zep-python 0.21

Zep stores, manages, enriches, and searches long-term memory for conversational AI applications. This is the Python client for the Zep service.

Zep: A long-term memory store for conversational AI applications

This is the Python client package for the Zep service. For more information about Zep, see https://github.com/getzep/zep.

Installation

pip install zep-python

or

poetry add zep-python

Quick Start

Ensure that you have a Zep server running. See https://github.com/getzep/zep.

import asyncio

from zep_python import Memory, Message, SearchPayload, ZepClient

base_url = "http://localhost:8000"  # TODO: Replace with Zep API URL
session_id = "2a2a2a" # an identifier for your user's session.

async with ZepClient(base_url) as client:

    role = "user"
    content = "who was the first man to go to space?"
    message = Message(role=role, content=content)
    memory = Memory()
    memory.messages = [message]
    # Add a memory
    result = await client.aadd_memory(session_id, memory)

    # Long chat histories will automatically be summarized.
    # A summary and chat history are returned with a `get_memory`
    memories = await client.aget_memory(session_id)
    for memory in memories:
        for message in memory.messages:
            print(message.to_dict())

    # Search memory
    # Messages uploaded to Zep are automatically embedded and made available
    # for vector-based similarity search.
    search_payload = SearchPayload({}, "Who is Yuri Gagarin?")
    search_results = await client.asearch_memory(session_id, search_payload)
    for search_result in search_results:
        # Access the 'content' field within the 'message' object.
        message_content = search_result.message
        print(message_content)

Zep Python

Zep Python has both an async and sync API. We've provided code examples for the async API, but theyre easily enough modified for sync usage.

Class Attributes

• base_url (str): The base URL of the API.
• client (httpx.AsyncClient): The HTTP client used for making API requests.

Methods

• get_memory(session_id: str, lastn: Optional[int] = None) -> List[Memory]: Retrieve memory for the specified session.
• add_memory(session_id: str, memory_messages: Memory) -> str: Add memory to the specified session.
• delete_memory(session_id: str) -> str: Delete memory for the specified session.
• search_memory(session_id: str, search_payload: SearchPayload, limit: Optional[int] = None) -> List[SearchResult]: Search memory for the specified session.
• close() -> None: Close the HTTP client.

__init__(self, base_url: str) -> None

Initialize the ZepClient with the specified base URL.

Parameters

• base_url (str): The base URL of the API.

base_url = "http://localhost:8000"  # TODO: Replace with your Zep API URL
async with ZepClient(base_url) as client:
    # ...

---

get_memory(self, session_id: str, lastn: Optional[int] = None) -> List[Memory]

Retrieve memory for the specified session. This method is a synchronous wrapper for the asynchronous method aget_memory.

Parameters

• session_id (str): The ID of the session for which to retrieve memory.
• lastn (Optional[int], optional): The number of most recent memory entries to retrieve. Defaults to None (all entries).

Returns

• List[Memory]: A list of Memory objects representing the retrieved memory entries.

Raises

• APIError: If the API response format is unexpected.
• NotFoundError: If no results were found.

---

aget_memory(self, session_id: str, lastn: Optional[int] = None) -> List[Memory]

Asynchronously retrieve memory for the specified session.

Parameters

• session_id (str): The ID of the session for which to retrieve memory.
• lastn (Optional[int], optional): The number of most recent memory entries to retrieve. Defaults to None (all entries).

Returns

• List[Memory]: A list of Memory objects representing the retrieved memory entries.

Raises

• APIError: If the API response format is unexpected.
• NotFoundError: If no results were found.

memories = await client.aget_memory("3a3a3a")
for memory in memories:
    for message in memory.messages:
        print(message.to_dict())

---

search_memory(self, session_id: str, search_payload: SearchPayload, limit: Optional[int] = None) -> List[SearchResult]

Search memory for the specified session. This method is a synchronous wrapper for the asynchronous method asearch_memory.

Parameters

• session_id (str): The ID of the session for which memory should be searched.
• search_payload (SearchPayload): A SearchPayload object representing the search query.
• limit (Optional[int], optional): The maximum number of search results to return. Defaults to None (no limit).

Returns

• List[SearchResult]: A list of SearchResult objects representing the search results.

Raises

• APIError: If the API response format is unexpected.

