mm-webhook

oproxy.py

@@ -1,5 +1,5 @@
 # バージョン情報: Python 3.12+ / FastAPI 0.115.0 / uvicorn 0.30.0 / httpx 0.28.0
-# [2026-02-10] 3060(12GB)戦術支援：全機能完全統合・非短縮版
+# [2026-02-14] 3060(12GB)戦術支援：Mattermost Webhook連携・遅延送信実装版
 import argparse
 import asyncio
 import json
@@ -35,6 +35,10 @@ CONFIG = {
     "models_cache": [],
     "loop": None,
     "vram_total": 12.0,
+    # --- Mattermost 連携設定 ---
+    "webhook_url": "https://mm.ka.sugeee.com/hooks/ctjisw6ugjg85nhddo9ox9zt4c",
+    "enable_webhook": True,
+    "max_log_len": 4000
 }
 
 # 高速抽出用正規表現
@@ -49,7 +53,40 @@ def pulse(char, color=C_RESET):
     print(f"{color}{char}{C_RESET}", end="", flush=True)
 
 
-# --- 🧠 戦略的モデル分析 (2行分割レイアウト) ---
+# --- 🛰️ Mattermost 送信エンジン (非同期・遅延送信) ---
+async def post_to_mattermost(model_name, path, req_body, resp_text):
+    """通信完了後にバックグラウンドでMattermostへログを飛ばす"""
+    if not CONFIG["webhook_url"] or not CONFIG["enable_webhook"]:
+        return
+    try:
+        req_obj = json.loads(req_body)
+        req_clean = json.dumps(req_obj, indent=2, ensure_ascii=False)
+    except:
+        req_clean = req_body
+
+    resp_display = (
+        resp_text if len(resp_text) < CONFIG["max_log_len"] 
+        else resp_text[:CONFIG["max_log_len"]] + "\n\n*(Truncated due to length)*"
+    )
+
+    payload = {
+        "username": f"oproxy:{model_name}",
+        "icon_url": "https://ollama.com/public/ollama.png",
+        "text": (
+            f"### 🛡️ LLM Traffic Log: `{path}`\n"
+            f"- **Model:** `{model_name}` | **Time:** {datetime.now().strftime('%H:%M:%S')}\n"
+            f"#### 📥 Request\n```json\n{req_clean[:800]}\n```\n"
+            f"#### 📤 Response\n{resp_display}"
+        )
+    }
+    try:
+        async with httpx.AsyncClient(timeout=10.0) as client:
+            await client.post(CONFIG["webhook_url"], json=payload)
+    except Exception as e:
+        print(f"\n{C_RED}[Webhook Fail] {e}{C_RESET}")
+
+
+# --- 🧠 戦略적モデル分析 (2行分割レイアウト) ---
 async def fetch_detailed_models():
     """Ollamaの内部情報を深掘りし、メタデータを完全取得する"""
     try:
@@ -193,6 +230,13 @@ async def sticky_proxy(path: str, request: Request):
     print(f"{get_ts()} {C_WHITE}/{path: <10}{C_RESET} ", end="", flush=True)
     body = await request.body()
     
+    # モデル名の特定
+    model_name = "unknown"
+    try:
+        model_name = json.loads(body).get("model", "unknown")
+    except:
+        pass
+
     if do_dump:
         print(
             f"\n{C_YELLOW}{'=' * 60}\n[DUMP REQUEST: {request.method} /{path}]\n{'=' * 60}{C_RESET}"
@@ -224,25 +268,29 @@ async def sticky_proxy(path: str, request: Request):
                 ) as response:
                     pulse("v", C_GREEN)
                     
+                    full_text_buffer = []
                     async for chunk in response.aiter_bytes():
-                        if do_dump:
-                            # チャンクから正規表現で高速テキスト抽出（リアルタイム表示）
                         raw_data = chunk.decode(errors="ignore")
                         matches = RE_CONTENT.findall(raw_data)
                         for m in matches:
                             try:
-                                    text = m.encode().decode(
+                                text = m.encode().decode("unicode_escape", errors="ignore")
-                                        "unicode_escape", errors="ignore"
+                                full_text_buffer.append(text)
-                                    )
+                                if do_dump:
-                                    print(
+                                    print(f"{C_WHITE}{text}{C_RESET}", end="", flush=True)
-                                        f"{C_WHITE}{text}{C_RESET}", end="", flush=True
-                                    )
                             except:
                                 pass
 
                         pulse("v", C_GREEN)
                         yield chunk
 
+                    # 通信完了後に非同期タスクとしてWebhook送信
+                    if full_text_buffer:
+                        combined_text = "".join(full_text_buffer)
+                        asyncio.create_task(post_to_mattermost(
+                            model_name, path, body.decode(errors='ignore'), combined_text
+                        ))
+
                     if do_dump:
                         print(f"\n{C_GREEN}{'=' * 60}{C_RESET}")
                     pulse("*", C_YELLOW)