""" Copyright 2024, Zep Software, Inc. Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. """ import asyncio import logging from datetime import datetime from time import time from dotenv import load_dotenv from neo4j import AsyncGraphDatabase from graphiti_core.edges import EntityEdge, EpisodicEdge from graphiti_core.llm_client import LLMClient, OpenAIClient from graphiti_core.llm_client.utils import generate_embedding from graphiti_core.nodes import EntityNode, EpisodeType, EpisodicNode from graphiti_core.search.search import Reranker, SearchConfig, SearchMethod, hybrid_search from graphiti_core.search.search_utils import ( RELEVANT_SCHEMA_LIMIT, get_relevant_edges, get_relevant_nodes, hybrid_node_search, ) from graphiti_core.utils import ( build_episodic_edges, retrieve_episodes, ) from graphiti_core.utils.bulk_utils import ( RawEpisode, dedupe_edges_bulk, dedupe_nodes_bulk, extract_edge_dates_bulk, extract_nodes_and_edges_bulk, resolve_edge_pointers, retrieve_previous_episodes_bulk, ) from graphiti_core.utils.maintenance.edge_operations import ( extract_edges, resolve_extracted_edges, ) from graphiti_core.utils.maintenance.graph_data_operations import ( EPISODE_WINDOW_LEN, build_indices_and_constraints, ) from graphiti_core.utils.maintenance.node_operations import ( extract_nodes, resolve_extracted_nodes, ) logger = logging.getLogger(__name__) load_dotenv() class Graphiti: def __init__(self, uri: str, user: str, password: str, llm_client: LLMClient | None = None): """ Initialize a Graphiti instance. This constructor sets up a connection to the Neo4j database and initializes the LLM client for natural language processing tasks. Parameters ---------- uri : str The URI of the Neo4j database. user : str The username for authenticating with the Neo4j database. password : str The password for authenticating with the Neo4j database. llm_client : LLMClient | None, optional An instance of LLMClient for natural language processing tasks. If not provided, a default OpenAIClient will be initialized. Returns ------- None Notes ----- This method establishes a connection to the Neo4j database using the provided credentials. It also sets up the LLM client, either using the provided client or by creating a default OpenAIClient. The default database name is set to 'neo4j'. If a different database name is required, it should be specified in the URI or set separately after initialization. The OpenAI API key is expected to be set in the environment variables. Make sure to set the OPENAI_API_KEY environment variable before initializing Graphiti if you're using the default OpenAIClient. """ self.driver = AsyncGraphDatabase.driver(uri, auth=(user, password)) self.database = 'neo4j' if llm_client: self.llm_client = llm_client else: self.llm_client = OpenAIClient() def close(self): """ Close the connection to the Neo4j database. This method safely closes the driver connection to the Neo4j database. It should be called when the Graphiti instance is no longer needed or when the application is shutting down. Parameters ---------- self Returns ------- None Notes ----- It's important to close the driver connection to release system resources and ensure that all pending transactions are completed or rolled back. This method should be called as part of a cleanup process, potentially in a context manager or a shutdown hook. Example: graphiti = Graphiti(uri, user, password) try: # Use graphiti... finally: graphiti.close() self.driver.close() """ async def build_indices_and_constraints(self): """ Build indices and constraints in the Neo4j database. This method sets up the necessary indices and constraints in the Neo4j database to optimize query performance and ensure data integrity for the knowledge graph. Parameters ---------- self Returns ------- None Notes ----- This method should typically be called once during the initial setup of the knowledge graph or when updating the database schema. It uses the `build_indices_and_constraints` function from the `graphiti_core.utils.maintenance.graph_data_operations` module to perform the actual database operations. The specific indices and constraints created depend on the implementation of the `build_indices_and_constraints` function. Refer to that function's documentation for details on the exact database schema modifications. Caution: Running this method on a large