diff --git a/src/codeinterpreterapi/crewai/crew_agent.py b/src/codeinterpreterapi/crewai/crew_agent.py
new file mode 100644
index 00000000..1e44479b
--- /dev/null
+++ b/src/codeinterpreterapi/crewai/crew_agent.py
@@ -0,0 +1,60 @@
+from typing import Dict
+
+from crewai import Agent, Crew, Task
+
+from codeinterpreterapi.agents.agents import CodeInterpreterAgent
+from codeinterpreterapi.brain.params import CodeInterpreterParams
+from codeinterpreterapi.crewai.custom_agent import (
+    CustomAgent,  # You need to build and extend your own agent logic with the CrewAI BaseAgent class then import it here.
+)
+from codeinterpreterapi.graphs.agent_wrapper_tool import AgentWrapperTool
+from codeinterpreterapi.llm.llm import prepare_test_llm
+from codeinterpreterapi.planners.planners import CodeInterpreterPlanner
+from codeinterpreterapi.supervisors.supervisors import CodeInterpreterSupervisor
+from codeinterpreterapi.test_prompts.test_prompt import TestPrompt
+
+
+def run(ci_params: CodeInterpreterParams, inputs: Dict):
+    agents = []
+    tasks = []
+    tools = AgentWrapperTool.create_agent_wrapper_tools(ci_params)
+    for agent_def in ci_params.agent_def_list:
+        agent_executor = agent_def.agent_executor
+        role = agent_def.agent_name
+        goal = "clear information provide for user about " + agent_def.agent_name
+        backstory = agent_def.agent_role
+        agent_custom = CustomAgent(agent_executor=agent_executor, role=role, goal=goal, backstory=backstory)
+        agent = Agent(
+            role=role,
+            goal=goal,
+            backstory=backstory,
+            tools=[tools[0]],
+            llm=ci_params.llm_fast,
+        )
+        agents.append(agent_custom)
+
+        task = Task(
+            expected_output=agent_def.agent_role + " of {input}",
+            description=agent_def.agent_role,
+            agent=agent,
+        )
+        tasks.append(task)
+
+    my_crew = Crew(agents=agents, tasks=tasks)
+    result = my_crew.kickoff(inputs=inputs)
+    return result
+
+
+def test():
+    llm, llm_tools, runnable_config = prepare_test_llm()
+    ci_params = CodeInterpreterParams.get_test_params(llm=llm, llm_tools=llm_tools, runnable_config=runnable_config)
+    _ = CodeInterpreterAgent.choose_agent_executors(ci_params=ci_params)
+    planner = CodeInterpreterPlanner.choose_planner(ci_params=ci_params)
+    _ = CodeInterpreterSupervisor.choose_supervisor(planner=planner, ci_params=ci_params)
+    inputs = {"input": TestPrompt.svg_input_str}
+    result = run(ci_params, inputs)
+    print(result)
+
+
+if __name__ == "__main__":
+    test()
diff --git a/src/codeinterpreterapi/crewai/custom_agent.py b/src/codeinterpreterapi/crewai/custom_agent.py
new file mode 100644
index 00000000..34d4c2a7
--- /dev/null
+++ b/src/codeinterpreterapi/crewai/custom_agent.py
@@ -0,0 +1,62 @@
+from typing import Any, Dict, List, Optional
+
+from crewai.agents.agent_builder.base_agent import BaseAgent
+from langchain_core.tools import BaseTool
+from pydantic import Field
+
+
+class CustomAgent(BaseAgent):
+    agent_executor: Any = Field(default=None, description="Verbose mode for the Agent Execution")
+    function_calling_llm: Optional[Any] = Field(description="Language model that will run the agent.", default=None)
+    allow_code_execution: Optional[bool] = Field(default=False, description="Enable code execution for the agent.")
+    step_callback: Optional[Any] = Field(
+        default=None,
+        description="Callback to be executed after each step of the agent execution.",
+    )
+
+    def __init__(self, agent_executor: Any, **data):
+        config = data.pop("config", {})
+
+        super().__init__(**config, **data)
+        self.agent_executor = agent_executor
+        self.function_calling_llm = "dummy"  # This is not used
+        self.allow_code_execution = False
+        self.step_callback = None
+
+    def interpolate_inputs(self, inputs: Dict[str, Any]) -> None:
+        """Interpolate inputs into the task description and expected output."""
+        print("interpolate_inputs inputs=", inputs)
+        super().interpolate_inputs(inputs)
+
+    def execute_task(self, task: Any, context: Optional[str] = None, tools: Optional[List[Any]] = None):
+        # AgentExecutorを使用してタスクを実行
+        print("execute_task task=", task)
+        print("execute_task context=", context)
+        input_dict = {}
+        input_dict["input"] = context
+        input_dict["question"] = context
+        input_dict["message"] = task.description + "を実行してください。\n" + task.expected_output
+        result = self.agent_executor.invoke(input=input_dict)
+        return result
+
+    def create_agent_executor(self, tools=None) -> None:
+        pass
+
+    def _parse_tools(self, tools: List[Any]) -> List[Any]:
+        return []
+
+    def parse_tools(self, tools: Optional[List[BaseTool]]) -> List[BaseTool]:
+        # ツールのパースロジックを実装
+        return tools or []
+
+    def get_delegation_tools(self, agents: List[BaseAgent]):
+        return []
+
+    def get_output_converter(self, llm, text, model, instructions):
+        return lambda x: x  # デフォルトでは変換なし
+
+    def execute(self, task_description: str, context: Optional[List[str]] = None):
+        # タスクの実行ロジックを実装
+        full_context = "\n".join(context) if context else ""
+        full_input = f"{full_context}\n\nTask: {task_description}"
+        return self.executor.run(input=full_input)