---

asearch_memory(self, session_id: str, search_payload: SearchPayload, limit: Optional[int] = None) -> List[SearchResult]

Asynchronously search memory for the specified session.

Parameters

• session_id (str): The ID of the session for which memory should be searched.
• search_payload (SearchPayload): A SearchPayload object representing the search query.
• limit (Optional[int], optional): The maximum number of search results to return. Defaults to None (no limit).

Returns

• List[SearchResult]: A list of SearchResult objects representing the search results.

Raises

• APIError: If the API response format is unexpected.

search_payload = SearchPayload({}, "What food is served in Iceland?")
search_results = await client.asearch_memory(session_id, search_payload)
for search_result in search_results:
    # Access the 'content' field within the 'message' object.
    message_content = search_result.message
    print(message_content)

---

add_memory(self, session_id: str, memory_messages: Memory) -> str

Add memory to the specified session. This method is a synchronous wrapper for the asynchronous method aadd_memory.

Parameters

• session_id (str): The ID of the session to which memory should be added.
• memory_messages (Memory): A Memory object representing the memory messages to be added.

Returns

• str: The response text from the API.

Raises

• APIError: If the API response format is unexpected.

---

aadd_memory(self, session_id: str, memory_messages: Memory) -> str

Asynchronously add memory to the specified session.

Parameters

• session_id (str): The ID of the session to which memory should be added.
• memory_messages (Memory): A Memory object representing the memory messages to be added.

Returns

• str: The response text from the API.

Raises

• APIError: If the API response format is unexpected.

message = Message(role="user", content="who was the first man to go to space?")

memory = Memory()
memory.messages = [message]

result = await client.aadd_memory(session_id, memory)

---

delete_memory(self, session_id: str) -> str

Delete memory for the specified session. This method is a synchronous wrapper for the asynchronous method adelete_memory.

Parameters

• session_id (str): The ID of the session for which memory should be deleted.

Returns

• str: The response text from the API.

Raises

• APIError: If the API response format is unexpected.

---

adelete_memory(self, session_id: str) -> str

Asynchronously delete memory for the specified session.

Parameters

• session_id (str): The ID of the session for which memory should be deleted

result = await client.adelete_memory(session_id)

---

close(self) -> None

Asynchronously close the HTTP client.

Note: This method may be called when the ZepClient is no longer needed to release resources.

Models

Memory

Represents a memory object with messages, metadata, and other attributes.

Attributes:

• messages (Optional[List[Dict[str, Any]] | Memory]): A list of message objects, where each message contains a role and content.
• metadata (Optional[Dict[str, Any]]): A dictionary containing metadata associated with the memory.
• summary (Optional[Dict[str, Any]]): A dictionary containing a summary of the memory.
• uuid (Optional[str]): A unique identifier for the memory.
• created_at (Optional[str]): The timestamp when the memory was created.
• token_count (Optional[int]): The token count of the memory.

---

Message

Represents a message in a conversation.

Attributes:

• uuid (Optional[str]): The unique identifier of the message.
• created_at (Optional[str]): The timestamp of when the message was created.
• role (str): The role of the sender of the message (e.g., "user", "assistant").
• content (str): The content of the message.
• token_count (Optional[int]): The number of tokens in the message.

Methods:

• to_dict() -> Dict[str, Any]: Returns a dictionary representation of the message.

---

Summary

Represents a summary of a conversation.

Attributes:

• uuid (str): The unique identifier of the summary.
• created_at (str): The timestamp of when the summary was created.
• content (str): The content of the summary.
• recent_message_uuid (str): The unique identifier of the most recent message in the conversation.
• token_count (int): The number of tokens in the summary.

Methods:

• to_dict() -> Dict[str, Any]: Returns a dictionary representation of the summary.

---

SearchPayload

Represents a search payload for querying memory.

Attributes:

• meta (Dict[str, Any]): Metadata associated with the search query.
• text (str): The text of the search query.

---

SearchResult

Represents a search result from querying memory.

Attributes:

• message (Optional[Dict[str, Any]]): The message associated with the search result.
• meta (Optional[Dict[str, Any]]): Metadata associated with the search result.
• score (Optional[float]): The score of the search result.
• summary (Optional[str]): The summary of the search result.
• dist (Optional[float]): The distance metric of the search result.

---

APIError

Represents an API error.

Attributes:

• code (int): The error code associated with the API error.
• message (str): The error message associated with the API error.