diff --git a/examples/README.md b/examples/README.md
index eae06ec2..6dbadbc9 100644
--- a/examples/README.md
+++ b/examples/README.md
@@ -6,6 +6,7 @@ Examples for common OpenSandbox use cases. Each subdirectory contains runnable c
 - [**echo**](echo): Minimal example using Ubuntu image to execute a simple echo command
 - [**aio-sandbox**](aio-sandbox): Basic example for agent_sandbox
 - [**code-interpreter**](code-interpreter): Code Interpreter SDK singleton example
+- [**rl-training**](rl-training): Reinforcement learning training loop inside a sandbox
 - [**claude-code**](claude-code): Call Claude (Anthropic) API/CLI within the sandbox
 - [**iflow-cli**](iflow-cli): CLI invocation template for iFlow/custom HTTP LLM services
 - [**gemini-cli**](gemini-cli): Call Google Gemini within the sandbox
diff --git a/examples/rl-training/README.md b/examples/rl-training/README.md
new file mode 100644
index 00000000..c057a89d
--- /dev/null
+++ b/examples/rl-training/README.md
@@ -0,0 +1,44 @@
+# Reinforcement Learning Sandbox Example
+
+Demonstrates running a basic RL training loop (CartPole + DQN) inside an isolated OpenSandbox container. The example installs RL dependencies in the sandbox, trains a policy, saves a checkpoint, and returns a training summary.
+
+## Start OpenSandbox server [local]
+
+Start the local OpenSandbox server:
+
+```shell
+git clone git@github.com:alibaba/OpenSandbox.git
+cd OpenSandbox/server
+cp example.config.toml ~/.sandbox.toml
+uv sync
+uv run python -m src.main
+```
+
+## Run the Example
+
+```shell
+# Install OpenSandbox package
+uv pip install opensandbox
+
+# Run the example
+uv run python examples/rl-training/main.py
+```
+
+The script provisions a sandbox, installs RL dependencies, trains a DQN agent on CartPole, saves a checkpoint, and prints the JSON training summary.
+
+![RL training screenshot](./screenshot.jpg)
+
+## Environment Variables
+
+- `SANDBOX_DOMAIN`: Sandbox service address (default: `localhost:8080`)
+- `SANDBOX_API_KEY`: API key if your server requires authentication
+- `SANDBOX_IMAGE`: Docker image to use (default: `sandbox-registry.cn-zhangjiakou.cr.aliyuncs.com/opensandbox/code-interpreter:latest`)
+- `RL_TIMESTEPS`: Training timesteps to run (default: `5000`)
+
+## TensorBoard
+
+The training script logs to `runs/`. To visualize metrics, open a shell in the sandbox and run:
+
+```shell
+tensorboard --logdir runs --host 0.0.0.0 --port 6006
+```
diff --git a/examples/rl-training/main.py b/examples/rl-training/main.py
new file mode 100644
index 00000000..18f67e1e
--- /dev/null
+++ b/examples/rl-training/main.py
@@ -0,0 +1,184 @@
+# Copyright 2025 Alibaba Group Holding Ltd.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import asyncio
+import os
+import textwrap
+from datetime import timedelta
+from pathlib import Path
+
+from opensandbox import Sandbox
+from opensandbox.config import ConnectionConfig
+
+
+def _load_requirements() -> str:
+    requirements_path = Path(__file__).with_name("requirements.txt")
+    return requirements_path.read_text(encoding="utf-8")
+
+
+def _training_script() -> str:
+    return textwrap.dedent(
+        """
+        import json
+        import os
+
+        import gymnasium as gym
+        from stable_baselines3 import DQN
+        from stable_baselines3.common.evaluation import evaluate_policy
+
+        timesteps = int(os.getenv("RL_TIMESTEPS", "5000"))
+        tensorboard_log = os.getenv("RL_TENSORBOARD_LOG", "runs")
+
+        env = gym.make("CartPole-v1")
+        model = DQN(
+            "MlpPolicy",
+            env,
+            verbose=1,
+            tensorboard_log=tensorboard_log,
+            learning_rate=1e-3,
+            buffer_size=10000,
+            learning_starts=1000,
+            batch_size=32,
+            train_freq=4,
+            gradient_steps=1,
+        )
+
+        model.learn(total_timesteps=timesteps)
+
+        os.makedirs("checkpoints", exist_ok=True)
+        checkpoint_path = "checkpoints/cartpole_dqn"
+        model.save(checkpoint_path)
+
+        mean_reward, std_reward = evaluate_policy(model, env, n_eval_episodes=5)
+        summary = {
+            "timesteps": timesteps,
+            "mean_reward": float(mean_reward),
+            "std_reward": float(std_reward),
+            "checkpoint_path": f"{checkpoint_path}.zip",
