diff --git a/main.py b/main.py index 0a7335a..1edd68c 100644 --- a/main.py +++ b/main.py @@ -1,5 +1,10 @@ -import asyncio, os +import os +import asyncio +import json from pathlib import Path +from typing import List + +import httpx from langchain_openai import ChatOpenAI from langchain_core.messages import HumanMessage from langchain.tools import tool @@ -7,10 +12,62 @@ from deepagents import create_deep_agent from deepagents.backends import FilesystemBackend, LocalShellBackend, CompositeBackend from langchain_chroma import Chroma from langchain_ollama import OllamaEmbeddings -import httpx -import json -# ---------- LLM ---------- +# --------------------- +# 1. Chroma DB helpers +# --------------------- +CHROMA_PATH = Path("./chroma_faq") +DATA_DIR = Path("./data") + + +def load_faq_to_chroma() -> None: + """Load all .md files from data/ into a persistent Chroma store.""" + if CHROMA_PATH.exists(): + # already loaded + return + CHROMA_PATH.mkdir(parents=True, exist_ok=True) + embeddings = OllamaEmbeddings(model="nomic-embed-text") + db = Chroma.from_folder(str(DATA_DIR), embedding=embeddings, persist_directory=str(CHROMA_PATH)) + db.persist() + +@tool +def search_course_docs(query: str, k: int = 3) -> str: + """Search local course FAQ documents in Chroma DB.""" + load_faq_to_chroma() + embeddings = OllamaEmbeddings(model="nomic-embed-text") + db = Chroma(persist_directory=str(CHROMA_PATH), embedding=embeddings) + docs = db.similarity_search(query, k=k) + if not docs: + return "No relevant FAQ found." + return "\n\n---\n\n".join(doc.page_content for doc in docs) + +# --------------------- +# 2. MCP‑style tool +# --------------------- +# For demo we use a local JSON file served by python -m http.server +# The file is located at ./meta/course_meta.json + +META_URL = "http://localhost:8000/course_meta.json" + +@tool +def fetch_course_meta(query: str) -> str: + """Fetch course metadata (e.g., schedule) from a mock MCP server.""" + try: + response = httpx.get(META_URL, timeout=5.0) + response.raise_for_status() + data = response.json() + except Exception as e: + return f"Error fetching metadata: {e}" + # Simple keyword search in the JSON + results: List[str] = [] + for key, value in data.items(): + if query.lower() in key.lower() or query.lower() in str(value).lower(): + results.append(f"{key}: {value}") + return "\n".join(results) if results else "No metadata matches your query." + +# --------------------- +# 3. Agent setup +# --------------------- llm = ChatOpenAI( model="openai/gpt-oss-20b:free", base_url="https://openrouter.ai/api/v1", @@ -18,97 +75,56 @@ llm = ChatOpenAI( temperature=0.0, ) -# ---------- Backend ---------- backend = CompositeBackend([ LocalShellBackend(workspace_dir="./workspace"), FilesystemBackend(), ]) -# ---------- Chroma DB ---------- -CHROMA_PATH = Path("./chroma_faq") -CHROMA_PATH.mkdir(exist_ok=True) +SYSTEM_PROMPT = ( + "You are a helpful FAQ assistant for the course.\n" + "When a user asks a question, first decide whether the answer is best found in the local FAQ documents or in the course metadata.\n" + "If the answer is in the FAQ, use the tool `search_course_docs`.\n" + "If the answer requires schedule or other metadata, use the tool `fetch_course_meta`.\n" + "Do not call both tools unless absolutely necessary.\n" + "In your final response, prepend the source: `source: chroma` or `source: mcp_meta`." +) -embeddings = OllamaEmbeddings(model="nomic-embed-text") - -# Load or create vector store -if CHROMA_PATH.exists() and any(CHROMA_PATH.iterdir()): - chroma = Chroma(persist_directory=str(CHROMA_PATH), embedding_function=embeddings) -else: - # Load markdown files - docs = [] - for md_file in Path("data").glob("*.md"): - text = md_file.read_text(encoding="utf-8") - docs.append(text) - chroma = Chroma.from_texts(docs, embedding=embeddings, persist_directory=str(CHROMA_PATH)) - chroma.persist() - -@tool -def search_course_docs(query: str, k: int = 3) -> str: - """Search local course documents in Chroma.""" - results = chroma.similarity_search(query, k=k) - return "\n---\n".join(doc.page_content for doc in results) if results else "No relevant docs found." - -# ---------- MCP‑style tool ---------- -# For demo we use a static JSON file. In production this would be an HTTP call. -META_JSON = Path("meta.json") -if not META_JSON.exists(): - # Create a simple mock meta file - META_JSON.write_text(json.dumps({ - "schedule": "Mon 10-12, Wed 14-16, Fri 9-11", - "instructor": "Dr. Smith", - "location": "Room 101" - })) - -@tool -def fetch_course_meta(query: str) -> str: - """Return course metadata matching the query keyword.""" - data = json.loads(META_JSON.read_text()) - # Simple keyword search in values - for key, value in data.items(): - if query.lower() in key.lower() or query.lower() in str(value).lower(): - return f"{key}: {value}" - return "No metadata found for the query." - -# ---------- 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 the course.\n" - "If the question is about course content, use search_course_docs.\n" - "If the question is about schedule, instructor, or location, use fetch_course_meta.\n" - "Do not call both tools unless necessary.\n" - "In your answer, prefix the source with 'source: chroma' or 'source: mcp_meta'." - ), + system_prompt=SYSTEM_PROMPT, ) -# ---------- CLI ---------- +# --------------------- +# 4. CLI +# --------------------- PRESET_QUESTIONS = [ - "What is the main topic of the first lecture?", - "How can I access the lecture slides?", - "When is the next class?" + "What is the deadline for the final project?", # FAQ + "When does the next lecture on deep learning start?", # metadata + "Explain the concept of attention mechanism.", # FAQ ] -async def run_cli(): - print("--- FAQ Bot CLI ---") +async def run_agent(question: str) -> str: + result = await agent.ainvoke( + {"messages": [HumanMessage(content=question)]}, + {"configurable": {"thread_id": "session-1"}}, + ) + return result["messages"][-1].content + +async def main(): + print("--- FAQ Bot Demo ---\n") for i, q in enumerate(PRESET_QUESTIONS, 1): - print(f"\nPreset {i}: {q}") - result = await agent.ainvoke( - {"messages": [HumanMessage(content=q)]}, - {"configurable": {"thread_id": f"preset-{i}"}}, - ) - print(result["messages"][-1].content) - print("\nEnter your own question (or 'exit'): ") + print(f"Q{i}: {q}") + ans = await run_agent(q) + print(f"A{i}: {ans}\n") + print("Enter your own question (or 'exit' to quit):") while True: user_q = input("> ") if user_q.lower() in {"exit", "quit"}: break - result = await agent.ainvoke( - {"messages": [HumanMessage(content=user_q)]}, - {"configurable": {"thread_id": "interactive"}}, - ) - print(result["messages"][-1].content) + ans = await run_agent(user_q) + print(ans) if __name__ == "__main__": - asyncio.run(run_cli()) + asyncio.run(main())