Overview
LangChatEngine is the internal engine that powers LangChat. In most cases you don’t interact with it directly — use the LangChat SDK class instead.
Access it via lc.engine when you need low-level control.
from langchat.core.engine import LangChatEngine
Constructor
LangChatEngine(
*,
llm,
vector_db,
db,
reranker=None,
prompt_template=None,
standalone_question_prompt=None,
verbose=False,
max_chat_history=20,
)
LangChat. All are keyword-only.
Methods
async chat()
Process a chat query. Returns a raw dict (not a ChatResponse).
async def chat(
self,
query: str,
user_id: str,
platform: str = "default",
standalone_question: str | None = None,
) -> dict
If provided, skips the standalone question generation step and uses this value directly. Useful for pre-processing.
{
"response": str, # AI response text
"user_id": str, # echoed user_id
"timestamp": str, # ISO 8601 UTC
"status": "success"|"error",
"response_time": float, # seconds
"error": str | None, # error message on failure
}
get_session()
Get or create a UserSession.
def get_session(
self,
user_id: str,
platform: str = "default",
) -> UserSession
self.sessions dict keyed by "{user_id}_{platform}".
Attributes
| Attribute | Type | Description |
|---|
llm | Any | LLM provider instance |
vector_adapter | Any | Vector DB provider instance |
history_store | Any | Database provider instance |
reranker_adapter | Any | Reranker instance |
sessions | dict[str, UserSession] | In-memory session cache |
id_manager | IDManager | Database ID sequence manager |
prompt_template | str | System prompt template |
verbose | bool | Verbose logging flag |
max_chat_history | int | History window size |
Advanced: skip standalone question generation
If you’re building a pipeline that pre-processes questions, you can skip the standalone question step:
# Pre-process the question yourself
standalone = f"Question about Acme Corp: {user_query}"
result = await lc.engine.chat(
query=user_query,
user_id="alice",
standalone_question=standalone,
)
API server mode
When running as an API server, the engine suppresses Rich console output (the formatted response panel):
from langchat.core.engine import set_api_server_mode
set_api_server_mode(True) # called automatically by create_app()
