hud-python 0.4.35__py3-none-any.whl → 0.4.37__py3-none-any.whl

hud/server/tests/test_server_extra.py

@@ -0,0 +1,166 @@
+# filename: hud/server/tests/test_server_extra.py
+from __future__ import annotations
+
+import asyncio
+from contextlib import asynccontextmanager, suppress
+
+import anyio
+import pytest
+
+from hud.server import MCPServer
+from hud.server import server as server_mod
+
+
+@asynccontextmanager
+async def _fake_stdio_server():
+    """
+    Stand-in for stdio_server that avoids reading real stdin.
+
+    It yields a pair of in-memory streams (receive, send) so the low-level server
+    can start and idle without touching sys.stdin/sys.stdout.
+    """
+    send_in, recv_in = anyio.create_memory_object_stream(100)
+    send_out, recv_out = anyio.create_memory_object_stream(100)
+    try:
+        yield recv_in, send_out
+    finally:
+        # best-effort close across anyio versions
+        for s in (send_in, recv_in, send_out, recv_out):
+            close = getattr(s, "close", None) or getattr(s, "aclose", None)
+            try:
+                if close is not None:
+                    res = close()
+                    if asyncio.iscoroutine(res):
+                        await res
+            except Exception:
+                pass
+
+
+@pytest.fixture
+def patch_stdio(monkeypatch: pytest.MonkeyPatch):
+    """Patch stdio server to avoid stdin issues during tests."""
+    monkeypatch.setenv("FASTMCP_DISABLE_BANNER", "1")
+    monkeypatch.setattr("mcp.server.stdio.stdio_server", _fake_stdio_server)
+    monkeypatch.setattr("fastmcp.server.server.stdio_server", _fake_stdio_server)
+
+
+@pytest.mark.asyncio
+async def test_sigterm_flag_remains_true_without_shutdown_handler(patch_stdio):
+    """
+    When no @mcp.shutdown is registered, neither the lifespan.finally nor run_async.finally
+    should reset the global SIGTERM flag. This exercises the 'no handler' branches.
+    """
+    mcp = MCPServer(name="NoShutdownHandler")
+
+    task = asyncio.create_task(mcp.run_async(transport="stdio", show_banner=False))
+    try:
+        await asyncio.sleep(0.05)
+        # Simulate SIGTERM path
+        server_mod._sigterm_received = True  # type: ignore[attr-defined]
+    finally:
+        with suppress(asyncio.CancelledError):
+            task.cancel()
+            await task
+
+    # Flag must remain set since no shutdown handler was installed
+    assert getattr(server_mod, "_sigterm_received") is True
+
+    # Always reset for other tests
+    server_mod._sigterm_received = False  # type: ignore[attr-defined]
+
+
+@pytest.mark.asyncio
+async def test_last_shutdown_handler_wins(patch_stdio):
+    """
+    If multiple @mcp.shutdown decorators are applied, the last one should be the one that runs.
+    """
+    mcp = MCPServer(name="ShutdownOverride")
+    calls: list[str] = []
+
+    @mcp.shutdown
+    async def _first() -> None:
+        calls.append("first")
+
+    @mcp.shutdown
+    async def _second() -> None:
+        calls.append("second")
+
+    task = asyncio.create_task(mcp.run_async(transport="stdio", show_banner=False))
+    try:
+        await asyncio.sleep(0.05)
+        server_mod._sigterm_received = True  # type: ignore[attr-defined]
+    finally:
+        with suppress(asyncio.CancelledError):
+            task.cancel()
+            await task
+
+    assert calls == ["second"], "Only the last registered shutdown handler should run"
+    server_mod._sigterm_received = False  # type: ignore[attr-defined]
+
+
+def test__run_with_sigterm_registers_handlers_when_enabled(monkeypatch: pytest.MonkeyPatch):
+    """
+    Verify that _run_with_sigterm attempts to register SIGTERM/SIGINT handlers
+    when the env var does NOT disable the handler. We stub AnyIO's TaskGroup so
+    the watcher doesn't block and the test returns immediately.
+    """
+    # Ensure handler is enabled
+    monkeypatch.delenv("FASTMCP_DISABLE_SIGTERM_HANDLER", raising=False)
+
+    # Record what the server tries to register
+    added_signals: list[int] = []
+
+    import asyncio as _asyncio
+
+    orig_get_running_loop = _asyncio.get_running_loop
+
+    def proxy_get_running_loop():
+        real = orig_get_running_loop()
+
+        class _LoopProxy:
+            __slots__ = ("_inner",)
+
+            def __init__(self, inner):
+                self._inner = inner
+
+            def add_signal_handler(self, signum, callback, *args):
+                added_signals.append(signum)  # don't actually install
+                # no-op: skip calling inner.add_signal_handler to avoid OS constraints
+
+            def __getattr__(self, name):
+                # delegate everything else (create_task, call_soon, etc.)
+                return getattr(self._inner, name)
+
+        return _LoopProxy(real)
+
+    # Patch globally so both the test and hud.server.server see the proxy
+    monkeypatch.setattr(_asyncio, "get_running_loop", proxy_get_running_loop)
+
+    # Dummy TaskGroup that runs the work but skips _watch
+    class _DummyTG:
+        async def __aenter__(self):
+            return self
+
+        async def __aexit__(self, exc_type, exc, tb):
+            return False
+
+        def start_soon(self, fn, *args, **kwargs):
+            if getattr(fn, "__name__", "") == "_watch":
+                return
+            _asyncio.get_running_loop().create_task(fn(*args, **kwargs))
+
+    monkeypatch.setattr("anyio.create_task_group", lambda: _DummyTG())
+
+    # Simple coroutine that should run to completion
+    hit = {"v": False}
+
+    async def work():
+        hit["v"] = True
+
+    server_mod._run_with_sigterm(work)
+    assert hit["v"] is True
+
+    import signal as _signal
+
+    assert _signal.SIGTERM in added_signals
+    assert _signal.SIGINT in added_signals

