autogpt 还原，开新分支编写

autogpt/memory/__init__.py

@@ -0,0 +1,99 @@
+from autogpt.memory.local import LocalCache
+from autogpt.memory.no_memory import NoMemory
+
+# List of supported memory backends
+# Add a backend to this list if the import attempt is successful
+supported_memory = ["local", "no_memory"]
+
+try:
+    from autogpt.memory.redismem import RedisMemory
+
+    supported_memory.append("redis")
+except ImportError:
+    # print("Redis not installed. Skipping import.")
+    RedisMemory = None
+
+try:
+    from autogpt.memory.pinecone import PineconeMemory
+
+    supported_memory.append("pinecone")
+except ImportError:
+    # print("Pinecone not installed. Skipping import.")
+    PineconeMemory = None
+
+try:
+    from autogpt.memory.weaviate import WeaviateMemory
+
+    supported_memory.append("weaviate")
+except ImportError:
+    # print("Weaviate not installed. Skipping import.")
+    WeaviateMemory = None
+
+try:
+    from autogpt.memory.milvus import MilvusMemory
+
+    supported_memory.append("milvus")
+except ImportError:
+    # print("pymilvus not installed. Skipping import.")
+    MilvusMemory = None
+
+
+def get_memory(cfg, init=False):
+    memory = None
+    if cfg.memory_backend == "pinecone":
+        if not PineconeMemory:
+            print(
+                "Error: Pinecone is not installed. Please install pinecone"
+                " to use Pinecone as a memory backend."
+            )
+        else:
+            memory = PineconeMemory(cfg)
+            if init:
+                memory.clear()
+    elif cfg.memory_backend == "redis":
+        if not RedisMemory:
+            print(
+                "Error: Redis is not installed. Please install redis-py to"
+                " use Redis as a memory backend."
+            )
+        else:
+            memory = RedisMemory(cfg)
+    elif cfg.memory_backend == "weaviate":
+        if not WeaviateMemory:
+            print(
+                "Error: Weaviate is not installed. Please install weaviate-client to"
+                " use Weaviate as a memory backend."
+            )
+        else:
+            memory = WeaviateMemory(cfg)
+    elif cfg.memory_backend == "milvus":
+        if not MilvusMemory:
+            print(
+                "Error: pymilvus sdk is not installed."
+                "Please install pymilvus to use Milvus or Zilliz Cloud as memory backend."
+            )
+        else:
+            memory = MilvusMemory(cfg)
+    elif cfg.memory_backend == "no_memory":
+        memory = NoMemory(cfg)
+
+    if memory is None:
+        memory = LocalCache(cfg)
+        if init:
+            memory.clear()
+    return memory
+
+
+def get_supported_memory_backends():
+    return supported_memory
+
+
+__all__ = [
+    "get_memory",
+    "LocalCache",
+    "RedisMemory",
+    "PineconeMemory",
+    "NoMemory",
+    "MilvusMemory",
+    "WeaviateMemory",
+]

autogpt/memory/base.py

@@ -0,0 +1,28 @@
+"""Base class for memory providers."""
+import abc
+
+from autogpt.config import AbstractSingleton, Config
+
+cfg = Config()
+
+
+class MemoryProviderSingleton(AbstractSingleton):
+    @abc.abstractmethod
+    def add(self, data):
+        pass
+
+    @abc.abstractmethod
+    def get(self, data):
+        pass
+
+    @abc.abstractmethod
+    def clear(self):
+        pass
+
+    @abc.abstractmethod
+    def get_relevant(self, data, num_relevant=5):
+        pass
+
+    @abc.abstractmethod
+    def get_stats(self):
+        pass

