Fix Embedder Schemas & Migrate Default Embedding Model

- Fixed missing output schema in Dev UI for embedder actions by explicitly setting input_schema and output_schema on Action objects.
- Added default sampleInput to embedder actions for easier testing.
- Resolved a deadlock in registry.py affecting dotprompt schema resolution (switched to async).
- Migrated default embedding model references from text-embedding-004 to gemini-embedding-001 across plugins, samples, and tests due to model deprecation/availability.
- Fixed a redundant cast lint warning in _aio.py.

Author: MengqinShen-GL

py/engdoc/parity-analysis/plugin_api_consistency.md

@@ -237,7 +237,7 @@ model='ollama/llama3'
 
 **JS Pattern:**
 ```typescript
-const embedder = googleAI.embedder('text-embedding-004');
+const embedder = googleAI.embedder('gemini-embedding-001');
 ```
 
 **Python:** ❌ No equivalent

py/packages/genkit/src/genkit/ai/_aio.py

@@ -619,7 +619,7 @@ async def embed(
         {version, ...config, ...options} (options take precedence).
 
         Args:
-            embedder: Embedder name (e.g., 'googleai/text-embedding-004') or
+            embedder: Embedder name (e.g., 'googleai/gemini-embedding-001') or
                 an EmbedderRef with configuration.
             content: A single string, Document, or DocumentData to embed.
             metadata: Optional metadata to apply to the document. Only used
@@ -634,25 +634,25 @@ async def embed(
             ValueError: If content is not specified.
 
         Example - Basic string embedding:
-            >>> embeddings = await ai.embed(embedder='googleai/text-embedding-004', content='Hello, world!')
+            >>> embeddings = await ai.embed(embedder='googleai/gemini-embedding-001', content='Hello, world!')
             >>> print(len(embeddings[0].embedding))  # Vector dimensions
 
         Example - With metadata:
             >>> embeddings = await ai.embed(
-            ...     embedder='googleai/text-embedding-004',
+            ...     embedder='googleai/gemini-embedding-001',
             ...     content='Product description',
             ...     metadata={'category': 'electronics'},
             ... )
 
         Example - With embedder options:
             >>> embeddings = await ai.embed(
-            ...     embedder='googleai/text-embedding-004',
+            ...     embedder='googleai/gemini-embedding-001',
             ...     content='Search query',
             ...     options={'task_type': 'RETRIEVAL_QUERY'},
             ... )
 
         Example - Using EmbedderRef:
-            >>> ref = create_embedder_ref('googleai/text-embedding-004', config={'task_type': 'CLUSTERING'})
+            >>> ref = create_embedder_ref('googleai/gemini-embedding-001', config={'task_type': 'CLUSTERING'})
             >>> embeddings = await ai.embed(embedder=ref, content='Text')
         """
         embedder_name = self._resolve_embedder_name(embedder)
@@ -706,7 +706,7 @@ async def embed_many(
         and passes options directly. This matches the JS canonical behavior.
 
         Args:
-            embedder: Embedder name (e.g., 'googleai/text-embedding-004') or
+            embedder: Embedder name (e.g., 'googleai/gemini-embedding-001') or
                 an EmbedderRef.
             content: List of strings, Documents, or DocumentData to embed.
             metadata: Optional metadata to apply to all items. Only used when
@@ -722,22 +722,22 @@ async def embed_many(
 
         Example - Basic batch embedding:
             >>> embeddings = await ai.embed_many(
-            ...     embedder='googleai/text-embedding-004',
+            ...     embedder='googleai/gemini-embedding-001',
             ...     content=['Doc 1', 'Doc 2', 'Doc 3'],
             ... )
             >>> for i, emb in enumerate(embeddings):
             ...     print(f'Doc {i}: {len(emb.embedding)} dims')
 
         Example - With shared metadata:
             >>> embeddings = await ai.embed_many(
-            ...     embedder='googleai/text-embedding-004',
+            ...     embedder='googleai/gemini-embedding-001',
             ...     content=['text1', 'text2'],
             ...     metadata={'batch_id': 'batch-001'},
             ... )
 
