feat: ChatOllama(llama3) ReAct agent + CLI

main.py

@@ -1,105 +1,76 @@
-import os
+"""RAG-агент с ChromaDB (Ollama) и веб-поиском (Tavily). CLI."""
-import asyncio
+from __future__ import annotations
-from pathlib import Path
+
-from dotenv import load_dotenv
+from pathlib import Path
-from langchain_openai import ChatOpenAI, OpenAIEmbeddings
+
-from langchain_chroma import Chroma
+from dotenv import load_dotenv
-from langchain_text_splitters import RecursiveCharacterTextSplitter
+from langchain_ollama import ChatOllama
-from langchain_core.documents import Document
+from langchain.agents import AgentExecutor, create_react_agent
-from langchain.tools import tool
+from langchain import hub
-from deepagents import create_deep_agent
+
-from deepagents.backends import FilesystemBackend, LocalShellBackend, CompositeBackend
+from vectorstore import create_vectorstore, load_documents
+from tools import init_tools, search_local_kb, web_search
 
-# Load environment variables
 load_dotenv()
 
-# ---------- LLM and Embeddings ----------
+# ── LLM: Ollama llama3 ────────────────────────────────────────────────────
-llm = ChatOpenAI(
+llm = ChatOllama(model="llama3", temperature=0)
-    model="openai/gpt-oss-20b:free",
+
-    base_url="https://openrouter.ai/api/v1",
+# ── Векторное хранилище ───────────────────────────────────────────────────
-    api_key=os.getenv("OPENAI_API_KEY"),
+CHROMA_DIR = "./chroma_db"
-    temperature=0.0,
+vectorstore = create_vectorstore(CHROMA_DIR)
+init_tools(vectorstore)
+
+# Автозагрузка документов из documents/ (если есть и БД пустая)
+docs_dir = Path("documents")
+if docs_dir.exists() and not any(Path(CHROMA_DIR).rglob("*.sqlite3")):
+    count = load_documents(str(docs_dir), vectorstore)
+    print(f"[init] Загружено {count} чанков из {docs_dir}/")
+
+# ── ReAct-агент ───────────────────────────────────────────────────────────
+tools = [search_local_kb, web_search]
+prompt = hub.pull("hwchase17/react")
+agent = create_react_agent(llm, tools, prompt)
+agent_executor = AgentExecutor(
+    agent=agent,
+    tools=tools,
+    verbose=True,
+    max_iterations=10,
+    handle_parsing_errors=True,
 )
 
-embeddings = OpenAIEmbeddings(
+ROUTING_HINT = (
-    model="text-embedding-3-small",
+    "Для вопросов по локальным документам/конспектам используй search_local_kb. "
-    base_url="https://openrouter.ai/api/v1",
+    "Для актуальных новостей и фактов из интернета — web_search. "
-    api_key=os.getenv("OPENAI_API_KEY"),
+    "В ответе всегда укажи источник: chromadb или tavily."
 )
 
-# ---------- Vector Store ----------
-CHROMA_DIR = Path("./chroma_db")
-vector_store = Chroma(
-    collection_name="knowledge",
-    embedding_function=embeddings,
-    persist_directory=str(CHROMA_DIR),
-)
 
-# ---------- Document Loader ----------
+def ask(question: str) -> str:
+    """Задать вопрос агенту."""
+    result = agent_executor.invoke({"input": f"{ROUTING_HINT}
 
-def load_documents(directory: str, vectorstore: Chroma):
+Вопрос: {question}"})
-    splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
+    return result.get("output", str(result))
-    docs = []
-    for file_path in Path(directory).glob("**/*.*"):
-        if file_path.suffix.lower() not in {".txt", ".md"}:
-            continue
-        text = file_path.read_text(encoding="utf-8")
-        chunks = splitter.split_text(text)
-        docs.extend([Document(page_content=c, metadata={"source": str(file_path)}) for c in chunks])
-    if docs:
-        vectorstore.add_documents(docs)
-        vectorstore.persist()
 
-# Load initial documents if collection is empty
-if not CHROMA_DIR.exists() or not list(CHROMA_DIR.iterdir()):
-    load_documents("documents", vector_store)
-
-# ---------- Tools ----------
-@tool
-def search_local_kb(query: str, top_k: int = 3) -> str:
-    """Semantic search in the local knowledge base."""
-    docs = vector_store.similarity_search(query, k=top_k)
-    if not docs:
-        return "No relevant information found in local knowledge base."
-    return "\n---\n".join([f"{d.metadata.get('source', 'unknown')}\n{d.page_content}" for d in docs])
-
-@tool
-def web_search(query: str) -> str:
-    """Web search using Tavily."""
-    from langchain_tavily import TavilySearchResults
-    tavily = TavilySearchResults()
-    results = tavily.run(query)
-    return "\n---\n".join([f"{r['title']}\n{r['content']}" for r in results])
-
-# ---------- Backend ----------
-backend = CompositeBackend([
-    LocalShellBackend(workspace_dir="./workspace"),
-    FilesystemBackend(),
-])
-
-# ---------- Agent ----------
-agent = create_deep_agent(
-    model=llm,
-    tools=[search_local_kb, web_search],
-    backend=backend,
-    system_prompt="You are a helpful assistant that can search both a local knowledge base and the web.\nWhen answering, always indicate the source: either 'chromadb' or 'tavily'.\nUse the appropriate tool based on the query context.",
-)
-
-# ---------- CLI ----------
-async def main():
-    print("RAG Agent with ChromaDB and Tavily. Type 'exit' to quit.")
-    while True:
-        user_input = input("\nЗапрос: ")
-        if user_input.lower() in {"exit", "quit"}:
-            print("Goodbye!")
-            break
-        result = await agent.ainvoke(
-            {"messages": [{"role": "user", "content": user_input}]},
-            {"configurable": {"thread_id": "session-1"}},
-        )
-        # Extract last message content
-        content = result["messages"][-1]["content"]
-        print(content)
 
 if __name__ == "__main__":
-    asyncio.run(main())
+    print("RAG-агент: Ollama llama3 + nomic-embed-text + ChromaDB + Tavily")
+    print("Введите 'exit' для выхода.")
+    print("-" * 60)
+    while True:
+        try:
+            user_input = input("
+Запрос: ").strip()
+        except (EOFError, KeyboardInterrupt):
+            print("
+Завершение.")
+            break
+        if user_input.lower() in ("exit", "quit", "выход"):
+            print("До свидания!")
+            break
+        if not user_input:
+            continue
+        answer = ask(user_input)
+        print(f"
+{answer}")