autogpt/memory/local.py

@@ -0,0 +1,126 @@
+from __future__ import annotations
+
+import dataclasses
+from pathlib import Path
+from typing import Any, List
+
+import numpy as np
+import orjson
+
+from autogpt.llm_utils import create_embedding_with_ada
+from autogpt.memory.base import MemoryProviderSingleton
+
+EMBED_DIM = 1536
+SAVE_OPTIONS = orjson.OPT_SERIALIZE_NUMPY | orjson.OPT_SERIALIZE_DATACLASS
+
+
+def create_default_embeddings():
+    return np.zeros((0, EMBED_DIM)).astype(np.float32)
+
+
+@dataclasses.dataclass
+class CacheContent:
+    texts: List[str] = dataclasses.field(default_factory=list)
+    embeddings: np.ndarray = dataclasses.field(
+        default_factory=create_default_embeddings
+    )
+
+
+class LocalCache(MemoryProviderSingleton):
+    """A class that stores the memory in a local file"""
+
+    def __init__(self, cfg) -> None:
+        """Initialize a class instance
+
+        Args:
+            cfg: Config object
+
+        Returns:
+            None
+        """
+        workspace_path = Path(cfg.workspace_path)
+        self.filename = workspace_path / f"{cfg.memory_index}.json"
+
+        self.filename.touch(exist_ok=True)
+
+        file_content = b"{}"
+        with self.filename.open("w+b") as f:
+            f.write(file_content)
+
+        self.data = CacheContent()
+
+    def add(self, text: str):
+        """
+        Add text to our list of texts, add embedding as row to our
+            embeddings-matrix
+
+        Args:
+            text: str
+
+        Returns: None
+        """
+        if "Command Error:" in text:
+            return ""
+        self.data.texts.append(text)
+
+        embedding = create_embedding_with_ada(text)
+
+        vector = np.array(embedding).astype(np.float32)
+        vector = vector[np.newaxis, :]
+        self.data.embeddings = np.concatenate(
+            [
+                self.data.embeddings,
+                vector,
+            ],
+            axis=0,
+        )
+
+        with open(self.filename, "wb") as f:
+            out = orjson.dumps(self.data, option=SAVE_OPTIONS)
+            f.write(out)
+        return text
+
+    def clear(self) -> str:
+        """
+        Clears the redis server.
+
+        Returns: A message indicating that the memory has been cleared.
+        """
+        self.data = CacheContent()
+        return "Obliviated"
+
+    def get(self, data: str) -> list[Any] | None:
+        """
+        Gets the data from the memory that is most relevant to the given data.
+
+        Args:
+            data: The data to compare to.
+
+        Returns: The most relevant data.
+        """
+        return self.get_relevant(data, 1)
+
+    def get_relevant(self, text: str, k: int) -> list[Any]:
+        """ "
+        matrix-vector mult to find score-for-each-row-of-matrix
+         get indices for top-k winning scores
+         return texts for those indices
+        Args:
+            text: str
+            k: int
+
+        Returns: List[str]
+        """
+        embedding = create_embedding_with_ada(text)
+
+        scores = np.dot(self.data.embeddings, embedding)
+
+        top_k_indices = np.argsort(scores)[-k:][::-1]
+
+        return [self.data.texts[i] for i in top_k_indices]
+
+    def get_stats(self) -> tuple[int, tuple[int, ...]]:
+        """
+        Returns: The stats of the local cache.
+        """
+        return len(self.data.texts), self.data.embeddings.shape

