diff --git a/atroposlib/tests/test_trl_vllm_server.py b/atroposlib/tests/test_trl_vllm_server.py
new file mode 100644
index 00000000..c829f853
--- /dev/null
+++ b/atroposlib/tests/test_trl_vllm_server.py
@@ -0,0 +1,90 @@
+"""Regression tests for TrlVllmServer wrappers (issue #183)."""
+
+from unittest.mock import patch
+
+import pytest
+from openai.types.chat.chat_completion import ChatCompletion
+from openai.types.completion import Completion
+
+from atroposlib.envs.server_handling.server_baseline import APIServerConfig
+from atroposlib.envs.server_handling.trl_vllm_server import TrlVllmServer
+
+
+class MockTokenizer:
+    """Minimal tokenizer stub for wrapper tests."""
+
+    eos_token_id = 2
+
+    def apply_chat_template(self, messages, tokenize=False, add_generation_prompt=True):
+        prompt = "".join(f"<{m['role']}>{m['content']}</{m['role']}>" for m in messages)
+        if add_generation_prompt:
+            prompt += "<assistant>"
+        return prompt
+
+    def decode(self, tokens, skip_special_tokens=False):
+        return "decoded"
+
+
+class MockResponse:
+    def __init__(self, payload):
+        self._payload = payload
+
+    async def json(self):
+        return self._payload
+
+    async def __aenter__(self):
+        return self
+
+    async def __aexit__(self, exc_type, exc, tb):
+        return False
+
+
+class MockSession:
+    def __init__(self, payload):
+        self._payload = payload
+
+    async def __aenter__(self):
+        return self
+
+    async def __aexit__(self, exc_type, exc, tb):
+        return False
+
+    def post(self, *args, **kwargs):
+        return MockResponse(self._payload)
+
+
+@pytest.fixture
+def mock_server():
+    config = APIServerConfig(
+        api_key="x",
+        base_url="http://localhost:8000",
+        model_name="test-model",
+        timeout=30,
+    )
+    with patch(
+        "atroposlib.envs.server_handling.trl_vllm_server.AutoTokenizer"
+    ) as mock_auto:
+        mock_auto.from_pretrained.return_value = MockTokenizer()
+        yield TrlVllmServer(config)
+
+
+@patch("atroposlib.envs.server_handling.trl_vllm_server.aiohttp.ClientSession")
+@pytest.mark.asyncio
+async def test_completion_wrapper_returns_text_completion(
+    mock_session_cls, mock_server
+):
+    mock_session_cls.return_value = MockSession({"completion_ids": [[1, 2, 3]]})
+    result = await mock_server._completion_wrapper(prompt="Hello")
+    assert isinstance(result, Completion)
+    assert result.choices[0].text == "decoded"
+
+
+@patch("atroposlib.envs.server_handling.trl_vllm_server.aiohttp.ClientSession")
+@pytest.mark.asyncio
+async def test_chat_completion_wrapper_returns_message(mock_session_cls, mock_server):
+    mock_session_cls.return_value = MockSession({"completion_ids": [[1, 2, 3]]})
+    result = await mock_server._chat_completion_wrapper(
+        messages=[{"role": "user", "content": "Hi"}]
+    )
+    assert isinstance(result, ChatCompletion)
+    assert result.choices[0].message.content == "decoded"