feat: JASON_BACKEND toggle — lokal Ollama fallback for /run

main.py

@@ -70,6 +70,22 @@ EMMA_BACKEND   = os.environ.get("EMMA_BACKEND", "vertex")
 EMMA_LOCAL_URL = os.environ.get("EMMA_LOCAL_URL", "http://localhost:8000/generate")
 EMMA_MODEL     = os.environ.get("EMMA_MODEL", "gemini-2.5-pro")
 
+# ── JASON: backend-flagg ────────────────────────────────────────────────────
+JASON_BACKEND   = os.environ.get("JASON_BACKEND", "vertex")
+JASON_LOCAL_URL = os.environ.get("JASON_LOCAL_URL", "http://34.158.137.245:11434/api/generate")
+JASON_MODEL     = os.environ.get("JASON_MODEL", "gemma3:4b")
+
+async def _jason_local(message: str) -> str:
+    import httpx
+    async with httpx.AsyncClient(timeout=120.0) as client:
+        r = await client.post(JASON_LOCAL_URL, json={
+            "model": JASON_MODEL,
+            "prompt": message,
+            "stream": False
+        })
+        r.raise_for_status()
+        return r.json().get("response", "")
+
 _WORLD_MD_PATH = pathlib.Path(__file__).parent / "docs" / "gemma" / "world.md"
 try:
     _EMMA_SYSTEM_PROMPT = _WORLD_MD_PATH.read_text(encoding="utf-8")
@@ -987,8 +1003,11 @@ async def run_agent(req: RunRequest):                          # ← async def
     error_msg = None
     response  = None
     try:
-        from agent import _run_async                           # ← importer async-funksjonen
-        response = await _run_async(                           # ← await direkte
+        if JASON_BACKEND == "local":
+            response = await _jason_local(req.message)
+        else:
+            from agent import _run_async                           # ← importer async-funksjonen
+            response = await _run_async(                           # ← await direkte
             message=req.message,
             user_id=req.user_id,
             session_id=req.session_id,