diff --git a/agent.py b/agent.py
new file mode 100644
index 0000000..47dfe77
--- /dev/null
+++ b/agent.py
@@ -0,0 +1,136 @@
+import os
+import json
+import asyncio
+from typing import TypedDict, List, Dict, Any
+
+import questionary
+from langchain_openai import ChatOpenAI
+from langgraph.graph import StateGraph, START, END
+from langgraph.checkpoint.memory import InMemorySaver
+from langgraph import interrupt, Command
+
+# Define the state schema according to the plan
+class StoryState(TypedDict):
+    theme: str
+    scene_text: str
+    options: List[str]
+    chosen_option: str
+    ending_text: str
+
+# Configure the LLM via environment variables
+openai_api_key = os.getenv("OPENAI_API_KEY")
+openai_base_url = os.getenv("OPENAI_BASE_URL")
+llm = ChatOpenAI(
+    model="gpt-4o",
+    temperature=0,
+    openai_api_key=openai_api_key,
+    base_url=openai_base_url,
+)
+
+async def call_llm(prompt: str) -> str:
+    """Invoke the LLM and return the text content, retrying once on failure."""
+    for attempt in range(2):
+        try:
+            response = await llm.invoke(prompt)
+            # ChatResult contains a list of messages
+            if hasattr(response, "messages") and response.messages:
+                content = response.messages[0].content
+            else:
+                # Fallback to direct attribute
+                content = getattr(response, "content", "")
+            if not content:
+                raise ValueError("Empty LLM response")
+            return content
+        except Exception as e:
+            if attempt == 1:
+                raise ValueError(f"LLM call failed twice: {e}") from e
+            await asyncio.sleep(1)
+    raise ValueError("LLM call failed")
+
+async def generate_scene(state: StoryState) -> Dict[str, Any]:
+    theme = state.get("theme", "Магический лес")
+    prompt = (
+        f"Сгенерируй сцену на тему '{theme}'.\n"
+        "Ответ должен быть в формате JSON:\n"
+        "{\"scene\": \"<описание>\", \"options\": [\"<вариант1>\",\"<вариант2>\",\"<вариант3>\"]}\n"
+    )
+    text = await call_llm(prompt)
+    try:
+        data = json.loads(text)
+    except Exception as e:
+        raise ValueError(f"Failed to parse scene JSON: {e}")
+    scene = data.get("scene", "")
+    options = data.get("options", [])
+    if not isinstance(options, list) or len(options) != 3:
+        raise ValueError("LLM must return exactly 3 options")
+    return {"scene_text": scene, "options": options}
+
+async def interrupt_choice(state: StoryState) -> Dict[str, Any]:
+    scene = state["scene_text"]
+    options = state["options"]
+    payload = {
+        "type": "choice",
+        "question": f"{scene} Что делаем?",
+        "options": options,
+    }
+    # Interrupt the graph, returning a special dictionary
+    return interrupt(Command(resume=payload))
+
+async def generate_ending(state: StoryState) -> Dict[str, Any]:
+    theme = state.get("theme", "Магический лес")
+    scene = state.get("scene_text", "")
+    choice = state.get("chosen_option", "")
+    prompt = (
+        f"На основе темы '{theme}', сцены '{scene}' и выбранного варианта '{choice}',\n"
+        "сгенерируй короткую концовку истории.\n"
+        "Ответ должен быть только текстом."
+    )
+    ending = (await call_llm(prompt)).strip()
+    return {"ending_text": ending}
+
+# Build the graph
+builder = StateGraph(StoryState)
+builder.add_node("generate_scene", generate_scene)
+builder.add_node("interrupt_choice", interrupt_choice)
+builder.add_node("generate_ending", generate_ending)
+
+builder.set_entry_point("generate_scene")
+builder.add_edge("generate_scene", "interrupt_choice")
+builder.add_edge("interrupt_choice", "generate_ending")
+builder.add_edge("generate_ending", END)
+
+# Persist state across interrupts
+builder.add_persisted_state(saver=InMemorySaver())
+
+graph = builder.compile()
+
+async def main() -> None:
+    theme = questionary.text("Введите тему истории:").ask()
+    if not theme:
+        theme = "Магический лес"
+    initial_state: StoryState = {
+        "theme": theme,
+        "scene_text": "",
+        "options": [],
+        "chosen_option": "",
+        "ending_text": "",
+    }
+    # Run the graph until the first interrupt
+    partial_state = await graph.ainvoke(initial_state)
+    if "_interrupt" in partial_state:
+        payload = partial_state["_interrupt"]
+        choice = questionary.select(payload["question"], choices=payload["options"]).ask()
+        partial_state["chosen_option"] = choice
+        thread_id = partial_state.get("thread_id")
+        config = {"configurable": {"thread_id": thread_id}} if thread_id else {}
+        final_state = await graph.ainvoke(partial_state, config=config)
+    else:
+        final_state = partial_state
+    # Print the final story
+    print("\n=== Итоговая история ===\n")
+    print(final_state.get("scene_text", ""))
+    print(f"\nВы выбрали: {final_state.get('chosen_option', '')}\n")
+    print(final_state.get("ending_text", ""))
+
+if __name__ == "__main__":
+    asyncio.run(main())