hud/server/tests/test_sigterm_runner.py

@@ -0,0 +1,78 @@
+from __future__ import annotations
+
+import asyncio
+from contextlib import asynccontextmanager, suppress
+
+import anyio
+import pytest
+
+from hud.server import MCPServer
+from hud.server import server as server_mod
+
+
+def test__run_with_sigterm_executes_coro_when_handler_disabled(monkeypatch: pytest.MonkeyPatch):
+    """With FASTMCP_DISABLE_SIGTERM_HANDLER=1, _run_with_sigterm should just run the task."""
+    monkeypatch.setenv("FASTMCP_DISABLE_SIGTERM_HANDLER", "1")
+
+    hit = {"v": False}
+
+    async def work(arg, *, kw=None):
+        assert arg == 123 and kw == "ok"
+        hit["v"] = True
+
+    # Wrapper to exercise kwargs since TaskGroup.start_soon only accepts positional args
+    async def wrapper(arg):
+        await work(arg, kw="ok")
+
+    # Should return cleanly and mark hit
+    server_mod._run_with_sigterm(wrapper, 123)
+    assert hit["v"] is True
+
+
+@asynccontextmanager
+async def _fake_stdio_server():
+    """Stand-in for stdio_server that avoids reading real stdin."""
+    send_in, recv_in = anyio.create_memory_object_stream(100)
+    send_out, recv_out = anyio.create_memory_object_stream(100)
+    try:
+        yield recv_in, send_out
+    finally:
+        for s in (send_in, recv_in, send_out, recv_out):
+            close = getattr(s, "close", None) or getattr(s, "aclose", None)
+            try:
+                if close is not None:
+                    res = close()
+                    if asyncio.iscoroutine(res):
+                        await res
+            except Exception:
+                pass
+
+
+@pytest.fixture
+def patch_stdio(monkeypatch: pytest.MonkeyPatch):
+    """Patch stdio server to avoid stdin issues during tests."""
+    monkeypatch.setenv("FASTMCP_DISABLE_BANNER", "1")
+    monkeypatch.setattr("mcp.server.stdio.stdio_server", _fake_stdio_server)
+    monkeypatch.setattr("fastmcp.server.server.stdio_server", _fake_stdio_server)
+
+
+@pytest.mark.asyncio
+async def test_shutdown_handler_exception_does_not_crash_and_resets_flag(patch_stdio):
+    """If @shutdown raises, run_async must swallow it and still reset the SIGTERM flag."""
+    mcp = MCPServer(name="ShutdownRaises")
+
+    @mcp.shutdown
+    async def _boom() -> None:
+        raise RuntimeError("kaboom")
+
+    task = asyncio.create_task(mcp.run_async(transport="stdio", show_banner=False))
+    try:
+        await asyncio.sleep(0.05)
+        server_mod._sigterm_received = True  # trigger shutdown path
+    finally:
+        with suppress(asyncio.CancelledError):
+            task.cancel()
+            await task
+
+    # No exception propagated; flag must be reset
+    assert not getattr(server_mod, "_sigterm_received")

hud/settings.py

@@ -1,7 +1,10 @@
 from __future__ import annotations
 
+from pathlib import Path
+
 from pydantic import Field
 from pydantic_settings import BaseSettings, SettingsConfigDict
+from pydantic_settings.sources import DotEnvSettingsSource, PydanticBaseSettingsSource
 
 
 class Settings(BaseSettings):
@@ -14,6 +17,41 @@ class Settings(BaseSettings):
 
     model_config = SettingsConfigDict(env_file=".env", env_file_encoding="utf-8", extra="allow")
 
+    @classmethod
+    def settings_customise_sources(
+        cls,
+        settings_cls: type[BaseSettings],
+        init_settings: PydanticBaseSettingsSource,
+        env_settings: PydanticBaseSettingsSource,
+        dotenv_settings: PydanticBaseSettingsSource,
+        file_secret_settings: PydanticBaseSettingsSource,
+    ) -> tuple[PydanticBaseSettingsSource, ...]:
+        """
+        Customize settings source precedence to include a user-level env file.
+
+        Precedence (highest to lowest):
+        - init_settings (explicit kwargs)
+        - env_settings (process environment)
+        - dotenv_settings (project .env)
+        - user_dotenv_settings (~/.hud/.env)  ← added
+        - file_secret_settings
+        """
+
+        user_env_path = Path.home() / ".hud" / ".env"
+        user_dotenv_settings = DotEnvSettingsSource(
+            settings_cls,
+            env_file=user_env_path,
+            env_file_encoding="utf-8",
+        )
+
+        return (
+            init_settings,
+            env_settings,
+            dotenv_settings,
+            user_dotenv_settings,
+            file_secret_settings,
+        )
+
     hud_telemetry_url: str = Field(
         default="https://telemetry.hud.so/v3/api",
         description="Base URL for the HUD API",

hud/shared/hints.py

@@ -37,8 +37,8 @@ HUD_API_KEY_MISSING = Hint(
     title="HUD API key required",
     message="Missing or invalid HUD_API_KEY.",
     tips=[
-        "Set HUD_API_KEY environment variable",
-        "Get a key at https://app.hud.so",
+        "Set HUD_API_KEY in your environment or run: hud set HUD_API_KEY=your-key-here",
+        "Get a key at https://hud.so",
         "Check for whitespace or truncation",
     ],
     docs_url=None,

hud/telemetry/job.py

@@ -143,7 +143,7 @@ def _print_job_url(job_id: str, job_name: str) -> None:
     if not (settings.telemetry_enabled and settings.api_key):
         return
 
-    url = f"https://app.hud.so/jobs/{job_id}"
+    url = f"https://hud.so/jobs/{job_id}"
     header = f"🚀 Job '{job_name}' started:"
 
     # ANSI color codes
@@ -182,7 +182,7 @@ def _print_job_complete_url(job_id: str, job_name: str, error_occurred: bool = F
     if not (settings.telemetry_enabled and settings.api_key):
         return
 
-    url = f"https://app.hud.so/jobs/{job_id}"
+    url = f"https://hud.so/jobs/{job_id}"
 
     # ANSI color codes
     GREEN = "\033[92m"

hud/types.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import contextlib
 import json
 import logging
 import uuid
@@ -107,7 +108,13 @@ class Task(BaseModel):
 
         # Start with current environment variables
         mapping = dict(os.environ)
-        mapping.update(settings.model_dump())
+        # Include settings (from process env, project .env, and user .env)
+        settings_dict = settings.model_dump()
+        mapping.update(settings_dict)
+        # Add UPPERCASE aliases for settings keys
+        for _key, _val in settings_dict.items():
+            with contextlib.suppress(Exception):
+                mapping[_key.upper()] = _val
 
         if settings.api_key:
             mapping["HUD_API_KEY"] = settings.api_key
@@ -208,7 +215,7 @@ class AgentResponse(BaseModel):
     tool_calls: list[MCPToolCall] = Field(default_factory=list)
     done: bool = Field(default=False)
 
-    # --- TELEMETRY [app.hud.so] ---
+    # --- TELEMETRY [hud.so] ---
     # Responses
     content: str | None = Field(default=None)
     reasoning: str | None = Field(default=None)

hud/utils/tasks.py

@@ -9,7 +9,7 @@ from hud.utils.hud_console import HUDConsole
 hud_console = HUDConsole()
 
 
-def load_tasks(tasks_input: str | list[dict]) -> list[Task]:
+def load_tasks(tasks_input: str | list[dict], *, raw: bool = False) -> list[Task] | list[dict]:
     """Load tasks from various sources.
 
     Args:
@@ -18,16 +18,19 @@ def load_tasks(tasks_input: str | list[dict]) -> list[Task]:
             - Path to a JSONL file (one task per line)
             - HuggingFace dataset name (format: "username/dataset" or "username/dataset:split")
             - List of task dictionaries
-        system_prompt: Default system prompt to use if not specified in task
+        raw: If True, return raw dicts without validation or env substitution
 
     Returns:
-        List of validated HUD Task objects
+        - If raw=False (default): list[Task]
+        - If raw=True: list[dict]
     """
-    tasks = []
+    tasks: list[Task] | list[dict] = []
 
     if isinstance(tasks_input, list):
         # Direct list of task dicts
         hud_console.info(f"Loading {len(tasks_input)} tasks from provided list")
+        if raw:
+            return [item for item in tasks_input if isinstance(item, dict)]
         for item in tasks_input:
             task = Task(**item)
             tasks.append(task)
@@ -36,7 +39,6 @@ def load_tasks(tasks_input: str | list[dict]) -> list[Task]:
         # Check if it's a file path
         if Path(tasks_input).exists():
             file_path = Path(tasks_input)
-            hud_console.info(f"Loading tasks from file: {tasks_input}")
 
             with open(file_path) as f:
                 # Handle JSON files (array of tasks)
@@ -46,31 +48,33 @@ def load_tasks(tasks_input: str | list[dict]) -> list[Task]:
                         raise ValueError(
                             f"JSON file must contain an array of tasks, got {type(data)}"
                         )
-
+                    if raw:
+                        return [item for item in data if isinstance(item, dict)]
                     for item in data:
                         task = Task(**item)
                         tasks.append(task)
 
                 # Handle JSONL files (one task per line)
                 else:
+                    raw_items: list[dict] = []
                     for line in f:
                         line = line.strip()
-                        if line:  # Skip empty lines
-                            item = json.loads(line)
-
-                            # Handle case where line contains an array of tasks
-                            if isinstance(item, list):
-                                for task_item in item:
-                                    task = Task(**task_item)
-                                    tasks.append(task)
-                            # Handle normal case where line contains a single task object
-                            elif isinstance(item, dict):
-                                task = Task(**item)
-                                tasks.append(task)
-                            else:
-                                raise ValueError(
-                                    f"Invalid JSONL format: expected dict or list of dicts, got {type(item)}"  # noqa: E501
-                                )
+                        if not line:
+                            continue
+                        item = json.loads(line)
+                        if isinstance(item, list):
+                            raw_items.extend([it for it in item if isinstance(it, dict)])
+                        elif isinstance(item, dict):
+                            raw_items.append(item)
+                        else:
+                            raise ValueError(
+                                f"Invalid JSONL format: expected dict or list of dicts, got {type(item)}"  # noqa: E501
+                            )
+                    if raw:
+                        return raw_items
+                    for it in raw_items:
+                        task = Task(**it)
+                        tasks.append(task)
 
         # Check if it's a HuggingFace dataset
         elif "/" in tasks_input:
@@ -88,6 +92,7 @@ def load_tasks(tasks_input: str | list[dict]) -> list[Task]:
                 dataset = load_dataset(dataset_name, split=split)
 
                 # Convert dataset rows to Task objects
+                raw_rows: list[dict] = []
                 for item in dataset:
                     if not isinstance(item, dict):
                         raise ValueError(
@@ -97,7 +102,11 @@ def load_tasks(tasks_input: str | list[dict]) -> list[Task]:
                         raise ValueError(
                             f"Invalid HuggingFace dataset: expected mcp_config and prompt, got {item}"  # noqa: E501
                         )
-                    task = Task(**item)
+                    raw_rows.append(item)
+                if raw:
+                    return raw_rows
+                for row in raw_rows:
+                    task = Task(**row)
                     tasks.append(task)
 
             except ImportError as e:
@@ -115,5 +124,4 @@ def load_tasks(tasks_input: str | list[dict]) -> list[Task]:
     else:
         raise TypeError(f"tasks_input must be str or list, got {type(tasks_input)}")
 
-    hud_console.info(f"Loaded {len(tasks)} tasks")
     return tasks

hud/utils/tests/test_version.py

@@ -5,4 +5,4 @@ def test_import():
     """Test that the package can be imported."""
     import hud
 
-    assert hud.__version__ == "0.4.35"
+    assert hud.__version__ == "0.4.37"

hud/version.py

@@ -4,4 +4,4 @@ Version information for the HUD SDK.
 
 from __future__ import annotations
 
-__version__ = "0.4.35"
+__version__ = "0.4.37"

{hud_python-0.4.35.dist-info → hud_python-0.4.37.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: hud-python
-Version: 0.4.35
+Version: 0.4.37
 Summary: SDK for the HUD platform.
 Project-URL: Homepage, https://github.com/hud-evals/hud-python
 Project-URL: Bug Tracker, https://github.com/hud-evals/hud-python/issues
@@ -36,11 +36,13 @@ Classifier: Programming Language :: Python :: 3.12
 Classifier: Programming Language :: Python :: 3.13
 Requires-Python: <3.13,>=3.11
 Requires-Dist: anthropic
+Requires-Dist: blessed>=1.20.0
 Requires-Dist: datasets>=2.14.0
 Requires-Dist: httpx<1,>=0.23.0
 Requires-Dist: hud-fastmcp-python-sdk>=0.1.2
 Requires-Dist: hud-mcp-python-sdk>=3.13.2
-Requires-Dist: hud-mcp-use-python-sdk>=2.3.16
+Requires-Dist: hud-mcp-use-python-sdk==2.3.19
+Requires-Dist: litellm>=1.55.0
 Requires-Dist: numpy>=1.24.0
 Requires-Dist: openai
 Requires-Dist: opentelemetry-api>=1.34.1
@@ -50,8 +52,8 @@ Requires-Dist: opentelemetry-sdk>=1.34.1
 Requires-Dist: pathspec>=0.12.1
 Requires-Dist: pillow>=11.1.0
 Requires-Dist: prompt-toolkit==3.0.51
-Requires-Dist: pydantic-settings<3,>=2
-Requires-Dist: pydantic<3,>=2
+Requires-Dist: pydantic-settings<3,>=2.2
+Requires-Dist: pydantic<3,>=2.6
 Requires-Dist: questionary==2.1.0
 Requires-Dist: rich>=13.0.0
 Requires-Dist: toml>=0.10.2
@@ -59,7 +61,9 @@ Requires-Dist: typer>=0.9.0
 Requires-Dist: watchfiles>=0.21.0
 Requires-Dist: wrapt>=1.14.0
 Provides-Extra: agent
+Requires-Dist: aiodocker>=0.24.0; extra == 'agent'
 Requires-Dist: dotenv>=0.9.9; extra == 'agent'
+Requires-Dist: inspect-ai>=0.3.80; extra == 'agent'
 Requires-Dist: ipykernel; extra == 'agent'
 Requires-Dist: ipython<9; extra == 'agent'
 Requires-Dist: jupyter-client; extra == 'agent'
@@ -67,8 +71,21 @@ Requires-Dist: jupyter-core; extra == 'agent'
 Requires-Dist: langchain; extra == 'agent'
 Requires-Dist: langchain-anthropic; extra == 'agent'
 Requires-Dist: langchain-openai; extra == 'agent'
+Requires-Dist: pillow>=11.1.0; extra == 'agent'
+Requires-Dist: playwright; extra == 'agent'
+Requires-Dist: pyautogui>=0.9.54; extra == 'agent'
+Requires-Dist: pyright==1.1.401; extra == 'agent'
+Requires-Dist: pytest-asyncio; extra == 'agent'
+Requires-Dist: pytest-cov; extra == 'agent'
+Requires-Dist: pytest-mock; extra == 'agent'
+Requires-Dist: pytest<9,>=8.1.1; extra == 'agent'
+Requires-Dist: ruff>=0.11.8; extra == 'agent'
+Requires-Dist: setuptools; extra == 'agent'
+Requires-Dist: textdistance<5,>=4.5.0; extra == 'agent'
 Provides-Extra: agents
+Requires-Dist: aiodocker>=0.24.0; extra == 'agents'
 Requires-Dist: dotenv>=0.9.9; extra == 'agents'
+Requires-Dist: inspect-ai>=0.3.80; extra == 'agents'
 Requires-Dist: ipykernel; extra == 'agents'
 Requires-Dist: ipython<9; extra == 'agents'
 Requires-Dist: jupyter-client; extra == 'agents'
@@ -76,6 +93,17 @@ Requires-Dist: jupyter-core; extra == 'agents'
 Requires-Dist: langchain; extra == 'agents'
 Requires-Dist: langchain-anthropic; extra == 'agents'
 Requires-Dist: langchain-openai; extra == 'agents'
+Requires-Dist: pillow>=11.1.0; extra == 'agents'
+Requires-Dist: playwright; extra == 'agents'
+Requires-Dist: pyautogui>=0.9.54; extra == 'agents'
+Requires-Dist: pyright==1.1.401; extra == 'agents'
+Requires-Dist: pytest-asyncio; extra == 'agents'
+Requires-Dist: pytest-cov; extra == 'agents'
+Requires-Dist: pytest-mock; extra == 'agents'
+Requires-Dist: pytest<9,>=8.1.1; extra == 'agents'
+Requires-Dist: ruff>=0.11.8; extra == 'agents'
+Requires-Dist: setuptools; extra == 'agents'
+Requires-Dist: textdistance<5,>=4.5.0; extra == 'agents'
 Provides-Extra: dev
 Requires-Dist: aiodocker>=0.24.0; extra == 'dev'
 Requires-Dist: dotenv>=0.9.9; extra == 'dev'
@@ -100,14 +128,6 @@ Requires-Dist: setuptools; extra == 'dev'
 Requires-Dist: textdistance<5,>=4.5.0; extra == 'dev'
 Provides-Extra: rl
 Requires-Dist: bitsandbytes>=0.41.0; (sys_platform == 'linux') and extra == 'rl'
-Requires-Dist: dotenv>=0.9.9; extra == 'rl'
-Requires-Dist: ipykernel; extra == 'rl'
-Requires-Dist: ipython<9; extra == 'rl'
-Requires-Dist: jupyter-client; extra == 'rl'
-Requires-Dist: jupyter-core; extra == 'rl'
-Requires-Dist: langchain; extra == 'rl'
-Requires-Dist: langchain-anthropic; extra == 'rl'
-Requires-Dist: langchain-openai; extra == 'rl'
 Requires-Dist: liger-kernel>=0.5.0; (sys_platform == 'linux') and extra == 'rl'
 Requires-Dist: peft>=0.17.1; extra == 'rl'
 Requires-Dist: vllm==0.10.1.1; extra == 'rl'
@@ -138,8 +158,8 @@ OSS RL environment + evals toolkit. Wrap software as environments, run benchmark
 ## Highlights
 
 - 🚀 **[MCP environment skeleton](https://docs.hud.so/core-concepts/mcp-protocol)** – any agent can call any environment.
-- ⚡️ **[Live telemetry](https://app.hud.so)** – inspect every tool call, observation, and reward in real time.
-- 🗂️ **[Public benchmarks](https://app.hud.so/leaderboards)** – OSWorld-Verified, SheetBench-50, and more.
+- ⚡️ **[Live telemetry](https://hud.so)** – inspect every tool call, observation, and reward in real time.
+- 🗂️ **[Public benchmarks](https://hud.so/leaderboards)** – OSWorld-Verified, SheetBench-50, and more.
 - 🌱 **[Reinforcement learning built-in](rl/)** – Verifiers gym pipelines for GRPO on any environment.
 - 🌐 **[Cloud browsers](environments/remote_browser/)** – AnchorBrowser, Steel, BrowserBase integrations for browser automation.
 - 🛠️ **[Hot-reload dev loop](environments/README.md#phase-5-hot-reload-development-with-cursor-agent)** – `hud dev` for iterating on environments without rebuilds.
@@ -185,14 +205,14 @@ from hud.agents import ClaudeAgent
 from hud.datasets import Task  # See docs: https://docs.hud.so/reference/tasks
 
 async def main() -> None:
-    with hud.trace("Quick Start 2048"): # All telemetry works for any MCP-based agent (see https://app.hud.so)
+    with hud.trace("Quick Start 2048"): # All telemetry works for any MCP-based agent (see https://hud.so)
         task = {
             "prompt": "Reach 64 in 2048.",
             "mcp_config": {
                 "hud": {
                     "url": "https://mcp.hud.so/v3/mcp",  # HUD's cloud MCP server (see https://docs.hud.so/core-concepts/architecture)
                     "headers": {
-                        "Authorization": f"Bearer {settings.api_key}",  # Get your key at https://app.hud.so
+                        "Authorization": f"Bearer {settings.api_key}",  # Get your key at https://hud.so
                         "Mcp-Image": "hudpython/hud-text-2048:v1.2"  # Docker image from https://hub.docker.com/u/hudpython
                     }
                 }
@@ -219,7 +239,7 @@ async def main() -> None:
 asyncio.run(main())
 ```
 
-The above example let's the agent play 2048 ([See replay](https://app.hud.so/trace/6feed7bd-5f67-4d66-b77f-eb1e3164604f))
+The above example let's the agent play 2048 ([See replay](https://hud.so/trace/6feed7bd-5f67-4d66-b77f-eb1e3164604f))
 
 ![Agent playing 2048](https://raw.githubusercontent.com/hud-evals/hud-python/main/docs/src/images/2048_1.gif)
 
@@ -250,7 +270,7 @@ Supports multi‑turn RL for both:
 - Language‑only models (e.g., `Qwen/Qwen2.5-7B-Instruct`)
 - Vision‑Language models (e.g., `Qwen/Qwen2.5-VL-3B-Instruct`)
 
-By default, `hud rl` provisions a persistant server and trainer in the cloud, streams telemetry to `app.hud.so`, and lets you monitor/manage models at `app.hud.so/models`. Use `--local` to run entirely on your machines (typically 2+ GPUs: one for vLLM, the rest for training).