autogpt/memory/milvus.py

@@ -0,0 +1,162 @@
+""" Milvus memory storage provider."""
+import re
+
+from pymilvus import Collection, CollectionSchema, DataType, FieldSchema, connections
+
+from autogpt.config import Config
+from autogpt.llm_utils import get_ada_embedding
+from autogpt.memory.base import MemoryProviderSingleton
+
+
+class MilvusMemory(MemoryProviderSingleton):
+    """Milvus memory storage provider."""
+
+    def __init__(self, cfg: Config) -> None:
+        """Construct a milvus memory storage connection.
+
+        Args:
+            cfg (Config): Auto-GPT global config.
+        """
+        self.configure(cfg)
+
+        connect_kwargs = {}
+        if self.username:
+            connect_kwargs["user"] = self.username
+            connect_kwargs["password"] = self.password
+
+        connections.connect(
+            **connect_kwargs,
+            uri=self.uri or "",
+            address=self.address or "",
+            secure=self.secure,
+        )
+
+        self.init_collection()
+
+    def configure(self, cfg: Config) -> None:
+        # init with configuration.
+        self.uri = None
+        self.address = cfg.milvus_addr
+        self.secure = cfg.milvus_secure
+        self.username = cfg.milvus_username
+        self.password = cfg.milvus_password
+        self.collection_name = cfg.milvus_collection
+        # use HNSW by default.
+        self.index_params = {
+            "metric_type": "IP",
+            "index_type": "HNSW",
+            "params": {"M": 8, "efConstruction": 64},
+        }
+
+        if (self.username is None) != (self.password is None):
+            raise ValueError(
+                "Both username and password must be set to use authentication for Milvus"
+            )
+
+        # configured address may be a full URL.
+        if re.match(r"^(https?|tcp)://", self.address) is not None:
+            self.uri = self.address
+            self.address = None
+
+            if self.uri.startswith("https"):
+                self.secure = True
+
+        # Zilliz Cloud requires AutoIndex.
+        if re.match(r"^https://(.*)\.zillizcloud\.(com|cn)", self.address) is not None:
+            self.index_params = {
+                "metric_type": "IP",
+                "index_type": "AUTOINDEX",
+                "params": {},
+            }
+
+    def init_collection(self) -> None:
+        """Initialize collection in vector database."""
+        fields = [
+            FieldSchema(name="pk", dtype=DataType.INT64, is_primary=True, auto_id=True),
+            FieldSchema(name="embeddings", dtype=DataType.FLOAT_VECTOR, dim=1536),
+            FieldSchema(name="raw_text", dtype=DataType.VARCHAR, max_length=65535),
+        ]
+
+        # create collection if not exist and load it.
+        self.schema = CollectionSchema(fields, "auto-gpt memory storage")
+        self.collection = Collection(self.collection_name, self.schema)
+        # create index if not exist.
+        if not self.collection.has_index():
+            self.collection.release()
+            self.collection.create_index(
+                "embeddings",
+                self.index_params,
+                index_name="embeddings",
+            )
+        self.collection.load()
+
+    def add(self, data) -> str:
+        """Add an embedding of data into memory.
+
+        Args:
+            data (str): The raw text to construct embedding index.
+
+        Returns:
+            str: log.
+        """
+        embedding = get_ada_embedding(data)
+        result = self.collection.insert([[embedding], [data]])
+        _text = (
+            "Inserting data into memory at primary key: "
+            f"{result.primary_keys[0]}:\n data: {data}"
+        )
+        return _text
+
+    def get(self, data):
+        """Return the most relevant data in memory.
+        Args:
+            data: The data to compare to.
+        """
+        return self.get_relevant(data, 1)
+
+    def clear(self) -> str:
+        """Drop the index in memory.
+
+        Returns:
+            str: log.
+        """
+        self.collection.drop()
+        self.collection = Collection(self.collection_name, self.schema)
+        self.collection.create_index(
+            "embeddings",
+            self.index_params,
+            index_name="embeddings",
+        )
+        self.collection.load()
+        return "Obliviated"
+
+    def get_relevant(self, data: str, num_relevant: int = 5):
+        """Return the top-k relevant data in memory.
+        Args:
+            data: The data to compare to.
+            num_relevant (int, optional): The max number of relevant data.
+                Defaults to 5.
+
+        Returns:
+            list: The top-k relevant data.
+        """
+        # search the embedding and return the most relevant text.
+        embedding = get_ada_embedding(data)
+        search_params = {
+            "metrics_type": "IP",
+            "params": {"nprobe": 8},
+        }
+        result = self.collection.search(
+            [embedding],
+            "embeddings",
+            search_params,
+            num_relevant,
+            output_fields=["raw_text"],
+        )
+        return [item.entity.value_of_field("raw_text") for item in result[0]]
+
+    def get_stats(self) -> str:
+        """
+        Returns: The stats of the milvus cache.
+        """
+        return f"Entities num: {self.collection.num_entities}"