existing database may take some time and could impact database performance during execution. """ await build_indices_and_constraints(self.driver) async def retrieve_episodes( self, reference_time: datetime, last_n: int = EPISODE_WINDOW_LEN, group_ids: list[str | None] | None = None, ) -> list[EpisodicNode]: """ Retrieve the last n episodic nodes from the graph. This method fetches a specified number of the most recent episodic nodes from the graph, relative to the given reference time. Parameters ---------- reference_time : datetime The reference time to retrieve episodes before. last_n : int, optional The number of episodes to retrieve. Defaults to EPISODE_WINDOW_LEN. group_ids : list[str | None], optional The group ids to return data from. Returns ------- list[EpisodicNode] A list of the most recent EpisodicNode objects. Notes ----- The actual retrieval is performed by the `retrieve_episodes` function from the `graphiti_core.utils` module. """ return await retrieve_episodes(self.driver, reference_time, last_n, group_ids) async def add_episode( self, name: str, episode_body: str, source_description: str, reference_time: datetime, source: EpisodeType = EpisodeType.message, group_id: str | None = None, uuid: str | None = None, ): """ Process an episode and update the graph. This method extracts information from the episode, creates nodes and edges, and updates the graph database accordingly. Parameters ---------- name : str The name of the episode. episode_body : str The content of the episode. source_description : str A description of the episode's source. reference_time : datetime The reference time for the episode. source : EpisodeType, optional The type of the episode. Defaults to EpisodeType.message. group_id : str | None An id for the graph partition the episode is a part of. uuid : str | None Optional uuid of the episode. Returns ------- None Notes ----- This method performs several steps including node extraction, edge extraction, deduplication, and database updates. It also handles embedding generation and edge invalidation. It is recommended to run this method as a background process, such as in a queue. It's important that each episode is added sequentially and awaited before adding the next one. For web applications, consider using FastAPI's background tasks or a dedicated task queue like Celery for this purpose. Example using FastAPI background tasks: @app.post("/add_episode") async def add_episode_endpoint(episode_data: EpisodeData): background_tasks.add_task(graphiti.add_episode, **episode_data.dict()) return {"message": "Episode processing started"} """ try: start = time() nodes: list[EntityNode] = [] entity_edges: list[EntityEdge] = [] embedder = self.llm_client.get_embedder() now = datetime.now() previous_episodes = await self.retrieve_episodes( reference_time, last_n=3, group_ids=[group_id] ) episode = EpisodicNode( name=name, group_id=group_id, labels=[], source=source, content=episode_body, source_description=source_description, created_at=now, valid_at=reference_time, ) episode.uuid = uuid if uuid is not None else episode.uuid # Extract entities as nodes extracted_nodes = await extract_nodes(self.llm_client, episode, previous_episodes) logger.info(f'Extracted nodes: {[(n.name, n.uuid) for n in extracted_nodes]}') # Calculate Embeddings await asyncio.gather( *[node.generate_name_embedding(embedder) for node in extracted_nodes] ) # Resolve extracted nodes with nodes already in the graph and extract facts existing_nodes_lists: list[list[EntityNode]] = list( await asyncio.gather( *[get_relevant_nodes([node], self.driver) for node in extracted_nodes] ) ) logger.info(f'Extracted nodes: {[(n.name, n.uuid) for n in extracted_nodes]}') (mentioned_nodes, uuid_map), extracted_edges = await asyncio.gather( resolve_extracted_nodes(self.llm_client, extracted_nodes, existing_nodes_lists), extract_edges( self.llm_client, episode, extracted_nodes, previous_episodes, group_id ), ) logger.info(f'Adjusted mentioned nodes: {[(n.name, n.uuid) for n in mentioned_nodes]}') nodes.extend(mentioned_nodes) extracted_edges_with_resolved_pointers = resolve_edge_pointers( extracted_edges, uuid_map ) # calculate embeddings await asyncio.gather( *[ edge.generate_embedding(embedder) for edge in extracted_edges_with_resolved_pointers ] ) # Resolve extracted edges with related edges already in