Update app.py

app.py

@@ -14,6 +14,8 @@ import traceback
 import inspect
 import re
 import logging
+import base64
+import tempfile
 
 # Setup simple logging
 logging.basicConfig(level=logging.INFO)
@@ -44,6 +46,12 @@ except Exception as e:
         "Make sure config.py exists with CLIENT_ID, CLIENT_SECRET, REFRESH_TOKEN, API_BASE, LOCAL_MODEL (or set LOCAL_MODEL=None)."
     )
 
+# Safe defaults for optional config values (prevents NameError)
+LOCAL_TOKENIZER = globals().get("LOCAL_TOKENIZER", None)
+OPENROUTER_BASE_URL = globals().get("OPENROUTER_BASE_URL", None)
+OPENROUTER_API_KEY = globals().get("OPENROUTER_API_KEY", None)
+OPENROUTER_MODEL = globals().get("OPENROUTER_MODEL", None)
+
 # ----------------------------
 # FastMCP init
 # ----------------------------
@@ -89,6 +97,33 @@ def _log_llm_call(confidence: Optional[float] = None):
 
 _init_analytics()
 
+# ----------------------------
+# Helper: normalize local file_path args
+# ----------------------------
+def _normalize_local_path_args(args: Any) -> Any:
+    """
+    If args is a dict containing 'file_path' that points to a local file (starting with /mnt/data/),
+    add file_url: file://<path> and file_b64: <base64> (optional) so tools can use either.
+    Returns modified args (same object or new).
+    """
+    if not isinstance(args, dict):
+        return args
+    fp = args.get("file_path")
+    if isinstance(fp, str) and fp.startswith("/mnt/data/") and os.path.exists(fp):
+        try:
+            args["file_url"] = f"file://{fp}"
+            # include base64 payload for tools that prefer raw bytes
+            try:
+                with open(fp, "rb") as f:
+                    args["file_b64"] = base64.b64encode(f.read()).decode("utf-8")
+            except Exception as e:
+                # If reading fails, still include file_url
+                logger.warning("Could not read file for base64 embedding: %s", e)
+                args.pop("file_b64", None)
+        except Exception as e:
+            logger.warning("Normalization error for file_path %s: %s", fp, e)
+    return args
+
 # ----------------------------
 # DeepSeek / HF model loader
 # ----------------------------
@@ -99,10 +134,6 @@ LOADED_MODEL_NAME = None
 def init_deepseek_model():
     """
     Try to load LOCAL_MODEL via transformers.pipeline.
-    Expected LOCAL_MODEL examples:
-      - "deepseek-ai/deepseek-r1-7b"  (may require GPU; big)
-      - "deepseek-ai/deepseek-r1-3b"  (smaller)
-      - "deepseek-ai/deepseek-r1-1.3b" (more likely to load on moderate machines)
     If loading fails, try a fallback small model (distilgpt2 or flan-t5-small if seq2seq).
     """
     global LLM_PIPELINE, TOKENIZER, LOADED_MODEL_NAME
@@ -192,7 +223,8 @@ def deepseek_generate(prompt: str, max_tokens: int = 256) -> Dict[str, Any]:
         if isinstance(out, list) and len(out) > 0:
             first = out[0]
             if isinstance(first, dict):
-                text = first.get("generated_text") or first.get("generated_text", "") or first.get("text") or str(first)
+                # many HF pipelines use 'generated_text' or 'text'
+                text = first.get("generated_text") or first.get("text") or str(first)
             else:
                 text = str(first)
         else:
@@ -369,13 +401,15 @@ def deepseek_response(message: str, history: Optional[List[Tuple[str,str]]] = No
     prompt = f"{system_prompt}\n{history_text}\nUser: {message}\nAssistant:"
     gen = deepseek_generate(prompt, max_tokens=256)
     text = gen.get("text", "")
-    # if text looks like JSON with a tool action, try to invoke
+    # if text looks like JSON with a tool action, try to invoke (normalize args first)
     payload = text.strip()
     if payload.startswith("{") or payload.startswith("["):
         try:
             parsed = json.loads(payload)
             if isinstance(parsed, dict) and "tool" in parsed:
                 tool_name = parsed.get("tool"); args = parsed.get("args", {})
+                # Normalize local path args if present
+                args = _normalize_local_path_args(args) if isinstance(args, dict) else args
                 if tool_name in globals() and callable(globals()[tool_name]):
                     try:
                         out = globals()[tool_name](**args) if isinstance(args, dict) else globals()[tool_name](args)