autogpt/memory/no_memory.py

@@ -0,0 +1,73 @@
+"""A class that does not store any data. This is the default memory provider."""
+from __future__ import annotations
+
+from typing import Any
+
+from autogpt.memory.base import MemoryProviderSingleton
+
+
+class NoMemory(MemoryProviderSingleton):
+    """
+    A class that does not store any data. This is the default memory provider.
+    """
+
+    def __init__(self, cfg):
+        """
+        Initializes the NoMemory provider.
+
+        Args:
+            cfg: The config object.
+
+        Returns: None
+        """
+        pass
+
+    def add(self, data: str) -> str:
+        """
+        Adds a data point to the memory. No action is taken in NoMemory.
+
+        Args:
+            data: The data to add.
+
+        Returns: An empty string.
+        """
+        return ""
+
+    def get(self, data: str) -> list[Any] | None:
+        """
+        Gets the data from the memory that is most relevant to the given data.
+        NoMemory always returns None.
+
+        Args:
+            data: The data to compare to.
+
+        Returns: None
+        """
+        return None
+
+    def clear(self) -> str:
+        """
+        Clears the memory. No action is taken in NoMemory.
+
+        Returns: An empty string.
+        """
+        return ""
+
+    def get_relevant(self, data: str, num_relevant: int = 5) -> list[Any] | None:
+        """
+        Returns all the data in the memory that is relevant to the given data.
+        NoMemory always returns None.
+
+        Args:
+            data: The data to compare to.
+            num_relevant: The number of relevant data to return.
+
+        Returns: None
+        """
+        return None
+
+    def get_stats(self):
+        """
+        Returns: An empty dictionary as there are no stats in NoMemory.
+        """
+        return {}

autogpt/memory/pinecone.py

@@ -0,0 +1,75 @@
+import pinecone
+from colorama import Fore, Style
+
+from autogpt.llm_utils import create_embedding_with_ada
+from autogpt.logs import logger
+from autogpt.memory.base import MemoryProviderSingleton
+
+
+class PineconeMemory(MemoryProviderSingleton):
+    def __init__(self, cfg):
+        pinecone_api_key = cfg.pinecone_api_key
+        pinecone_region = cfg.pinecone_region
+        pinecone.init(api_key=pinecone_api_key, environment=pinecone_region)
+        dimension = 1536
+        metric = "cosine"
+        pod_type = "p1"
+        table_name = "auto-gpt"
+        # this assumes we don't start with memory.
+        # for now this works.
+        # we'll need a more complicated and robust system if we want to start with
+        #  memory.
+        self.vec_num = 0
+
+        try:
+            pinecone.whoami()
+        except Exception as e:
+            logger.typewriter_log(
+                "FAILED TO CONNECT TO PINECONE",
+                Fore.RED,
+                Style.BRIGHT + str(e) + Style.RESET_ALL,
+            )
+            logger.double_check(
+                "Please ensure you have setup and configured Pinecone properly for use."
+                + f"You can check out {Fore.CYAN + Style.BRIGHT}"
+                "https://github.com/Torantulino/Auto-GPT#-pinecone-api-key-setup"
+                f"{Style.RESET_ALL} to ensure you've set up everything correctly."
+            )
+            exit(1)
+
+        if table_name not in pinecone.list_indexes():
+            pinecone.create_index(
+                table_name, dimension=dimension, metric=metric, pod_type=pod_type
+            )
+        self.index = pinecone.Index(table_name)
+
+    def add(self, data):
+        vector = create_embedding_with_ada(data)
+        # no metadata here. We may wish to change that long term.
+        self.index.upsert([(str(self.vec_num), vector, {"raw_text": data})])
+        _text = f"Inserting data into memory at index: {self.vec_num}:\n data: {data}"
+        self.vec_num += 1
+        return _text
+
+    def get(self, data):
+        return self.get_relevant(data, 1)
+
+    def clear(self):
+        self.index.delete(deleteAll=True)
+        return "Obliviated"
+
+    def get_relevant(self, data, num_relevant=5):
+        """
+        Returns all the data in the memory that is relevant to the given data.
+        :param data: The data to compare to.
+        :param num_relevant: The number of relevant data to return. Defaults to 5
+        """
+        query_embedding = create_embedding_with_ada(data)
+        results = self.index.query(
+            query_embedding, top_k=num_relevant, include_metadata=True
+        )
+        sorted_results = sorted(results.matches, key=lambda x: x.score)
+        return [str(item["metadata"]["raw_text"]) for item in sorted_results]
+
+    def get_stats(self):
+        return self.index.describe_index_stats()

