diff --git a/main.py b/main.py index c35d8ff..cd8a45b 100644 --- a/main.py +++ b/main.py @@ -1,20 +1,18 @@ import os import asyncio -import json from pathlib import Path -from typing import List - from langchain_openai import ChatOpenAI, OpenAIEmbeddings from langchain_chroma import Chroma from langchain_core.documents import Document -from langchain_core.messages import HumanMessage from langchain.tools import tool from deepagents import create_deep_agent from deepagents.backends import FilesystemBackend, LocalShellBackend, CompositeBackend -# --------------------------------------------------------------------------- -# 1. Настройка LLM и Embeddings (OpenRouter) -# --------------------------------------------------------------------------- +# ----------------- Configuration ----------------- +# Load API key from .env or environment variable +os.environ.setdefault("OPENAI_API_KEY", os.getenv("OPENAI_API_KEY", "")) + +# LLM via OpenRouter llm = ChatOpenAI( model="openai/gpt-oss-20b:free", base_url="https://openrouter.ai/api/v1", @@ -22,123 +20,105 @@ llm = ChatOpenAI( temperature=0.0, ) +# Embeddings for Chroma embeddings = OpenAIEmbeddings( model="text-embedding-3-small", base_url="https://openrouter.ai/api/v1", api_key=os.getenv("OPENAI_API_KEY"), ) -# --------------------------------------------------------------------------- -# 2. ChromaDB: загрузка .md файлов и поиск -# --------------------------------------------------------------------------- +# ----------------- Chroma DB ----------------- CHROMA_PATH = Path("./chroma_faq") CHROMA_COLLECTION = "faq_collection" - vector_store = Chroma( collection_name=CHROMA_COLLECTION, embedding_function=embeddings, persist_directory=str(CHROMA_PATH), ) -# Если коллекция пуста, загрузим данные из data/*.md -if not vector_store.get_collection().list_documents(): - md_files = list(Path("data").glob("*.md")) - docs: List[Document] = [] - for f in md_files: - text = f.read_text(encoding="utf-8") - docs.append(Document(page_content=text, metadata={"source": f.name})) +# Load markdown files into Chroma (idempotent) +DATA_DIR = Path("./data") +if not CHROMA_PATH.exists() or not list(CHROMA_PATH.iterdir()): + docs = [] + for md_file in DATA_DIR.glob("*.md"): + text = md_file.read_text(encoding="utf-8") + docs.append(Document(page_content=text, metadata={"source": md_file.name})) vector_store.add_documents(docs) vector_store.persist() +# ----------------- Tools ----------------- @tool -def search_course_docs(query: str, k: int = 3) -> str: +def search_course_docs(query: str) -> str: """Search the local FAQ collection for relevant passages.""" - results = vector_store.similarity_search(query, k=k) + results = vector_store.similarity_search(query, k=3) if not results: return "No relevant information found in the course materials." - return "\n\n---\n\n".join(f"{doc.metadata.get('source', 'unknown')}\n{doc.page_content}" for doc in results) + return "\n---\n".join(f"{doc.metadata.get('source', 'unknown')}\n{doc.page_content}" for doc in results) -# --------------------------------------------------------------------------- -# 3. MCP‑style tool (mocked via local JSON file) -# --------------------------------------------------------------------------- -META_JSON = Path("course_meta.json") -if not META_JSON.exists(): - # Создаём простую статическую мета‑информацию - META_JSON.write_text(json.dumps({ - "schedule": { - "Monday": "Lecture 1", - "Wednesday": "Lecture 2", - "Friday": "Lab" - }, - "instructor": "Dr. Example" - }, indent=2)) +# Mock MCP tool – static JSON data +COURSE_META = { + "schedule": "Monday 10:00-12:00, Wednesday 14:00-16:00", + "instructor": "Dr. Ivanov", + "location": "Room 101", +} @tool def fetch_course_meta(query: str) -> str: - """Return course metadata that matches the query. - The function simply looks for the query string in the keys of the JSON. + """Return course metadata matching the query keyword. + For example, query="schedule" returns the schedule string. """ - data = json.loads(META_JSON.read_text()) - for key, value in data.items(): - if query.lower() in key.lower(): - return json.dumps({key: value}, indent=2) - return "No metadata found for the given query." + key = query.lower().strip() + return COURSE_META.get(key, f"No metadata found for '{query}'.") -# --------------------------------------------------------------------------- -# 4. DeepAgent с маршрутизацией -# --------------------------------------------------------------------------- +# ----------------- Backend ----------------- backend = CompositeBackend([ LocalShellBackend(workspace_dir="./workspace"), FilesystemBackend(), ]) +# ----------------- Agent ----------------- agent = create_deep_agent( model=llm, tools=[search_course_docs, fetch_course_meta], backend=backend, system_prompt=( - "You are a helpful FAQ bot for a course.\n" - "If the question is about course content, use the search_course_docs tool.\n" - "If the question is about schedule, instructor, or other metadata, use fetch_course_meta.\n" + "You are a helpful FAQ assistant for the course.\n" + "When answering a question, first decide whether the answer comes from the course materials (use search_course_docs)\n" + "or from course metadata (use fetch_course_meta).\n" "Do not call both tools unless absolutely necessary.\n" - "In your final answer, prepend 'source: chroma' or 'source: mcp_meta' to indicate where the answer came from." + "In your final answer, prepend 'source: chroma' or 'source: mcp_meta' to indicate the origin." ), ) -# --------------------------------------------------------------------------- -# 5. CLI -# --------------------------------------------------------------------------- +# ----------------- CLI ----------------- PRESET_QUESTIONS = [ - "What topics are covered in Lecture 1?", # should hit chroma - "Who is the instructor for this course?", # should hit mcp_meta - "Explain the concept of polymorphism." + "What topics are covered in the first lecture?", # should hit chroma + "Who is the instructor for this course?", # should hit mcp_meta + "When is the next class?", # should hit chroma (or meta if schedule) ] -async def run_agent(question: str) -> str: +async def run_question(question: str): result = await agent.ainvoke( - {"messages": [HumanMessage(content=question)]}, + {"messages": ["HumanMessage(content=\"{}\")".format(question)]}, {"configurable": {"thread_id": "session-1"}}, ) - return result["messages"][-1].content + # The agent returns a dict with 'messages'; extract last content + content = result["messages"][-1].content + print(f"\nQ: {question}\nA: {content}\n") + +async def interactive(): + print("Enter a question (or 'exit' to quit):") + while True: + q = input("> ") + if q.lower() in {"exit", "quit"}: + break + await run_question(q) async def main(): - print("\n--- FAQ Bot Demo ---\n") - for i, q in enumerate(PRESET_QUESTIONS, 1): - print(f"Q{i}: {q}") - ans = await run_agent(q) - print(f"A{i}: {ans}\n") - - print("Enter your own question (or press Ctrl+C to exit):") - while True: - try: - user_q = input("> ") - if not user_q.strip(): - continue - ans = await run_agent(user_q) - print(ans) - except KeyboardInterrupt: - print("\nExiting.") - break + print("Running preset questions...") + for q in PRESET_QUESTIONS: + await run_question(q) + await interactive() if __name__ == "__main__": asyncio.run(main())