the graph related_edges_list: list[list[EntityEdge]] = list( await asyncio.gather( *[ get_relevant_edges( self.driver, [edge], edge.source_node_uuid, edge.target_node_uuid, RELEVANT_SCHEMA_LIMIT, ) for edge in extracted_edges_with_resolved_pointers ] ) ) logger.info( f'Related edges lists: {[(e.name, e.uuid) for edges_lst in related_edges_list for e in edges_lst]}' ) logger.info( f'Extracted edges: {[(e.name, e.uuid) for e in extracted_edges_with_resolved_pointers]}' ) existing_source_edges_list: list[list[EntityEdge]] = list( await asyncio.gather( *[ get_relevant_edges( self.driver, [edge], edge.source_node_uuid, None, RELEVANT_SCHEMA_LIMIT, ) for edge in extracted_edges_with_resolved_pointers ] ) ) existing_target_edges_list: list[list[EntityEdge]] = list( await asyncio.gather( *[ get_relevant_edges( self.driver, [edge], None, edge.target_node_uuid, RELEVANT_SCHEMA_LIMIT, ) for edge in extracted_edges_with_resolved_pointers ] ) ) existing_edges_list: list[list[EntityEdge]] = [ source_lst + target_lst for source_lst, target_lst in zip( existing_source_edges_list, existing_target_edges_list ) ] resolved_edges, invalidated_edges = await resolve_extracted_edges( self.llm_client, extracted_edges_with_resolved_pointers, related_edges_list, existing_edges_list, episode, previous_episodes, ) entity_edges.extend(resolved_edges + invalidated_edges) logger.info(f'Resolved edges: {[(e.name, e.uuid) for e in resolved_edges]}') episodic_edges: list[EpisodicEdge] = build_episodic_edges(mentioned_nodes, episode, now) logger.info(f'Built episodic edges: {episodic_edges}') # Future optimization would be using batch operations to save nodes and edges await episode.save(self.driver) await asyncio.gather(*[node.save(self.driver) for node in nodes]) await asyncio.gather(*[edge.save(self.driver) for edge in episodic_edges]) await asyncio.gather(*[edge.save(self.driver) for edge in entity_edges]) end = time() logger.info(f'Completed add_episode in {(end - start) * 1000} ms') except Exception as e: raise e async def add_episode_bulk(self, bulk_episodes: list[RawEpisode], group_id: str | None): """ Process multiple episodes in bulk and update the graph. This method extracts information from multiple episodes, creates nodes and edges, and updates the graph database accordingly, all in a single batch operation. Parameters ---------- bulk_episodes : list[RawEpisode] A list of RawEpisode objects to be processed and added to the graph. group_id : str | None An id for the graph partition the episode is a part of. Returns ------- None Notes ----- This method performs several steps including: - Saving all episodes to the database - Retrieving previous episode context for each new episode - Extracting nodes and edges from all episodes - Generating embeddings for nodes and edges - Deduplicating nodes and edges - Saving nodes, episodic edges, and entity edges to the knowledge graph This bulk operation is designed for efficiency when processing multiple episodes at once. However, it's important to ensure that the bulk operation doesn't overwhelm system resources. Consider implementing rate limiting or chunking for very large batches of episodes. Important: This method does not perform edge invalidation or date extraction steps. If these operations are required, use the `add_episode` method instead for each individual episode. """ try: start = time() embedder = self.llm_client.get_embedder() now = datetime.now() episodes = [ EpisodicNode( name=episode.name, labels=[], source=episode.source, content=episode.content, source_description=episode.source_description, group_id=group_id, created_at=now, valid_at=episode.reference_time, ) for episode in bulk_episodes ] # Save all the episodes await asyncio.gather(*[episode.save(self.driver) for episode in episodes]) # Get previous episode context for each episode episode_pairs = await retrieve_previous_episodes_bulk(self.driver, episodes) # Extract all nodes and edges ( extracted_nodes, extracted_edges, episodic_edges, ) = await extract_nodes_and_edges_bulk(self.llm_client, episode_pairs) # Generate embeddings await asyncio.gather( *[node.generate_name_embedding(embedder) for node in extracted_nodes], *[edge.generate_embedding(embedder) for edge in extracted_edges], ) # Dedupe extracted nodes, compress extracted edges (nodes, uuid_map), extracted_edges_timestamped = await