         Example - With options (EmbedderRef config is NOT extracted):
             >>> embeddings = await ai.embed_many(
-            ...     embedder='googleai/text-embedding-004',
+            ...     embedder='googleai/gemini-embedding-001',
             ...     content=documents,
             ...     options={'task_type': 'RETRIEVAL_DOCUMENT'},
             ... )

py/packages/genkit/src/genkit/blocks/embedding.py

@@ -31,7 +31,7 @@
     |                   | vectors. Contains 'embedding' field and metadata.    |
     +-------------------+------------------------------------------------------+
     | Embedder          | A model/service that converts text to embeddings.    |
-    |                   | Examples: 'googleai/text-embedding-004'. Registered  |
+    |                   | Examples: 'googleai/gemini-embedding-001'. Registered  |
     |                   | as actions, invoked via embed() and embed_many().    |
     +-------------------+------------------------------------------------------+
     | EmbedderRef       | Reference bundling embedder name with optional       |
@@ -68,14 +68,14 @@
     - ai.embed_many(): Embed multiple pieces of content in batch
 
 Example - Single embedding:
-    >>> embeddings = await ai.embed(embedder='googleai/text-embedding-004', content='Hello, world!')
+    >>> embeddings = await ai.embed(embedder='googleai/gemini-embedding-001', content='Hello, world!')
     >>> vector = embeddings[0].embedding
 
 Example - Batch embedding:
-    >>> embeddings = await ai.embed_many(embedder='googleai/text-embedding-004', content=['Doc 1', 'Doc 2', 'Doc 3'])
+    >>> embeddings = await ai.embed_many(embedder='googleai/gemini-embedding-001', content=['Doc 1', 'Doc 2', 'Doc 3'])
 
 Example - Using EmbedderRef with configuration:
-    >>> ref = create_embedder_ref('googleai/text-embedding-004', config={'task_type': 'CLUSTERING'}, version='v1')
+    >>> ref = create_embedder_ref('googleai/gemini-embedding-001', config={'task_type': 'CLUSTERING'}, version='v1')
     >>> embeddings = await ai.embed(embedder=ref, content='My text')
 
 Note on embed() vs embed_many():
@@ -195,12 +195,16 @@ def embedder_action_metadata(
 
     embedder_info['customOptions'] = options.config_schema if options.config_schema else None
 
+    # Default sample input for easier testing in Dev UI
+    sample_input: dict[str, object] = {'input': [{'content': [{'text': 'Hello, world!'}]}]}
+
     return ActionMetadata(
         kind=cast(ActionKind, ActionKind.EMBEDDER),
         name=name,
         input_json_schema=to_json_schema(EmbedRequest),
         output_json_schema=to_json_schema(EmbedResponse),
         metadata=embedder_metadata_dict,
+        sample_input=sample_input,
     )
 
 

py/packages/genkit/src/genkit/blocks/prompt.py

@@ -851,9 +851,11 @@ class RecipeInput(BaseModel):
         middleware = effective_opts.get('use') or self._use
         context = effective_opts.get('context')
 
+        rendered_options = await self.render(input=input, opts=effective_opts)
+
         result = await generate_action(
             self._registry,
-            await self.render(input=input, opts=effective_opts),
+            rendered_options,
             on_chunk=on_chunk,
             middleware=middleware,
             context=context if context else ActionRunContext._current_context(),  # pyright: ignore[reportPrivateUsage]

py/packages/genkit/src/genkit/core/action/_action.py

@@ -92,7 +92,8 @@
 from functools import cached_property
 from typing import Any, Generic, Protocol, cast, get_type_hints
 
-from pydantic import BaseModel, TypeAdapter, ValidationError
+from pydantic import BaseModel, ConfigDict, TypeAdapter, ValidationError
+from pydantic.alias_generators import to_camel
 from typing_extensions import Never, TypeVar
 
 from genkit.aio import Channel, ensure_async
@@ -524,6 +525,8 @@ def _initialize_io_schemas(
 class ActionMetadata(BaseModel):
     """Metadata for actions."""
 
+    model_config = ConfigDict(alias_generator=to_camel, populate_by_name=True)
+
     kind: ActionKind
     name: str
     description: str | None = None
@@ -533,6 +536,7 @@ class ActionMetadata(BaseModel):
     output_json_schema: dict[str, object] | None = None
     stream_schema: object | None = None
     metadata: dict[str, object] | None = None
+    sample_input: dict[str, object] | None = None
 
 
 _SyncTracingWrapper = Callable[[object | None, ActionRunContext], ActionResponse[Any]]

py/packages/genkit/src/genkit/core/reflection.py

@@ -122,6 +122,7 @@ def _build_actions_payload(
             'inputSchema': getattr(meta, 'input_json_schema', None),
             'outputSchema': getattr(meta, 'output_json_schema', None),
             'metadata': getattr(meta, 'metadata', None),
+            'sampleInput': getattr(meta, 'sample_input', None),
         }
 
         if key not in actions:

py/plugins/dev-local-vectorstore/src/genkit/plugins/dev_local_vectorstore/plugin_api.py

@@ -46,7 +46,7 @@ def define_dev_local_vector_store(
     Args:
         ai: The Genkit instance to register the retriever and indexer with.
         name: Name of the retriever and indexer.
-        embedder: The embedder to use (e.g., 'vertexai/text-embedding-004').
+        embedder: The embedder to use (e.g., 'vertexai/gemini-embedding-001').
         embedder_options: Optional configuration to pass to the embedder.
 
