templates: Add neo4j semantic layer with ollama template (#17192)

A template with JSON-based agent using Mixtral via Ollama.

---------

Co-authored-by: Erick Friis <erick@langchain.dev>

templates/neo4j-semantic-ollama/neo4j_semantic_ollama/memory_tool.py

@@ -0,0 +1,73 @@
+from typing import Optional, Type
+
+from langchain.callbacks.manager import (
+    AsyncCallbackManagerForToolRun,
+    CallbackManagerForToolRun,
+)
+
+# Import things that are needed generically
+from langchain.pydantic_v1 import BaseModel, Field
+from langchain.tools import BaseTool
+
+from neo4j_semantic_ollama.utils import get_candidates, get_user_id, graph
+
+store_rating_query = """
+MERGE (u:User {userId:$user_id}) 
+WITH u 
+UNWIND $candidates as row
+MATCH (m:Movie {title: row.candidate})
+MERGE (u)-[r:RATED]->(m)
+SET r.rating = toFloat($rating)
+RETURN distinct 
+   'Create a final answer saying that preference has been stored' AS response
+"""
+
+
+def store_movie_rating(movie: str, rating: int):
+    user_id = get_user_id()
+    candidates = get_candidates(movie, "movie")
+    if not candidates:
+        return "This movie is not in our database"
+    response = graph.query(
+        store_rating_query,
+        params={"user_id": user_id, "candidates": candidates, "rating": rating},
+    )
+    try:
+        return response[0]["response"]
+    except Exception as e:
+        print(e)
+        return "Something went wrong"
+
+
+class MemoryInput(BaseModel):
+    movie: str = Field(description="movie the user liked")
+    rating: int = Field(
+        description=(
+            "Rating from 1 to 5, where one represents heavy dislike "
+            "and 5 represent the user loved the movie"
+        )
+    )
+
+
+class MemoryTool(BaseTool):
+    name = "Memory"
+    description = "useful for memorizing which movies the user liked"
+    args_schema: Type[BaseModel] = MemoryInput
+
+    def _run(
+        self,
+        movie: str,
+        rating: int,
+        run_manager: Optional[CallbackManagerForToolRun] = None,
+    ) -> str:
+        """Use the tool."""
+        return store_movie_rating(movie, rating)
+
+    async def _arun(
+        self,
+        movie: str,
+        rating: int,
+        run_manager: Optional[AsyncCallbackManagerForToolRun] = None,
+    ) -> str:
+        """Use the tool asynchronously."""
+        return store_movie_rating(movie, rating)