asyncio.gather( dedupe_nodes_bulk(self.driver, self.llm_client, extracted_nodes), extract_edge_dates_bulk(self.llm_client, extracted_edges, episode_pairs), ) # save nodes to KG await asyncio.gather(*[node.save(self.driver) for node in nodes]) # re-map edge pointers so that they don't point to discard dupe nodes extracted_edges_with_resolved_pointers: list[EntityEdge] = resolve_edge_pointers( extracted_edges_timestamped, uuid_map ) episodic_edges_with_resolved_pointers: list[EpisodicEdge] = resolve_edge_pointers( episodic_edges, uuid_map ) # save episodic edges to KG await asyncio.gather( *[edge.save(self.driver) for edge in episodic_edges_with_resolved_pointers] ) # Dedupe extracted edges edges = await dedupe_edges_bulk( self.driver, self.llm_client, extracted_edges_with_resolved_pointers ) logger.info(f'extracted edge length: {len(edges)}') # invalidate edges # save edges to KG await asyncio.gather(*[edge.save(self.driver) for edge in edges]) end = time() logger.info(f'Completed add_episode_bulk in {(end - start) * 1000} ms') except Exception as e: raise e async def search( self, query: str, center_node_uuid: str | None = None, group_ids: list[str | None] | None = None, num_results=10, ): """ Perform a hybrid search on the knowledge graph. This method executes a search query on the graph, combining vector and text-based search techniques to retrieve relevant facts. Parameters ---------- query : str The search query string. center_node_uuid: str, optional Facts will be reranked based on proximity to this node group_ids : list[str | None] | None, optional The graph partitions to return data from. num_results : int, optional The maximum number of results to return. Defaults to 10. Returns ------- list A list of EntityEdge objects that are relevant to the search query. Notes ----- This method uses a SearchConfig with num_episodes set to 0 and num_results set to the provided num_results parameter. It then calls the hybrid_search function to perform the actual search operation. The search is performed using the current date and time as the reference point for temporal relevance. """ reranker = Reranker.rrf if center_node_uuid is None else Reranker.node_distance search_config = SearchConfig( num_episodes=0, num_edges=num_results, num_nodes=0, group_ids=group_ids, search_methods=[SearchMethod.bm25, SearchMethod.cosine_similarity], reranker=reranker, ) edges = ( await hybrid_search( self.driver, self.llm_client.get_embedder(), query, datetime.now(), search_config, center_node_uuid, ) ).edges return edges async def _search( self, query: str, timestamp: datetime, config: SearchConfig, center_node_uuid: str | None = None, ): return await hybrid_search( self.driver, self.llm_client.get_embedder(), query, timestamp, config, center_node_uuid ) async def get_nodes_by_query( self, query: str, group_ids: list[str | None] | None = None, limit: int = RELEVANT_SCHEMA_LIMIT, ) -> list[EntityNode]: """ Retrieve nodes from the graph database based on a text query. This method performs a hybrid search using both text-based and embedding-based approaches to find relevant nodes. Parameters ---------- query : str The text query to search for in the graph. group_ids : list[str | None] | None, optional The graph partitions to return data from. limit : int | None, optional The maximum number of results to return per search method. If None, a default limit will be applied. Returns ------- list[EntityNode] A list of EntityNode objects that match the search criteria. Notes ----- This method uses the following steps: 1. Generates an embedding for the input query using the LLM client's embedder. 2. Calls the hybrid_node_search function with both the text query and its embedding. 3. The hybrid search combines fulltext search and vector similarity search to find the most relevant nodes. The method leverages the LLM client's embedding capabilities to enhance the search with semantic similarity matching. The 'limit' parameter is applied to each individual search method before results are combined and deduplicated. If not specified, a default limit (defined in the search functions) will be used. """ embedder = self.llm_client.get_embedder() query_embedding = await generate_embedding(embedder, query) relevant_nodes = await hybrid_node_search( [query], [query_embedding], self.driver, group_ids, limit ) return relevant_nodes