     Returns:

py/plugins/firebase/src/genkit/plugins/firebase/firestore.py

@@ -63,7 +63,7 @@ def define_firestore_vector_store(
     Args:
         ai: The Genkit instance to register the retriever with.
         name: Name of the retriever.
-        embedder: The embedder to use (e.g., 'vertexai/text-embedding-004').
+        embedder: The embedder to use (e.g., 'vertexai/gemini-embedding-001').
         embedder_options: Optional configuration to pass to the embedder.
         collection: The name of the Firestore collection to query.
         vector_field: The name of the field containing the vector embeddings.

py/plugins/google-genai/src/genkit/plugins/google_genai/google.py

@@ -283,7 +283,7 @@ class GoogleAI(Plugin):
         +------------------+-------------------+--------------------------------+
         | Gemini/Gemma     | MODEL             | googleai/gemini-2.0-flash-001  |
         | Imagen           | MODEL             | googleai/imagen-3.0-generate   |
-        | Embedders        | EMBEDDER          | googleai/text-embedding-004    |
+        | Embedders        | EMBEDDER          | googleai/gemini-embedding-001  |
         | Veo (video)      | BACKGROUND_MODEL  | googleai/veo-2.0-generate-001  |
         +------------------+-------------------+--------------------------------+
 

py/plugins/google-genai/test/google_plugin_test.py

@@ -137,7 +137,7 @@ async def test_googleai_initialize(mock_client_cls: MagicMock) -> None:
     m1.description = ' Gemini Pro '
 
     m2 = MagicMock()
-    m2.name = 'models/text-embedding-004'
+    m2.name = 'models/gemini-embedding-001'
     m2.supported_actions = ['embedContent']
     m2.description = ' Embedding '
 
@@ -264,7 +264,7 @@ class MockModel:
 
     models_return_value = [
         MockModel(supported_actions=['generateContent'], name='models/gemini-pro'),
-        MockModel(supported_actions=['embedContent'], name='models/text-embedding-004'),
+        MockModel(supported_actions=['embedContent'], name='models/gemini-embedding-001'),
         MockModel(supported_actions=['generateContent'], name='models/gemini-2.0-flash-tts'),  # TTS
     ]
 
@@ -279,7 +279,7 @@ class MockModel:
     assert action1 is not None
 
     # Check Embedder
-    action2 = next(a for a in result if a.name == googleai_name('text-embedding-004'))
+    action2 = next(a for a in result if a.name == googleai_name('gemini-embedding-001'))
     assert action2 is not None
     assert action2.kind == ActionKind.EMBEDDER
 
@@ -490,7 +490,7 @@ async def test_vertexai_initialize(vertexai_plugin_instance: VertexAI) -> None:
     m1.supported_actions = ['generateContent']
 
     m2 = MagicMock()
-    m2.name = 'publishers/google/models/text-embedding-004'
+    m2.name = 'publishers/google/models/gemini-embedding-001'
     m2.supported_actions = ['embedContent']
 
     plugin._client.models.list.return_value = [m1, m2]  # type: ignore
@@ -661,7 +661,7 @@ class MockModel:
 
     [
         MockModel(name='publishers/google/models/gemini-1.5-flash'),
-        MockModel(name='publishers/google/models/text-embedding-004'),
+        MockModel(name='publishers/google/models/gemini-embedding-001'),
         MockModel(name='publishers/google/models/imagen-3.0-generate-001'),
         MockModel(name='publishers/google/models/veo-2.0-generate-001'),
     ]
@@ -674,7 +674,7 @@ class MockModel:
     m1.description = 'Gemini model'
 
     m2 = MagicMock()
-    m2.name = 'publishers/google/models/text-embedding-004'
+    m2.name = 'publishers/google/models/gemini-embedding-001'
     m2.supported_actions = ['embedContent']
     m2.description = 'Embedder'
 
@@ -698,7 +698,7 @@ class MockModel:
     assert action1 is not None
 
     # Verify Embedder
-    action2 = next(a for a in result if a.name == vertexai_name('text-embedding-004'))
+    action2 = next(a for a in result if a.name == vertexai_name('gemini-embedding-001'))
     assert action2 is not None
 
     # Verify Imagen