+        }
+        with open("training_summary.json", "w", encoding="utf-8") as handle:
+            json.dump(summary, handle, indent=2)
+
+        print("Training summary:", summary)
+        env.close()
+        """
+    ).lstrip()
+
+
+async def _print_execution_logs(execution) -> None:
+    for msg in execution.logs.stdout:
+        print(f"[stdout] {msg.text}")
+    for msg in execution.logs.stderr:
+        print(f"[stderr] {msg.text}")
+    if execution.error:
+        print(f"[error] {execution.error.name}: {execution.error.value}")
+
+
+def _execution_failed(execution) -> bool:
+    return execution.error is not None
+
+
+async def _run_command(sandbox: Sandbox, command: str) -> bool:
+    execution = await sandbox.commands.run(command)
+    await _print_execution_logs(execution)
+    return not _execution_failed(execution)
+
+
+def _with_python_env(command: str) -> str:
+    return (
+        "bash -lc '"
+        "source /opt/opensandbox/code-interpreter-env.sh "
+        "python ${PYTHON_VERSION:-3.14} >/dev/null "
+        "&& "
+        f"{command}"
+        "'"
+    )
+
+
+async def _ensure_pip(sandbox: Sandbox) -> bool:
+    bootstrap_commands = [
+        _with_python_env("python3 -m pip --version"),
+        _with_python_env("python3 -m ensurepip --upgrade"),
+        "apt-get update && apt-get install -y python3-pip",
+        "apk add --no-cache py3-pip",
+    ]
+    for command in bootstrap_commands:
+        if await _run_command(sandbox, command):
+            return True
+    return False
+
+
+async def _install_requirements(sandbox: Sandbox) -> bool:
+    install_commands = [
+        _with_python_env(
+            "python3 -m pip install --no-cache-dir --break-system-packages -r requirements.txt"
+        ),
+        "pip3 install --no-cache-dir -r requirements.txt",
+        "pip install --no-cache-dir -r requirements.txt",
+    ]
+    for command in install_commands:
+        if await _run_command(sandbox, command):
+            return True
+    return False
+
+
+async def main() -> None:
+    domain = os.getenv("SANDBOX_DOMAIN", "localhost:8080")
+    api_key = os.getenv("SANDBOX_API_KEY")
+    image = os.getenv("SANDBOX_IMAGE", "sandbox-registry.cn-zhangjiakou.cr.aliyuncs.com/opensandbox/code-interpreter:latest")
+    timesteps = os.getenv("RL_TIMESTEPS", "5000")
+
+    config = ConnectionConfig(
+        domain=domain,
+        api_key=api_key,
+        request_timeout=timedelta(minutes=10),
+    )
+
+    sandbox = await Sandbox.create(
+        image,
+        connection_config=config,
+        env={"RL_TIMESTEPS": timesteps},
+    )
+
+    async with sandbox:
+        try:
+            await sandbox.files.write_file("requirements.txt", _load_requirements())
+            if not await _ensure_pip(sandbox):
+                print("Failed to bootstrap pip inside the sandbox.")
+                return
+
+            if not await _install_requirements(sandbox):
+                print("Failed to install RL dependencies inside the sandbox.")
+                return
+
+            await sandbox.files.write_file("train.py", _training_script())
+            train_exec = await sandbox.commands.run(_with_python_env("python3 train.py"))
+            await _print_execution_logs(train_exec)
+            if _execution_failed(train_exec):
+                print("Training failed inside the sandbox.")
+                return
+
+            try:
+                summary = await sandbox.files.read_file("training_summary.json")
+            except Exception as exc:
+                print(f"\nFailed to read training summary: {exc}")
+            else:
+                print("\n=== Training summary ===")
+                print(summary)
+        finally:
+            await sandbox.kill()
+
+
+if __name__ == "__main__":
+    asyncio.run(main())
diff --git a/examples/rl-training/requirements.txt b/examples/rl-training/requirements.txt
new file mode 100644
index 00000000..e2e0340b
--- /dev/null
+++ b/examples/rl-training/requirements.txt
@@ -0,0 +1,4 @@
+gymnasium==0.29.1
+stable-baselines3==2.3.2
+tensorboard==2.16.2
+torch==2.9.1
diff --git a/examples/rl-training/screenshot.jpg b/examples/rl-training/screenshot.jpg
new file mode 100644
index 00000000..df6ff195
Binary files /dev/null and b/examples/rl-training/screenshot.jpg differ