⚡️ Added history

api.py

@@ -4,6 +4,9 @@
 import uvicorn
 
 
+from langchain.load.dump import dumps
+from langchain.load.load import loads
+
 from interface.response_models import ResponseModel, CTA, CTAType
 from interface.request_models import RequestModel
 from src.context import get_context
@@ -49,16 +52,23 @@ async def redirect():
     response_model=ResponseModel
 )
 def answer_a_question(query: RequestModel) -> ResponseModel:
+    question = query.question
+    if query.history is not None:
+        history = loads(query.history)
+    else:
+        history = []
+
     context = get_context(query.question)
     prompt = assemble_prompt(query.question, context)
-    answer = generate_answer(prompt)
+    answer, history = generate_answer(prompt, history)
     refs = [c.metadata["title"] for c in context]
     return ResponseModel(
         status="ok",
         msg="Successfully generated answer",
         answer=answer,
         cta=list(generate_cta(context)),
-        refs=refs
+        refs=refs,
+        history=dumps(history)
     )
 
 

interface/request_models.py

@@ -1,8 +1,11 @@
-from typing import Mapping, Optional
+from typing import Mapping, Optional, Union
 from pydantic import BaseModel
 
+from langchain_core.messages import HumanMessage, SystemMessage
+
 
 class RequestModel(BaseModel):
     question: str
     context: Optional[Mapping] = None
-    assistant_id: Optional[str] = None
+    assistant_id: Optional[str] = None
+    history: Optional[str] = None

interface/response_models.py

@@ -25,3 +25,4 @@ class ResponseModel(BaseModel):
     status: str
     msg: str
     cta: list[CTA]
+    history: str

src/generate_with_azure.py

@@ -1,5 +1,5 @@
 import os
-from typing import Optional
+from typing import Optional, Union, Tuple
 
 from langchain_core.messages import HumanMessage, SystemMessage
 from langchain_core.documents.base import Document
@@ -22,15 +22,20 @@
 )
 
 
-def generate_answer(prompt: str, system_prompt: str = SYSTEM_PROMPT) -> str:
-    message = SystemMessage(
-        content=system_prompt
-    )
-    message = HumanMessage(
-        content=prompt
-    )
-    result = MODEL.invoke([message])
-    return result.content
+def generate_answer(
+    prompt: str,
+    history: list[Optional[Union[SystemMessage, HumanMessage]]],
+    system_prompt: str = SYSTEM_PROMPT
+) -> Tuple[str, list[Union[SystemMessage, HumanMessage]]]:
+    if not history:
+        history.append(
+            SystemMessage(
+                content=system_prompt
+            )
+        )
+    history.append(HumanMessage(content=prompt))
+    result = MODEL.invoke(history)
+    return result.content, history
 
 
 if __name__ == "__main__":

src/tools.py

@@ -1,6 +1,6 @@
-from typing import Optional, Generator
+from typing import Optional, Generator, Mapping, Union
 from langchain_core.documents.base import Document
-
+from langchain_core.messages import HumanMessage, SystemMessage
 from interface.response_models import CTA, CTAType
 from urllib import parse
 
@@ -13,4 +13,3 @@ def generate_cta(context: list[Optional[Document]]) -> Generator[CTA, None, None
             text=c.metadata["title"],
             payload=f"https://brbuddy-api-service-volume.brdata-dev.de/share/URV7wHnY/{norm_name}"
         )
-