autogpt/memory/redismem.py

@@ -0,0 +1,156 @@
+"""Redis memory provider."""
+from __future__ import annotations
+
+from typing import Any
+
+import numpy as np
+import redis
+from colorama import Fore, Style
+from redis.commands.search.field import TextField, VectorField
+from redis.commands.search.indexDefinition import IndexDefinition, IndexType
+from redis.commands.search.query import Query
+
+from autogpt.llm_utils import create_embedding_with_ada
+from autogpt.logs import logger
+from autogpt.memory.base import MemoryProviderSingleton
+
+SCHEMA = [
+    TextField("data"),
+    VectorField(
+        "embedding",
+        "HNSW",
+        {"TYPE": "FLOAT32", "DIM": 1536, "DISTANCE_METRIC": "COSINE"},
+    ),
+]
+
+
+class RedisMemory(MemoryProviderSingleton):
+    def __init__(self, cfg):
+        """
+        Initializes the Redis memory provider.
+
+        Args:
+            cfg: The config object.
+
+        Returns: None
+        """
+        redis_host = cfg.redis_host
+        redis_port = cfg.redis_port
+        redis_password = cfg.redis_password
+        self.dimension = 1536
+        self.redis = redis.Redis(
+            host=redis_host,
+            port=redis_port,
+            password=redis_password,
+            db=0,  # Cannot be changed
+        )
+        self.cfg = cfg
+
+        # Check redis connection
+        try:
+            self.redis.ping()
+        except redis.ConnectionError as e:
+            logger.typewriter_log(
+                "FAILED TO CONNECT TO REDIS",
+                Fore.RED,
+                Style.BRIGHT + str(e) + Style.RESET_ALL,
+            )
+            logger.double_check(
+                "Please ensure you have setup and configured Redis properly for use. "
+                + f"You can check out {Fore.CYAN + Style.BRIGHT}"
+                f"https://github.com/Torantulino/Auto-GPT#redis-setup{Style.RESET_ALL}"
+                " to ensure you've set up everything correctly."
+            )
+            exit(1)
+
+        if cfg.wipe_redis_on_start:
+            self.redis.flushall()
+        try:
+            self.redis.ft(f"{cfg.memory_index}").create_index(
+                fields=SCHEMA,
+                definition=IndexDefinition(
+                    prefix=[f"{cfg.memory_index}:"], index_type=IndexType.HASH
+                ),
+            )
+        except Exception as e:
+            print("Error creating Redis search index: ", e)
+        existing_vec_num = self.redis.get(f"{cfg.memory_index}-vec_num")
+        self.vec_num = int(existing_vec_num.decode("utf-8")) if existing_vec_num else 0
+
+    def add(self, data: str) -> str:
+        """
+        Adds a data point to the memory.
+
+        Args:
+            data: The data to add.
+
+        Returns: Message indicating that the data has been added.
+        """
+        if "Command Error:" in data:
+            return ""
+        vector = create_embedding_with_ada(data)
+        vector = np.array(vector).astype(np.float32).tobytes()
+        data_dict = {b"data": data, "embedding": vector}
+        pipe = self.redis.pipeline()
+        pipe.hset(f"{self.cfg.memory_index}:{self.vec_num}", mapping=data_dict)
+        _text = (
+            f"Inserting data into memory at index: {self.vec_num}:\n" f"data: {data}"
+        )
+        self.vec_num += 1
+        pipe.set(f"{self.cfg.memory_index}-vec_num", self.vec_num)
+        pipe.execute()
+        return _text
+
+    def get(self, data: str) -> list[Any] | None:
+        """
+        Gets the data from the memory that is most relevant to the given data.
+
+        Args:
+            data: The data to compare to.
+
+        Returns: The most relevant data.
+        """
+        return self.get_relevant(data, 1)
+
+    def clear(self) -> str:
+        """
+        Clears the redis server.
+
+        Returns: A message indicating that the memory has been cleared.
+        """
+        self.redis.flushall()
+        return "Obliviated"
+
+    def get_relevant(self, data: str, num_relevant: int = 5) -> list[Any] | None:
+        """
+        Returns all the data in the memory that is relevant to the given data.
+        Args:
+            data: The data to compare to.
+            num_relevant: The number of relevant data to return.
+
+        Returns: A list of the most relevant data.
+        """
+        query_embedding = create_embedding_with_ada(data)
+        base_query = f"*=>[KNN {num_relevant} @embedding $vector AS vector_score]"
+        query = (
+            Query(base_query)
+            .return_fields("data", "vector_score")
+            .sort_by("vector_score")
+            .dialect(2)
+        )
+        query_vector = np.array(query_embedding).astype(np.float32).tobytes()
+
+        try:
+            results = self.redis.ft(f"{self.cfg.memory_index}").search(
+                query, query_params={"vector": query_vector}
+            )
+        except Exception as e:
+            print("Error calling Redis search: ", e)
+            return None
+        return [result.data for result in results.docs]
+
+    def get_stats(self):
+        """
+        Returns: The stats of the memory index.
+        """
+        return self.redis.ft(f"{self.cfg.memory_index}").info()

