Populate mime_type for assets in scanner and API paths

- Add custom MIME type registrations for model files (.safetensors, .pt, .ckpt, .gguf, .yaml)
- Pass mime_type through SeedAssetSpec to bulk_ingest
- Re-register types before use since server.py mimetypes.init() resets them
- Add tests for bulk ingest mime_type handling

Amp-Thread-ID: https://ampcode.com/threads/T-019c3626-c6ad-7139-a570-62da4e656a1a
Co-authored-by: Amp <amp@ampcode.com>

app/assets/scanner.py

@@ -320,6 +320,9 @@ def build_asset_specs(
             except Exception as e:
                 logging.warning("Failed to hash %s: %s", abs_p, e)
 
+        mime_type = metadata.content_type if metadata else None
+        if mime_type is None:
+            print(f"[build_asset_specs] no mime_type for {abs_p} (metadata={metadata is not None})")
         specs.append(
             {
                 "abs_path": abs_p,
@@ -330,6 +333,7 @@ def build_asset_specs(
                 "fname": rel_fname,
                 "metadata": metadata,
                 "hash": asset_hash,
+                "mime_type": mime_type,
             }
         )
         tag_pool.update(tags)

app/assets/services/bulk_ingest.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import logging
 import os
 import uuid
 from dataclasses import dataclass
@@ -38,6 +39,7 @@ class SeedAssetSpec(TypedDict):
     fname: str
     metadata: ExtractedMetadata | None
     hash: str | None
+    mime_type: str | None
 
 
 class AssetRow(TypedDict):
@@ -162,12 +164,15 @@ def batch_insert_seed_assets(
         absolute_path_list.append(absolute_path)
         path_to_asset_id[absolute_path] = asset_id
 
+        mime_type = spec.get("mime_type")
+        if mime_type is None:
+            logging.info("batch_insert_seed_assets: no mime_type for %s", absolute_path)
         asset_rows.append(
             {
                 "id": asset_id,
                 "hash": spec.get("hash"),
                 "size_bytes": spec["size_bytes"],
-                "mime_type": None,
+                "mime_type": mime_type,
                 "created_at": current_time,
             }
         )

app/assets/services/metadata_extract.py

@@ -20,6 +20,31 @@ SAFETENSORS_EXTENSIONS = frozenset({".safetensors", ".sft"})
 # Maximum safetensors header size to read (8MB)
 MAX_SAFETENSORS_HEADER_SIZE = 8 * 1024 * 1024
 
+def _register_custom_mime_types():
+    """Register custom MIME types for model and config files.
+    
+    Called before each use because mimetypes.init() in server.py resets the database.
+    Uses a quick check to avoid redundant registrations.
+    """
+    # Quick check if already registered (avoids redundant add_type calls)
+    test_result, _ = mimetypes.guess_type("test.safetensors")
+    if test_result == "application/safetensors":
+        return
+    
+    mimetypes.add_type("application/safetensors", ".safetensors")
+    mimetypes.add_type("application/safetensors", ".sft")
+    mimetypes.add_type("application/pytorch", ".pt")
+    mimetypes.add_type("application/pytorch", ".pth")
+    mimetypes.add_type("application/pickle", ".ckpt")
+    mimetypes.add_type("application/pickle", ".pkl")
+    mimetypes.add_type("application/gguf", ".gguf")
+    mimetypes.add_type("application/yaml", ".yaml")
+    mimetypes.add_type("application/yaml", ".yml")
+
+
+# Register custom types at module load
+_register_custom_mime_types()
+
 
 @dataclass
 class ExtractedMetadata:
@@ -284,9 +309,12 @@ def extract_file_metadata(
     _, ext = os.path.splitext(abs_path)
     meta.format = ext.lstrip(".").lower() if ext else ""
 
-    # MIME type guess
+    # MIME type guess (re-register in case mimetypes.init() was called elsewhere)
+    _register_custom_mime_types()
     mime_type, _ = mimetypes.guess_type(abs_path)
     meta.content_type = mime_type
+    if mime_type is None:
+        print(f"[extract_file_metadata] No mime_type for {abs_path}")
 
     # Size from stat
     if stat_result is None: