Fix OpenEnv examples broken by async-first client API

docs/source/openenv.md

@@ -70,7 +70,7 @@ ENV_URL = "https://openenv-echo-env.hf.space"
 
 class EchoToolEnv:
     def __init__(self):
-        self.env = EchoEnv(base_url=ENV_URL)
+        self.env = EchoEnv(base_url=ENV_URL).sync()
         self.reward = 0.0
 
     def reset(self, **kwargs) -> str | None:
@@ -248,7 +248,7 @@ from textarena_env import TextArenaAction, TextArenaEnv
 
 class WordleEnv:
     def __init__(self):
-        self.client = TextArenaEnv(base_url="https://openenv-wordle.hf.space")
+        self.client = TextArenaEnv(base_url="https://openenv-wordle.hf.space").sync()
 
     def reset(self, **kwargs) -> str | None:
         result = self.client.reset()
@@ -401,7 +401,7 @@ class MultiEnv:
                     self._wordle_client.close()
                 except Exception:
                     pass
-            self._wordle_client = TextArenaEnv(base_url=WORDLE_URL)
+            self._wordle_client = TextArenaEnv(base_url=WORDLE_URL).sync()
             result = self._wordle_client.reset()
             self._last_full_feedback = result.observation.messages[0].content
             self.reward = 0.0
@@ -412,7 +412,7 @@ class MultiEnv:
                     self._catch_client.close()
                 except Exception:
                     pass
-            self._catch_client = OpenSpielEnv(base_url=CATCH_URL)
+            self._catch_client = OpenSpielEnv(base_url=CATCH_URL).sync()
             result = self._catch_client.reset()
             self.done = result.observation.done
             return _format_catch_obs(result.observation.info_state)
@@ -492,6 +492,9 @@ python examples/scripts/openenv/multi_env.py \
 
 When using `environment_factory`, the trainer connects to the environment server automatically. You just need the server to be running. There are three ways to run an OpenEnv environment server:
 
+> [!NOTE]
+> OpenEnv's client API is async-first: calling `reset()` or `step()` directly returns a coroutine. Appending `.sync()` (as in the examples below) returns a blocking client, which is what the synchronous `environment_factory` and `rollout_func` code expects. The classmethod constructors such as `from_docker_image()` are coroutines too (`await` them). See the OpenEnv [async/sync guide](https://github.com/huggingface/OpenEnv/blob/main/docs/source/guides/async-sync.md).
+
 <hfoptions id="env_mode">
 
 <hfoption id="space">
@@ -501,7 +504,7 @@ When using `environment_factory`, the trainer connects to the environment server
 Most example scripts default to a hosted Space (no setup needed):
 
 ```python
-env = EchoEnv(base_url="https://openenv-echo-env.hf.space")
+env = EchoEnv(base_url="https://openenv-echo-env.hf.space").sync()
 ```
 
 > [!WARNING]
@@ -520,7 +523,7 @@ docker run -d -p 8001:8000 --platform linux/amd64 registry.hf.space/openenv-echo
 Then connect:
 
 ```python
-env = EchoEnv(base_url="http://0.0.0.0:8001")
+env = EchoEnv(base_url="http://0.0.0.0:8001").sync()
 ```
 
 We map port 8001 to 8000 to leave port 8000 available for a vLLM server.
@@ -550,7 +553,7 @@ python -m uvicorn echo_env.src.envs.echo_env.server.app:app --host 0.0.0.0 --por
 Then connect:
 
 ```python
-env = EchoEnv(base_url="http://0.0.0.0:8001")
+env = EchoEnv(base_url="http://0.0.0.0:8001").sync()
 ```
 
 For more details, see the [OpenEnv catalog](https://huggingface.co/docs/openenv/environments).
@@ -636,7 +639,7 @@ trainer = GRPOTrainer(..., rollout_func=rollout_func)
 ```python
 class EchoToolEnv:
     def __init__(self):
-        self.env = EchoEnv(base_url=url)
+        self.env = EchoEnv(base_url=url).sync()
         self.reward = 0.0
 
     def reset(self, **kwargs) -> str | None:

examples/notebooks/grpo_functiongemma_browsergym_openenv.ipynb

@@ -116,7 +116,10 @@
         "from browsergym_env import BrowserGymEnv\n",
         "space_url = \"https://openenv-browsergym-env.hf.space\"\n",
         "\n",
-        "client = BrowserGymEnv(base_url=space_url)"
+        "# OpenEnv's client API is async-first: `reset()` and `step()` are coroutines.\n",
+        "# `.sync()` returns a synchronous wrapper so we can call them as plain blocking\n",
+        "# `reset()`/`step()` throughout this notebook.\n",
+        "client = BrowserGymEnv(base_url=space_url).sync()"
       ]
     },
     {

examples/notebooks/openenv_sudoku_grpo.ipynb

@@ -325,7 +325,8 @@
     "\n",
     "class SudokuEnv:\n",
     "    def __init__(self):\n",
-    "        self.client = TextArenaEnv(base_url=\"https://openenv-sudoku.hf.space\")\n",
+    "        # OpenEnv's client API is async-first; .sync() exposes blocking reset()/step().\n",
+    "        self.client = TextArenaEnv(base_url=\"https://openenv-sudoku.hf.space\").sync()\n",
     "        self.difficulty = \"easy\"\n",
     "        self.max_turns = 100\n",
     "        self._turn = 0\n",

examples/notebooks/openenv_wordle_grpo.ipynb

@@ -139,7 +139,8 @@
     "\n",
     "class WordleEnv:\n",
     "    def __init__(self):\n",
-    "        self.client = TextArenaEnv(base_url=\"https://openenv-wordle.hf.space\")\n",
+    "        # OpenEnv's client API is async-first; .sync() exposes blocking reset()/step().\n",
+    "        self.client = TextArenaEnv(base_url=\"https://openenv-wordle.hf.space\").sync()\n",
     "\n",
     "    def reset(self, **kwargs) -> None | str:\n",
     "        result = self.client.reset()\n",

examples/scripts/openenv/browsergym.py

@@ -130,7 +130,8 @@ def main() -> None:
 
     class BrowserGymVLMEnv:
         def __init__(self):
-            self.client = BrowserGymEnv(base_url=space_url)
+            # OpenEnv's client API is async-first; .sync() exposes blocking reset()/step().
+            self.client = BrowserGymEnv(base_url=space_url).sync()
             self.reward = 0.0
             self.done = False
             self._step_count = 0

examples/scripts/openenv/browsergym_llm.py

@@ -285,7 +285,8 @@ def main() -> None:
 
     class BrowserGymLLMEnv:
         def __init__(self):
-            self.client = BrowserGymEnv(base_url=space_url)
+            # OpenEnv's client API is async-first; .sync() exposes blocking reset()/step().
+            self.client = BrowserGymEnv(base_url=space_url).sync()
             self.reward = 0.0
             self._done = False
             self._step_count = 0

examples/scripts/openenv/carla.py

@@ -110,7 +110,8 @@ class CarlaGRPOEnv:
 
     def __init__(self):
         url = next(CarlaGRPOEnv._env_url_iter)
-        self.client = CarlaEnv(base_url=url, connect_timeout_s=30, message_timeout_s=120)
+        # OpenEnv's client API is async-first; .sync() exposes blocking reset()/step().
+        self.client = CarlaEnv(base_url=url, connect_timeout_s=30, message_timeout_s=120).sync()
 
     @staticmethod
     def _describe(obs) -> str:

examples/scripts/openenv/carla_vlm.py

@@ -135,7 +135,8 @@ class CarlaGRPOEnv:
 
     def __init__(self):
         url = next(CarlaGRPOEnv._env_url_iter)
-        self.client = CarlaEnv(base_url=url, connect_timeout_s=30, message_timeout_s=120)
+        # OpenEnv's client API is async-first; .sync() exposes blocking reset()/step().
+        self.client = CarlaEnv(base_url=url, connect_timeout_s=30, message_timeout_s=120).sync()
 
     @staticmethod
     def _describe(obs) -> str:

examples/scripts/openenv/carla_vlm_gemma.py

@@ -114,7 +114,8 @@ class CarlaVLMEnv:
 
     def __init__(self):
         self.url = next(CarlaVLMEnv._env_url_iter)
-        self.client = CarlaEnv(base_url=self.url, connect_timeout_s=30, message_timeout_s=120)
+        # OpenEnv's client API is async-first; .sync() exposes blocking reset()/step().
+        self.client = CarlaEnv(base_url=self.url, connect_timeout_s=30, message_timeout_s=120).sync()
         self.reward = 0.0
 
     @staticmethod
@@ -166,7 +167,7 @@ def reset(self, **kwargs) -> str | None:
                 if attempt == 2:
                     raise
                 print(f"[WARN] reset failed (attempt {attempt + 1}/3): {e}. Reconnecting...")
-                self.client = CarlaEnv(base_url=self.url, connect_timeout_s=30, message_timeout_s=120)
+                self.client = CarlaEnv(base_url=self.url, connect_timeout_s=30, message_timeout_s=120).sync()
 
     def observe(self) -> list:
         """

examples/scripts/openenv/catch.py

@@ -237,7 +237,8 @@ class CatchEnv:
         COLS = 5
 
         def __init__(self):
-            self.client = OpenSpielEnv(base_url=env_url)
+            # OpenEnv's client API is async-first; .sync() exposes blocking reset()/step().
+            self.client = OpenSpielEnv(base_url=env_url).sync()
             self.reward = 0.0
             self.done = False
 

examples/scripts/openenv/echo.py

@@ -93,7 +93,8 @@ def main():
 
     class EchoToolEnv:
         def __init__(self):
-            self.env = EchoEnv(base_url=args.env_host)
+            # OpenEnv's client API is async-first; .sync() exposes blocking reset()/step().
+            self.env = EchoEnv(base_url=args.env_host).sync()
             self.reward = 0.0
 
         def reset(self, **kwargs) -> None | str:

examples/scripts/openenv/multi_env.py

@@ -135,7 +135,8 @@ def reset(self, **kwargs) -> str | None:
                     self._wordle_client.close()
                 except Exception:
                     pass
-            self._wordle_client = TextArenaEnv(base_url=MultiEnv.wordle_url)
+            # OpenEnv's client API is async-first; .sync() exposes blocking reset()/step().
+            self._wordle_client = TextArenaEnv(base_url=MultiEnv.wordle_url).sync()
             result = self._wordle_client.reset()
             self._last_full_feedback = result.observation.messages[0].content
             self.reward = 0.0
@@ -146,7 +147,7 @@ def reset(self, **kwargs) -> str | None:
                     self._catch_client.close()
                 except Exception:
                     pass
-            self._catch_client = OpenSpielEnv(base_url=MultiEnv.catch_url)
+            self._catch_client = OpenSpielEnv(base_url=MultiEnv.catch_url).sync()
             result = self._catch_client.reset()
             self.done = result.observation.done
             return _format_catch_obs(result.observation.info_state)

examples/scripts/openenv/sudoku.py

@@ -410,7 +410,8 @@ def main() -> None:
 
     class SudokuEnv:
         def __init__(self):
-            self.client = TextArenaEnv(base_url=env_url)
+            # OpenEnv's client API is async-first; .sync() exposes blocking reset()/step().
+            self.client = TextArenaEnv(base_url=env_url).sync()
             self._difficulty = difficulty
             self._max_turns = max_turns
             self._api_delay = api_delay

examples/scripts/openenv/wordle.py

@@ -206,7 +206,8 @@ def main() -> None:
 
     class WordleEnv:
         def __init__(self):
-            self.client = TextArenaEnv(base_url=env_url)
+            # OpenEnv's client API is async-first; .sync() exposes blocking reset()/step().
+            self.client = TextArenaEnv(base_url=env_url).sync()
 
         def reset(self, **kwargs) -> str | None:
             result = self.client.reset()