Scope vision model resolution by owner (#3009)

src/chat_handler.py

@@ -229,7 +229,7 @@ class ChatHandler:
                             except Exception:
                                 vl_desc = None
                         if not vl_desc:
-                            vl_result = analyze_image_with_vl_result(file_info["path"])
+                            vl_result = analyze_image_with_vl_result(file_info["path"], owner=owner)
                             vl_desc = vl_result.get("text", "")
                             vl_model = vl_result.get("model", "")
                             if vl_desc and not vl_desc.startswith("["):

src/document_processor.py

@@ -109,7 +109,7 @@ def _process_text_file(path: str) -> str:
         return result
 
 
-def _process_pdf(path: str) -> str:
+def _process_pdf(path: str, owner: str | None = None) -> str:
     """Process PDF file with text extraction (pypdf). Uses VL model for image-heavy pages."""
     try:
         from pypdf import PdfReader
@@ -133,7 +133,7 @@ def _process_pdf(path: str) -> str:
                             temp_img_path = tmp.name
                         try:
                             img.image.save(temp_img_path, "PNG")  # pypdf -> PIL image
-                            ocr_text = analyze_image_with_vl(temp_img_path)
+                            ocr_text = analyze_image_with_vl(temp_img_path, owner=owner)
                             if ocr_text and "unavailable" not in ocr_text.lower():
                                 pdf_text += f"\n\n[Page {page_num + 1} image {img_index + 1} text]: {ocr_text}"
                         finally:
@@ -254,7 +254,7 @@ def _load_vl_settings() -> dict:
         return {}
 
 
-def _resolve_vl_model(configured: str) -> tuple:
+def _resolve_vl_model(configured: str, owner: str | None = None) -> tuple:
     """Resolve the vision model to (url, model_id, headers).
 
     Uses admin-configured model if set, otherwise tries auto-detection
@@ -263,7 +263,7 @@ def _resolve_vl_model(configured: str) -> tuple:
     from src.ai_interaction import _resolve_model
 
     if configured:
-        return _resolve_model(configured)
+        return _resolve_model(configured, owner=owner)
 
     # Auto-detect: try known vision-capable models in priority order
     candidates = [
@@ -274,14 +274,14 @@ def _resolve_vl_model(configured: str) -> tuple:
     ]
     for candidate in candidates:
         try:
-            return _resolve_model(candidate)
+            return _resolve_model(candidate, owner=owner)
         except (ValueError, Exception):
             continue
 
     raise ValueError("No vision model available")
 
 
-def analyze_image_with_vl_result(image_path: str) -> dict:
+def analyze_image_with_vl_result(image_path: str, owner: str | None = None) -> dict:
     """Analyze an image and return both text and the model that produced it."""
     logger.info(f"Analyzing image with VL model: {image_path}")
     try:
@@ -291,7 +291,7 @@ def analyze_image_with_vl_result(image_path: str) -> dict:
         vl_model = settings.get("vision_model", "")
 
         try:
-            url, model_id, headers = _resolve_vl_model(vl_model)
+            url, model_id, headers = _resolve_vl_model(vl_model, owner=owner)
         except ValueError:
             return {"text": "[No vision model configured — set one in Settings → Vision]", "model": vl_model or ""}
 
@@ -316,7 +316,7 @@ def analyze_image_with_vl_result(image_path: str) -> dict:
         # — same shape as task/chat but its own list (`vision_model_fallbacks`).
         try:
             from src.endpoint_resolver import resolve_vision_fallback_candidates
-            _vl_candidates = [(url, model_id, headers)] + resolve_vision_fallback_candidates()
+            _vl_candidates = [(url, model_id, headers)] + resolve_vision_fallback_candidates(owner=owner)
         except Exception:
             _vl_candidates = [(url, model_id, headers)]
 
@@ -338,9 +338,9 @@ def analyze_image_with_vl_result(image_path: str) -> dict:
         return {"text": "[VL model unavailable - image not analyzed]", "model": ""}
 
 
-def analyze_image_with_vl(image_path: str) -> str:
+def analyze_image_with_vl(image_path: str, owner: str | None = None) -> str:
     """Analyze an image using the admin-configured Vision-Language model."""
-    return analyze_image_with_vl_result(image_path).get("text", "")
+    return analyze_image_with_vl_result(image_path, owner=owner).get("text", "")
 
 
 def build_user_content(
@@ -434,7 +434,7 @@ def build_user_content(
                         # Pull the PDF prose once — used as either intro_text
                         # (form path) or the doc body (plain path).
                         try:
-                            pdf_body_text = strip_pdf_content_marker(_process_pdf(path))
+                            pdf_body_text = strip_pdf_content_marker(_process_pdf(path, owner=owner))
                         except Exception:
                             pdf_body_text = None
 
@@ -517,7 +517,7 @@ def build_user_content(
                     except Exception as e:
                         logger.warning(f"PDF auto-doc creation failed for {path}: {e}")
                 if extracted_text is None:
-                    extracted_text = _process_pdf(path)
+                    extracted_text = _process_pdf(path, owner=owner)
             elif mime.startswith("text/") or _is_text_file(path):
                 extracted_text = _process_text_file(path)
             else: