import os
import asyncio
from langchain_openai import ChatOpenAI
from langchain.tools import tool
from deepagents import create_deep_agent
from deepagents.backends import FilesystemBackend, LocalShellBackend, CompositeBackend
from langchain_core.messages import HumanMessage
from langchain_core.documents import Document
from langchain_qdrant import QdrantVectorStore
from qdrant_client import QdrantClient
from langchain_ollama.embeddings import OllamaEmbeddings
from langchain_text_splitter import RecursiveCharacterTextSplitter

# DESIGN DECISION: Use OllamaEmbeddings for local embeddings to avoid external API calls.
# NECESSITY: Assignment requires local LLM and embeddings via Ollama.
# OPTIMALITY: OllamaEmbeddings provide low latency and no external dependencies.
# ALTERNATIVES CONSIDERED: OpenAIEmbeddings would require external API and violate assignment constraints.

embeddings = OllamaEmbeddings(model="nomic-embed-text")

# DESIGN DECISION: Initialize QdrantVectorStore with local Qdrant client and OllamaEmbeddings.
# NECESSITY: RAG requires a vector store; Qdrant is specified in the stack.
# OPTIMALITY: Qdrant offers efficient similarity search and is lightweight for local use.
# ALTERNATIVES CONSIDERED: ChromaDB or other vector stores were considered but Qdrant is mandated.

qdrant_client = QdrantClient(host="localhost", port=6333)
vector_store = QdrantVectorStore(
    client=qdrant_client,
    collection_name="knowledge",
    embedding_function=embeddings
)

# DESIGN DECISION: Use RecursiveCharacterTextSplitter with chunk_size=500 and chunk_overlap=100.
# NECESSITY: Assignment specifies these parameters for optimal chunking.
# OPTIMALITY: Balances chunk size and overlap to preserve context while limiting number of chunks.
# ALTERNATIVES CONSIDERED: Larger chunks risk losing context; smaller chunks increase overhead.

splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=100)

@tool
def search_knowledge_base(query: str, max_results: int = 3) -> str:
    """Search the knowledge base for relevant information."""
    docs = vector_store.similarity_search(query, k=max_results)
    return "\n".join(d.page_content for d in docs) if docs else "No results."

@tool
def add_to_knowledge_base(content: str, title: str = "doc") -> str:
    """Add content to the knowledge base."""
    chunks = splitter.split_text(content)
    docs = [Document(page_content=chunk, metadata={"title": title}) for chunk in chunks]
    vector_store.add_documents(docs)
    return f"Added {len(docs)} chunks for {title}."

# DESIGN DECISION: Use OpenRouter via langchain_openai for LLM.
# NECESSITY: Assignment mandates OpenRouter usage.
# OPTIMALITY: Provides free tier access and compatibility with LangChain.
# ALTERNATIVES CONSIDERED: Local LLMs would require GPU resources.

llm = ChatOpenAI(
    model="openai/gpt-oss-20b:free",
    base_url="https://openrouter.ai/api/v1",
    api_key=os.getenv("OPENAI_API_KEY"),
    temperature=0.0,
)

backend = CompositeBackend([
    LocalShellBackend(workspace_dir="./workspace"),
    FilesystemBackend(),
])

agent = create_deep_agent(
    model=llm,
    tools=[search_knowledge_base, add_to_knowledge_base],
    backend=backend,
    system_prompt="You are a helpful agent with a knowledge base. Use the tools search_knowledge_base and add_to_knowledge_base as needed.",
)

async def main():
    print("RAG Agent CLI. Commands: /add <title> <content>, /search <query>, /quit")
    while True:
        user_input = input(">> ").strip()
        if not user_input:
            continue
        if user_input.lower() == "/quit":
            print("Goodbye!")
            break
        if user_input.lower().startswith("/add"):
            parts = user_input.split(maxsplit=2)
            if len(parts) < 3:
                print("Usage: /add <title> <content>")
                continue
            title, content = parts[1], parts[2]
            message = f"Add document titled {title} with content: {content}"
        elif user_input.lower().startswith("/search"):
            parts = user_input.split(maxsplit=1)
            if len(parts) < 2:
                print("Usage: /search <query>")
                continue
            query = parts[1]
            message = f"Search for {query}"
        else:
            message = user_input

        result = await agent.ainvoke(
            {"messages": [HumanMessage(content=message)]},
            {"configurable": {"thread_id": "session-1"}},
        )
        print(result["messages"][-1].content)

if __name__ == "__main__":
    asyncio.run(main())