Migration 1.0.0 → 1.2.0

Migration Guide: 1.0.0 → 1.2.1

LongTrainer 1.2.1 is a backward-compatible feature release. No breaking changes — all existing code continues to work as-is.

Upgrade

pip install --upgrade longtrainer

What's New (No Migration Required)

Dynamic Tool Injection

You can now pass tools as strings instead of importing and instantiating them manually:

# 1.0.0 — manual tool registration
from longtrainer.tools import web_search
trainer.add_tool(web_search, bot_id)
trainer.create_bot(bot_id, agent_mode=True)

# 1.2.1 — zero-code string injection (ALSO works)
trainer.create_bot(bot_id, agent_mode=True, tools=["wikipedia", "arxiv", "tavily_search_results_json"])

Tool selections are automatically saved to MongoDB and restored on load_bot().

Enterprise Vector Databases

You can now use any of 9 vector store providers instead of only FAISS:

# 1.0.0 — FAISS only (still the default)
trainer = LongTrainer()

# 1.2.1 — choose your provider
trainer = LongTrainer(
    vector_store_provider="qdrant",  # or pinecone, chroma, pgvector, mongodb, milvus, weaviate, elasticsearch
    vector_store_kwargs={"url": "http://localhost:6333"}
)

Enterprise Document Loaders

New helper methods for loading from cloud and enterprise platforms:

Method	Source
add_document_from_directory()	Local folders with glob patterns
add_document_from_json()	JSON files with jq schema
add_document_from_confluence()	Atlassian Confluence
add_document_from_github()	GitHub repositories
add_document_from_aws_s3()	AWS S3 buckets
add_document_from_google_drive()	Google Drive folders
add_document_from_dynamic_loader()	Any LangChain loader by class name

Dynamic Model Factory

New providers supported out of the box:

trainer = LongTrainer(
    llm_provider="anthropic",        # or google, bedrock, huggingface, groq, together, ollama
    default_llm="claude-3-5-sonnet-20241022",
)

CLI & API

# New --tools flag
longtrainer bot create --agent --tools "wikipedia,arxiv"

// API: POST /bots/{bot_id}/build
{ "agent_mode": true, "tools": ["wikipedia", "arxiv"] }

Dependency Changes (1.0.0 → 1.2.1)

New Optional Dependency	Used For
langchain_community	Dynamic tools & loaders
langchain_experimental	Python REPL tool
Provider packages (e.g. langchain_pinecone)	Enterprise vector stores

---

Migration Guide: 0.3.4 → 1.0.0

LongTrainer 1.0.0 is a major rewrite with significant improvements and some breaking changes.

Upgrade

pip install --upgrade longtrainer

For agent mode support:

pip install --upgrade longtrainer[agent]

Breaking Changes

Area	0.3.4	1.0.0
Internal chain	ConversationalRetrievalChain	LCEL chain (RAGBot) or LangGraph agent (AgentBot)
Packaging	requirements.txt + setup.py	pyproject.toml (compatible with pip, UV, and Poetry)
Memory	langchain.memory.ConversationTokenBufferMemory	langchain_core.chat_history.InMemoryChatMessageHistory
Response format	get_response() returns (answer, sources, web_sources)	Returns (answer, web_sources)
LLM default	gpt-4-turbo	gpt-4o-2024-08-06

What Stays the Same

The core API surface is unchanged:

# These all work exactly as before
trainer = LongTrainer(mongo_endpoint="mongodb://localhost:27017/")
bot_id = trainer.initialize_bot_id()
trainer.add_document_from_path("data.pdf", bot_id)
trainer.create_bot(bot_id)
chat_id = trainer.new_chat(bot_id)
trainer.load_bot(bot_id)
trainer.update_chatbot(paths, bot_id)
trainer.delete_chatbot(bot_id)
trainer.list_chats(bot_id)
trainer.get_chat_by_id(chat_id)
trainer.train_chats(bot_id)

Code Changes Required

Response Unpacking

# 0.3.4
answer, sources, web_sources = trainer.get_response(query, bot_id, chat_id)

# 1.0.0
answer, web_sources = trainer.get_response(query, bot_id, chat_id)

Imports

# 0.3.4 — these still work
from longtrainer.trainer import LongTrainer

# 1.0.0 — new top-level imports available
from longtrainer import LongTrainer, ToolRegistry, web_search

New Features in 1.0.0

These are new capabilities that don't require migration — just start using them:

Streaming

# Sync streaming
for chunk in trainer.get_response(query, bot_id, chat_id, stream=True):
    print(chunk, end="", flush=True)

# Async streaming
async for chunk in trainer.aget_response(query, bot_id, chat_id):
    print(chunk, end="", flush=True)

Agent Mode with Tools

from longtrainer.tools import web_search
from langchain_core.tools import tool

@tool
def calculate(expression: str) -> str:
    """Evaluate a math expression."""
    return str(eval(expression))

trainer.add_tool(web_search, bot_id)
trainer.add_tool(calculate, bot_id)
trainer.create_bot(bot_id, agent_mode=True)

Per-Bot Customization

trainer.create_bot(
    bot_id,
    llm=ChatOpenAI(model="gpt-4o-mini"),
    embedding_model=OpenAIEmbeddings(model="text-embedding-3-small"),
    num_k=5,
    prompt_template="Custom prompt. {context}",
)

Dependency Changes

0.3.4 Dependency	1.0.0 Replacement
langchain (monolithic)	langchain_core, langchain_community, langchain_openai
setup.py	pyproject.toml
requirements.txt	pyproject.toml [project.dependencies]
N/A	langgraph (optional, for agent mode)