autogpt/memory/weaviate.py

@@ -0,0 +1,126 @@
+import weaviate
+from weaviate import Client
+from weaviate.embedded import EmbeddedOptions
+from weaviate.util import generate_uuid5
+
+from autogpt.llm_utils import get_ada_embedding
+from autogpt.memory.base import MemoryProviderSingleton
+
+
+def default_schema(weaviate_index):
+    return {
+        "class": weaviate_index,
+        "properties": [
+            {
+                "name": "raw_text",
+                "dataType": ["text"],
+                "description": "original text for the embedding",
+            }
+        ],
+    }
+
+
+class WeaviateMemory(MemoryProviderSingleton):
+    def __init__(self, cfg):
+        auth_credentials = self._build_auth_credentials(cfg)
+
+        url = f"{cfg.weaviate_protocol}://{cfg.weaviate_host}:{cfg.weaviate_port}"
+
+        if cfg.use_weaviate_embedded:
+            self.client = Client(
+                embedded_options=EmbeddedOptions(
+                    hostname=cfg.weaviate_host,
+                    port=int(cfg.weaviate_port),
+                    persistence_data_path=cfg.weaviate_embedded_path,
+                )
+            )
+
+            print(
+                f"Weaviate Embedded running on: {url} with persistence path: {cfg.weaviate_embedded_path}"
+            )
+        else:
+            self.client = Client(url, auth_client_secret=auth_credentials)
+
+        self.index = WeaviateMemory.format_classname(cfg.memory_index)
+        self._create_schema()
+
+    @staticmethod
+    def format_classname(index):
+        # weaviate uses capitalised index names
+        # The python client uses the following code to format
+        # index names before the corresponding class is created
+        index = index.replace("-", "_")
+        if len(index) == 1:
+            return index.capitalize()
+        return index[0].capitalize() + index[1:]
+
+    def _create_schema(self):
+        schema = default_schema(self.index)
+        if not self.client.schema.contains(schema):
+            self.client.schema.create_class(schema)
+
+    def _build_auth_credentials(self, cfg):
+        if cfg.weaviate_username and cfg.weaviate_password:
+            return weaviate.AuthClientPassword(
+                cfg.weaviate_username, cfg.weaviate_password
+            )
+        if cfg.weaviate_api_key:
+            return weaviate.AuthApiKey(api_key=cfg.weaviate_api_key)
+        else:
+            return None
+
+    def add(self, data):
+        vector = get_ada_embedding(data)
+
+        doc_uuid = generate_uuid5(data, self.index)
+        data_object = {"raw_text": data}
+
+        with self.client.batch as batch:
+            batch.add_data_object(
+                uuid=doc_uuid,
+                data_object=data_object,
+                class_name=self.index,
+                vector=vector,
+            )
+
+        return f"Inserting data into memory at uuid: {doc_uuid}:\n data: {data}"
+
+    def get(self, data):
+        return self.get_relevant(data, 1)
+
+    def clear(self):
+        self.client.schema.delete_all()
+
+        # weaviate does not yet have a neat way to just remove the items in an index
+        # without removing the entire schema, therefore we need to re-create it
+        # after a call to delete_all
+        self._create_schema()
+
+        return "Obliterated"
+
+    def get_relevant(self, data, num_relevant=5):
+        query_embedding = get_ada_embedding(data)
+        try:
+            results = (
+                self.client.query.get(self.index, ["raw_text"])
+                .with_near_vector({"vector": query_embedding, "certainty": 0.7})
+                .with_limit(num_relevant)
+                .do()
+            )
+
+            if len(results["data"]["Get"][self.index]) > 0:
+                return [
+                    str(item["raw_text"]) for item in results["data"]["Get"][self.index]
+                ]
+            else:
+                return []
+
+        except Exception as err:
+            print(f"Unexpected error {err=}, {type(err)=}")
+            return []
+
+    def get_stats(self):
+        result = self.client.query.aggregate(self.index).with_meta_count().do()
+        class_data = result["data"]["Aggregate"][self.index]
+
+        return class_data[0]["meta"] if class_data else {}