Block A: License-Check + Credits-Tracking fuer Enhance und Chat
- Neuer Helper charge_usage_to_tenant() in services/license_service.py: UPSERT in token_usage_monthly und Credits-Abzug aus licenses.credits_used. Wiederverwendbar fuer alle Claude-Call-Verursacher. - Orchestrator: Inline-Buchungslogik (35 Zeilen) durch Helper-Aufruf ersetzt. - routers/incidents.py POST /enhance-description: require_writable_license statt get_current_user, db_dependency hinzugefuegt, Credits-Buchung mit source="enhance" nach jedem Claude-Call. - routers/chat.py POST /: analog require_writable_license + Credits-Buchung mit source="chat". _call_claude_chat() gibt jetzt zusaetzlich ClaudeUsage zurueck. Abgelaufene/gesperrte Lizenzen koennen damit keine Haiku-Calls mehr ausloesen, und alle Kosten werden konsistent auf Tenant-Ebene verbucht.
Dieser Commit ist enthalten in:
@@ -12,6 +12,11 @@ from pydantic import BaseModel, Field
|
||||
|
||||
from auth import get_current_user
|
||||
from config import CLAUDE_PATH, CLAUDE_MODEL_FAST
|
||||
from database import db_dependency
|
||||
from middleware.license_check import require_writable_license
|
||||
from services.license_service import charge_usage_to_tenant
|
||||
from agents.claude_client import ClaudeUsage
|
||||
import aiosqlite
|
||||
|
||||
logger = logging.getLogger("osint.chat")
|
||||
|
||||
@@ -21,8 +26,8 @@ router = APIRouter(tags=["chat"])
|
||||
# Claude CLI Aufruf (Chat-spezifisch, kein JSON-Modus)
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
async def _call_claude_chat(prompt: str) -> tuple[str, int]:
|
||||
"""Ruft Claude CLI fuer Chat auf. Gibt (text, duration_ms) zurueck.
|
||||
async def _call_claude_chat(prompt: str) -> tuple[str, int, ClaudeUsage]:
|
||||
"""Ruft Claude CLI fuer Chat auf. Gibt (text, duration_ms, usage) zurueck.
|
||||
|
||||
Anders als call_claude(): kein JSON-Output-Modus, kein append-system-prompt.
|
||||
"""
|
||||
@@ -62,21 +67,29 @@ async def _call_claude_chat(prompt: str) -> tuple[str, int]:
|
||||
raw = stdout.decode("utf-8", errors="replace").strip()
|
||||
duration_ms = 0
|
||||
result_text = raw
|
||||
usage = ClaudeUsage()
|
||||
|
||||
try:
|
||||
data = _json.loads(raw)
|
||||
result_text = data.get("result", raw)
|
||||
duration_ms = data.get("duration_ms", 0)
|
||||
cost = data.get("total_cost_usd", 0.0)
|
||||
u = data.get("usage", {})
|
||||
usage = ClaudeUsage(
|
||||
input_tokens=u.get("input_tokens", 0),
|
||||
output_tokens=u.get("output_tokens", 0),
|
||||
cache_creation_tokens=u.get("cache_creation_input_tokens", 0),
|
||||
cache_read_tokens=u.get("cache_read_input_tokens", 0),
|
||||
cost_usd=data.get("total_cost_usd", 0.0),
|
||||
duration_ms=duration_ms,
|
||||
)
|
||||
logger.info(
|
||||
f"Chat Claude: {u.get('input_tokens', 0)} in / {u.get('output_tokens', 0)} out / "
|
||||
f"${cost:.4f} / {duration_ms}ms"
|
||||
f"Chat Claude: {usage.input_tokens} in / {usage.output_tokens} out / "
|
||||
f"${usage.cost_usd:.4f} / {duration_ms}ms"
|
||||
)
|
||||
except _json.JSONDecodeError:
|
||||
logger.warning("Chat Claude CLI Antwort kein JSON, nutze raw output")
|
||||
|
||||
return result_text, duration_ms
|
||||
return result_text, duration_ms, usage
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# Models
|
||||
@@ -395,7 +408,8 @@ def _build_prompt(user_message: str, history: list[dict]) -> str:
|
||||
@router.post("", response_model=ChatResponse)
|
||||
async def chat(
|
||||
req: ChatRequest,
|
||||
current_user: dict = Depends(get_current_user),
|
||||
current_user: dict = Depends(require_writable_license),
|
||||
db: aiosqlite.Connection = Depends(db_dependency),
|
||||
):
|
||||
"""Chat-Nachricht verarbeiten und Antwort generieren."""
|
||||
user_id = current_user["id"]
|
||||
@@ -420,7 +434,7 @@ async def chat(
|
||||
|
||||
# Claude CLI aufrufen
|
||||
try:
|
||||
result, duration_ms = await _call_claude_chat(prompt)
|
||||
result, duration_ms, usage = await _call_claude_chat(prompt)
|
||||
except TimeoutError:
|
||||
raise HTTPException(status_code=504, detail="Der Assistent antwortet gerade nicht. Bitte versuche es erneut.")
|
||||
except RuntimeError as e:
|
||||
@@ -430,6 +444,10 @@ async def chat(
|
||||
logger.error(f"Chat Claude-Fehler: {e}")
|
||||
raise HTTPException(status_code=502, detail="Der Assistent ist voruebergehend nicht erreichbar.")
|
||||
|
||||
# Credits buchen
|
||||
await charge_usage_to_tenant(db, current_user.get("tenant_id"), usage, source="chat")
|
||||
await db.commit()
|
||||
|
||||
# Output sanitieren
|
||||
reply = _sanitize_output(result)
|
||||
if not reply:
|
||||
|
||||
In neuem Issue referenzieren
Einen Benutzer sperren