+By default, `hud rl` provisions a persistant server and trainer in the cloud, streams telemetry to `hud.so`, and lets you monitor/manage models at `hud.so/models`. Use `--local` to run entirely on your machines (typically 2+ GPUs: one for vLLM, the rest for training).
 
 Any HUD MCP environment and evaluation works with our RL pipeline (including remote configurations). See the guided docs: `https://docs.hud.so/train-agents/quickstart`.
 
@@ -260,7 +280,7 @@ This is Claude Computer Use running on our proprietary financial analyst benchma
 
 ![Trace screenshot](https://raw.githubusercontent.com/hud-evals/hud-python/main/docs/src/images/trace_sheet.gif)
 
-> [See this trace on _app.hud.so_](https://app.hud.so/trace/9e212e9e-3627-4f1f-9eb5-c6d03c59070a)
+> [See this trace on _hud.so_](https://hud.so/trace/9e212e9e-3627-4f1f-9eb5-c6d03c59070a)
 
 This example runs the full dataset (only takes ~20 minutes) using [run_evaluation.py](examples/run_evaluation.py):
 
@@ -286,7 +306,7 @@ results = await run_dataset(
 print(f"Average reward: {sum(r.reward for r in results) / len(results):.2f}")
 ```
 
-> Running a dataset creates a job and streams results to the [app.hud.so](https://app.hud.so) platform for analysis and [leaderboard submission](https://docs.hud.so/evaluate-agents/leaderboards).
+> Running a dataset creates a job and streams results to the [hud.so](https://hud.so) platform for analysis and [leaderboard submission](https://docs.hud.so/evaluate-agents/leaderboards).
 
 ## Building Environments (MCP)
 
@@ -377,7 +397,7 @@ Tools
 hud push # needs docker login, hud api key
 ```
 
-5. Now you can use `mcp.hud.so` to launch 100s of instances of this environment in parallel with any agent, and see everything live on [app.hud.so](https://app.hud.so):
+5. Now you can use `mcp.hud.so` to launch 100s of instances of this environment in parallel with any agent, and see everything live on [hud.so](https://hud.so):
 
 ```python
 from hud.agents import ClaudeAgent
@@ -408,7 +428,7 @@ result = await ClaudeAgent().run({  # See all agents: https://docs.hud.so/refere
 
 ## Leaderboards & benchmarks
 
-All leaderboards are publicly available on [app.hud.so/leaderboards](https://app.hud.so/leaderboards) (see [docs](https://docs.hud.so/evaluate-agents/leaderboards))
+All leaderboards are publicly available on [hud.so/leaderboards](https://hud.so/leaderboards) (see [docs](https://docs.hud.so/evaluate-agents/leaderboards))
 
 ![Leaderboard](https://raw.githubusercontent.com/hud-evals/hud-python/main/docs/src/images/leaderboards_3.png)
 
@@ -422,7 +442,7 @@ Using the [`run_dataset`](https://docs.hud.so/reference/tasks#run_dataset) funct
 %%{init: {"theme": "neutral", "themeVariables": {"fontSize": "14px"}} }%%
 graph LR
     subgraph "Platform"
-        Dashboard["📊 app.hud.so"]
+        Dashboard["📊 hud.so"]
         API["🔌 mcp.hud.so"]
     end