Creating an Instance

Creating an Instance of LongTrainer

To start using LongTrainer, initialize an instance of the LongTrainer class. This instance manages all bots, chat sessions, document ingestion, and MongoDB persistence.

Basic Initialization

from longtrainer.trainer import LongTrainer

trainer = LongTrainer()

Constructor Parameters

Parameter	Type	Default	Description
mongo_endpoint	str	"mongodb://localhost:27017/"	MongoDB connection string
llm	BaseChatModel	ChatOpenAI(model="gpt-4o-2024-08-06")	Default language model for all bots
embedding_model	Embeddings	OpenAIEmbeddings()	Default embedding model for document vectorization
prompt_template	str	Built-in prompt	System prompt template (must include {context} placeholder)
max_token_limit	int	32000	Token buffer limit for conversation memory
num_k	int	3	Number of documents to retrieve per query
chunk_size	int	2048	Text splitter chunk size
chunk_overlap	int	200	Text splitter overlap between chunks
ensemble	bool	False	Enable multi-query ensemble retrieval for better recall
encrypt_chats	bool	False	Enable Fernet encryption for stored chats
encryption_key	bytes	Auto-generated	Custom Fernet encryption key

Custom Configuration

from longtrainer.trainer import LongTrainer
from langchain_openai import ChatOpenAI

trainer = LongTrainer(
    mongo_endpoint="mongodb://custom-host:27017/",
    llm=ChatOpenAI(model="gpt-4o-mini", temperature=0.3),
    max_token_limit=16000,
    num_k=5,
    chunk_size=1024,
    chunk_overlap=100,
    ensemble=True,
    encrypt_chats=True,
)

With Encryption

When encrypt_chats=True, all chat history stored in MongoDB is encrypted using Fernet symmetric encryption. You can provide your own key or let LongTrainer generate one:

from cryptography.fernet import Fernet

key = Fernet.generate_key()

trainer = LongTrainer(
    encrypt_chats=True,
    encryption_key=key,
)

# Save the key — you'll need it to decrypt chats later
print(f"Encryption key: {key.decode()}")

Warning

If you lose the encryption key, stored chats cannot be recovered.

Per-Bot Overrides

While the constructor sets global defaults, each bot can override LLM, embeddings, and retrieval settings individually via create_bot(). See Creating and Using a Bot for details.