95 lines
3.7 KiB
Python
95 lines
3.7 KiB
Python
from langchain_ollama import OllamaEmbeddings
|
|
from langchain_qdrant import QdrantVectorStore
|
|
from qdrant_client import QdrantClient
|
|
from qdrant_client.http.models import Distance, VectorParams
|
|
from langchain_core.documents import Document
|
|
from langchain_text_splitters import RecursiveCharacterTextSplitter
|
|
from langchain.tools import tool
|
|
from langchain.agents import create_agent
|
|
from langchain_openai import ChatOpenAI
|
|
from pydantic import SecretStr
|
|
|
|
# ---------- LLM and embeddings ----------
|
|
llm = ChatOpenAI(
|
|
model="openai/gpt-oss-20b",
|
|
base_url='https://platform.brojs.ru/jrnl-bh/api/inference/v1',
|
|
api_key=SecretStr("jrnl_30283ab953615cbb6846ff9940a1eedce0b76d7b2f59a2394f29e74643e6a90d"),
|
|
temperature=0.7,
|
|
)
|
|
|
|
embeddings = OllamaEmbeddings(model_name="nomic-embed-text")
|
|
|
|
# ---------- Qdrant client ----------
|
|
client = QdrantClient(":memory:")
|
|
client.create_collection(
|
|
collection_name="knowledge_base",
|
|
vectors_config=VectorParams(size=embeddings.embedding_size, distance=Distance.COSINE),
|
|
)
|
|
vector_store = QdrantVectorStore(client=client, collection_name="knowledge_base", embedding=embeddings)
|
|
|
|
# ---------- Text splitter ----------
|
|
splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50)
|
|
|
|
# ---------- Tools ----------
|
|
@tool
|
|
def add_to_knowledge_base(content: str, title: str) -> str:
|
|
"""Add a document to the knowledge base."""
|
|
docs = [Document(page_content=c, metadata={"title": title}) for c in splitter.split_text(content)]
|
|
vector_store.add_documents(docs)
|
|
return f"Added {len(docs)} chunks titled '{title}'."
|
|
|
|
@tool
|
|
def search_knowledge_base(query: str, max_results: int = 5) -> str:
|
|
"""Search the knowledge base for relevant information."""
|
|
results = vector_store.similarity_search_with_score(query, k=max_results)
|
|
if not results:
|
|
return "No relevant documents found."
|
|
reply = ""
|
|
for i, (doc, score) in enumerate(results, start=1):
|
|
reply += f"{i}. ({score:.2f}) {doc.metadata.get('title', 'Untitled')}: {doc.page_content[:200]}...\n"
|
|
return reply.strip()
|
|
|
|
# ---------- Agent ----------
|
|
system_prompt = (
|
|
"You are an assistant that can search and add information to a knowledge base. "
|
|
"Use the tools `add_to_knowledge_base` and `search_knowledge_base` as needed."
|
|
)
|
|
agent = create_agent(model=llm, tools=[add_to_knowledge_base, search_knowledge_base], system_prompt=system_prompt)
|
|
|
|
# ---------- CLI ----------
|
|
def main():
|
|
print("RAG Agent CLI. Commands: /add <title> <content>, /search <query>, /quit")
|
|
while True:
|
|
try:
|
|
inp = input("> ").strip()
|
|
except EOFError:
|
|
break
|
|
if not inp:
|
|
continue
|
|
if inp.lower() in ("/quit", "exit"):
|
|
print("Bye!")
|
|
break
|
|
if inp.startswith("/add"):
|
|
parts = inp.split(maxsplit=2)
|
|
if len(parts) < 3:
|
|
print("Usage: /add <title> <content>")
|
|
continue
|
|
title, content = parts[1], parts[2]
|
|
res_msg = agent.invoke({"messages": [{"role": "human", "content": f"/add {title} {content}"}]})
|
|
for msg in res_msg["messages"]:
|
|
if hasattr(msg, "tool_calls"):
|
|
print(msg.tool_calls[0]["output"])
|
|
elif inp.startswith("/search"):
|
|
query = inp[len("/search"):].strip()
|
|
if not query:
|
|
print("Usage: /search <query>")
|
|
continue
|
|
res_msg = agent.invoke({"messages": [{"role": "human", "content": f"/search {query}"}]})
|
|
for msg in res_msg["messages"]:
|
|
if hasattr(msg, "tool_calls"):
|
|
print(msg.tool_calls[0]["output"])
|
|
else:
|
|
print("Unknown command. Use /add, /search, or /quit.")
|
|
|
|
if __name__ == "__main__":
|
|
main() |