Updating Bots

Updating Bots

LongTrainer allows you to update bots with new documents, links, and prompt templates after creation.

Update with New Documents

trainer.update_chatbot(
    paths=["new_data/report.pdf", "new_data/notes.md"],
    bot_id=bot_id,
)

Update with Links and Search Queries

trainer.update_chatbot(
    paths=[],
    bot_id=bot_id,
    links=["https://example.com/latest-news"],
    search_query="Current trends in AI",
)

Parameters

Parameter	Type	Description
paths	list[str]	File paths to load
bot_id	str	The bot's unique identifier
links	list[str]	Optional web URLs or YouTube links
search_query	str	Optional Wikipedia search query
documents	list	Optional pre-loaded LangChain documents
prompt_template	str	Optional new prompt template
use_unstructured	bool	Use UnstructuredLoader for files

Update Behavior

When updating a bot:

1. New documents are added to MongoDB
2. The FAISS index is updated incrementally (existing index is preserved)
3. The retriever is rebuilt with the expanded index
4. Existing chat sessions continue to work with the updated knowledge base

Update the Prompt Template

Change the prompt without adding new documents:

trainer.set_custom_prompt_template(
    bot_id,
    "You are a customer support agent. Use only the provided context. {context}"
)

Full Update Example

trainer.update_chatbot(
    paths=["reports/q4_2024.pdf"],
    bot_id=bot_id,
    links=["https://example.com/blog/update"],
    search_query="quarterly earnings reports",
    prompt_template="You are a financial analyst. {context}",
    use_unstructured=True,
)
print("Bot updated successfully.")