
     except Exception as e:
         logger.error(f"Agent run failed: {e}", exc_info=True)
         raise HTTPException(status_code=500, detail=str(e))
+
+
+@app.get("/billing/tokens/summary")
+async def billing_tokens_summary():
+    """
+    CG3e — Aggreger LLM token-bruk og estimert kostnad per agent siste 30 dager.
+    Returnerer JSON-liste sortert etter total_cost DESC.
+    """
+    try:
+        from google.cloud import bigquery
+        client = bigquery.Client(project=PROJECT_ID)
+
+        query = f"""
+            SELECT
+              agent_name,
+              model_name,
+              SUM(input_tokens)       AS total_input_tokens,
+              SUM(output_tokens)      AS total_output_tokens,
+              SUM(total_tokens)       AS total_tokens,
+              SUM(estimated_cost_usd) AS total_cost_usd
+            FROM
+              `{PROJECT_ID}.{BQ_BILLING_DATASET}.llm_token_usage`
+            WHERE
+              timestamp >= TIMESTAMP_SUB(CURRENT_TIMESTAMP(), INTERVAL 30 DAY)
+            GROUP BY
+              agent_name, model_name
+            ORDER BY
+              total_cost_usd DESC
+        """
+
+        results = client.query(query).result()
+        rows = [
+            {
+                "agent_name":          row.agent_name,
+                "model_name":          row.model_name,
+                "total_input_tokens":  row.total_input_tokens,
+                "total_output_tokens": row.total_output_tokens,
+                "total_tokens":        row.total_tokens,
+                "total_cost_usd":      round(float(row.total_cost_usd), 6),
+            }
+            for row in results
+        ]
+        return JSONResponse({"period_days": 30, "rows": rows})
+
+    except Exception as e:
+        logger.error(f"[/billing/tokens/summary] Failed: {e}", exc_info=True)
+        raise HTTPException(status_code=500, detail=str(e))

agents/core-logic/token_logger.py

@@ -0,0 +1,82 @@
+# token_logger.py — Logg LLM token-bruk til BigQuery (CG3e)
+
+import os
+import logging
+import uuid
+from datetime import datetime, timezone
+from typing import Optional
+
+logger = logging.getLogger(__name__)
+
+PROJECT_ID = os.environ.get("GOOGLE_CLOUD_PROJECT", "propane-will-491900-m5")
+DATASET    = os.environ.get("BQ_BILLING_DATASET", "billing_data")
+TABLE      = "llm_token_usage"
+FULL_TABLE = f"{PROJECT_ID}.{DATASET}.{TABLE}"
+
+# Prismodell (USD per 1M tokens) — oppdater ved modellbytte
+MODEL_PRICING = {
+    "gemini-2.5-pro":   {"input": 1.25, "output": 10.00},
+    "gemini-2.5-flash": {"input": 0.075, "output": 0.30},
+}
+DEFAULT_PRICING = {"input": 1.25, "output": 10.00}
+
+
+def _estimate_cost(model_name: str, input_tokens: int, output_tokens: int) -> float:
+    pricing = MODEL_PRICING.get(model_name, DEFAULT_PRICING)
+    cost = (input_tokens / 1_000_000) * pricing["input"] + \
+           (output_tokens / 1_000_000) * pricing["output"]
+    return round(cost, 8)
+
+
+def log_token_usage(
+    agent_name: str,
+    model_name: str,
+    input_tokens: int,
+    output_tokens: int,
+    request_id: Optional[str] = None,
+) -> None:
+    """
+    Logg ett LLM-kall til BigQuery-tabellen llm_token_usage.
+    Feiler stille slik at applikasjonen aldri krasjer pga logging.
+    """
+    try:
+        from google.cloud import bigquery
+        client = bigquery.Client(project=PROJECT_ID)
+
+        total_tokens = input_tokens + output_tokens
+        estimated_cost = _estimate_cost(model_name, input_tokens, output_tokens)
+
+        row = {
+            "timestamp":          datetime.now(timezone.utc).isoformat(),
+            "agent_name":         agent_name,
+            "model_name":         model_name,
+            "input_tokens":       input_tokens,
+            "output_tokens":      output_tokens,
+            "total_tokens":       total_tokens,
+            "estimated_cost_usd": estimated_cost,
+            "request_id":         request_id or str(uuid.uuid4()),
+        }
+
+        errors = client.insert_rows_json(FULL_TABLE, [row])
+        if errors:
+            logger.warning(f"[token_logger] BQ insert errors: {errors}")
+        else:
+            logger.info(
+                f"[token_logger] Logged: agent={agent_name} model={model_name} "
+                f"in={input_tokens} out={output_tokens} cost=${estimated_cost:.6f}"
+            )
+    except Exception as e:
+        logger.warning(f"[token_logger] Failed to log token usage (non-fatal): {e}")
+
+
+# BQ table schema — brukes som referanse ved manuell oppretting eller Terraform
+BQ_SCHEMA = [
+    {"name": "timestamp",          "type": "TIMESTAMP", "mode": "REQUIRED"},
+    {"name": "agent_name",         "type": "STRING",    "mode": "REQUIRED"},
+    {"name": "model_name",         "type": "STRING",    "mode": "REQUIRED"},
+    {"name": "input_tokens",       "type": "INTEGER",   "mode": "REQUIRED"},
+    {"name": "output_tokens",      "type": "INTEGER",   "mode": "REQUIRED"},
+    {"name": "total_tokens",       "type": "INTEGER",   "mode": "REQUIRED"},
+    {"name": "estimated_cost_usd", "type": "FLOAT",     "mode": "REQUIRED"},
+    {"name": "request_id",         "type": "STRING",    "mode": "NULLABLE"},
+]

scripts/create_llm_token_usage_table.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+# CG3e — Opprett BigQuery-tabell llm_token_usage
+# Kjøres én gang manuelt: bash scripts/create_llm_token_usage_table.sh
+
+set -e
+
+PROJECT="propane-will-491900-m5"
+DATASET="billing_data"
+TABLE="llm_token_usage"
+
+echo "Oppretter BigQuery-tabell ${PROJECT}:${DATASET}.${TABLE} ..."
+
+bq mk --table \
+  --description "Logg over all LLM token-bruk for CostGuard (CG3e)" \
+  "${PROJECT}:${DATASET}.${TABLE}" \
+  timestamp:TIMESTAMP,agent_name:STRING,model_name:STRING,input_tokens:INTEGER,output_tokens:INTEGER,total_tokens:INTEGER,estimated_cost_usd:FLOAT,request_id:STRING
+
+echo "✅ Tabell ${FULL_TABLE} opprettet."