hud-python 0.4.27__py3-none-any.whl → 0.4.29__py3-none-any.whl

hud/cli/rl/vllm.py

@@ -0,0 +1,177 @@
+"""vLLM server management utilities."""
+
+from __future__ import annotations
+
+import asyncio
+import logging
+import os
+import subprocess
+import time
+from pathlib import Path
+
+import httpx
+from rich.console import Console
+
+from hud.utils.hud_console import HUDConsole
+
+logger = logging.getLogger(__name__)
+hud_console = HUDConsole(logger)
+
+console = Console()
+
+
+def get_vllm_args(model_name: str, chat_template_path: Path | None = None) -> list[str]:
+    """Get common vLLM server arguments for both local and remote deployments."""
+    args = [
+        "serve",
+        model_name,
+        "--api-key",
+        "token-abc123",
+        "--host",
+        "0.0.0.0",  # noqa: S104
+        "--port",
+        "8000",
+        "--tensor-parallel-size",
+        "1",
+        "--trust-remote-code",
+        "--max-model-len",
+        "16384",
+        "--enable-lora",
+        "--max-lora-rank",
+        "64",
+        "--max-cpu-loras",
+        "4",
+        "--enable-auto-tool-choice",
+        "--tool-call-parser",
+        "hermes",
+        "--disable-log-requests",
+        "--dtype",
+        "auto",
+    ]
+
+    # Add chat template if provided
+    if chat_template_path and chat_template_path.exists():
+        args.extend(["--chat-template", str(chat_template_path.absolute())])
+
+    return args
+
+
+def check_vllm_server() -> bool:
+    """Check if vLLM server is running."""
+    try:
+        response = httpx.get("http://localhost:8000/health", timeout=2.0)
+        return response.status_code == 200
+    except Exception:
+        return False
+
+
+def kill_vllm_server() -> None:
+    """Kill any running vLLM server processes."""
+    try:
+        # Check for PID file first
+        pid_file = Path("/tmp/vllm_server.pid")  # noqa: S108
+        if pid_file.exists():
+            try:
+                pid = int(pid_file.read_text().strip())
+                subprocess.run(["kill", "-TERM", str(pid)], check=False)  # noqa: S603, S607
+                time.sleep(2)
+                # Force kill if still running
+                subprocess.run(["kill", "-9", str(pid)], check=False)  # noqa: S603, S607
+                pid_file.unlink()
+            except Exception as e:
+                hud_console.error(f"Failed to kill vLLM server: {e}")
+
+        # Also try to kill by process name
+        subprocess.run(["pkill", "-f", "vllm serve"], check=False)  # noqa: S607
+        subprocess.run(["pkill", "-f", "vllm.entrypoints.openai.api_server"], check=False)  # noqa: S607
+        time.sleep(2)
+
+        # Check for any process using port 8000
+        result = subprocess.run(["lsof", "-ti:8000"], capture_output=True, text=True, check=False)  # noqa: S607
+
+        if result.stdout.strip():
+            for pid in result.stdout.strip().split("\n"):
+                try:
+                    subprocess.run(["kill", "-9", pid], check=False)  # noqa: S603, S607
+                except Exception as e:
+                    hud_console.error(f"Failed to kill vLLM server: {e}")
+
+        console.print("[yellow]Killed existing vLLM server processes[/yellow]")
+    except Exception as e:
+        hud_console.error(f"Error killing vLLM server: {e}")
+
+
+def start_vllm_server(model_name: str, gpu_index: int = 1, restart: bool = False) -> None:
+    """Start vLLM server in the background with dynamic GPU selection."""
+    if restart:
+        kill_vllm_server()
+        time.sleep(3)
+
+    # Check if already running
+    if check_vllm_server():
+        console.print("[green]vLLM server is already running[/green]")
+        return
+
+    console.print(f"[cyan]Starting vLLM server with {model_name} on GPU {gpu_index}...[/cyan]")
+
+    # Set up environment variables
+    env = os.environ.copy()
+    env.update(
+        {
+            "CUDA_VISIBLE_DEVICES": str(gpu_index),
+            "VLLM_ALLOW_RUNTIME_LORA_UPDATING": "True",
+            "TOKENIZERS_PARALLELISM": "false",
+            "VLLM_LOGGING_LEVEL": "INFO",  # Changed from DEBUG to reduce noise
+            "CUDA_LAUNCH_BLOCKING": "1",  # Better error messages
+        }
+    )
+
+    # Get the path to chat template
+    chat_template_path = Path(__file__).parent.parent.parent / "rl" / "chat_template.jinja"
+
+    # Build the vLLM command
+    vllm_args = get_vllm_args(model_name, chat_template_path)
+    cmd = ["uv", "run", "vllm", *vllm_args]
+
+    # Start the server in the background
+    with open("/tmp/vllm_server.log", "w") as log_file:  # noqa: S108,
+        process = subprocess.Popen(  # noqa: S603
+            cmd,
+            env=env,
+            stdout=log_file,
+            stderr=subprocess.STDOUT,
+            preexec_fn=os.setpgrp,  # type: ignore
+            cwd=Path.cwd(),  # Use current working directory
+        )
+
+    console.print("[yellow]vLLM server starting in background...[/yellow]")
+    console.print(f"[yellow]Process ID: {process.pid}[/yellow]")
+    console.print("[yellow]Check logs at: /tmp/vllm_server.log[/yellow]")
+
+    # Save PID for later management
+    pid_file = Path("/tmp/vllm_server.pid")  # noqa: S108
+    pid_file.write_text(str(process.pid))
+
+
+async def wait_for_vllm_server(timeout: int = 360) -> bool:  # noqa: ASYNC109
+    """Wait for vLLM server to be ready."""
+    start_time = time.time()
+    console.print("[yellow]Waiting for vLLM server to be ready (up to 6 minutes)...[/yellow]")
+
+    async with httpx.AsyncClient() as client:
+        while time.time() - start_time < timeout:
+            try:
+                response = await client.get("http://localhost:8000/health", timeout=2.0)
+                if response.status_code == 200:
+                    console.print("[green]✅ vLLM server is ready![/green]")
+                    return True
+            except Exception as e:
+                hud_console.error(f"Failed to connect to vLLM server: {e}")
+
+            await asyncio.sleep(2)
+            elapsed = int(time.time() - start_time)
+            console.print(f"[yellow]Waiting... ({elapsed}s / {timeout}s)[/yellow]", end="\r")
+
+    console.print("\n[red]❌ vLLM server failed to start within timeout[/red]")
+    console.print("[yellow]Check /tmp/vllm_server.log for details[/yellow]")
+    return False

hud/cli/tests/test_analyze_metadata.py

@@ -214,7 +214,6 @@ class TestAnalyzeFromMetadata:
 
     @mock.patch("hud.cli.utils.metadata.check_local_cache")
     @mock.patch("hud.cli.utils.metadata.fetch_lock_from_registry")
-    @mock.patch("hud.cli.utils.metadata.design")
     @mock.patch("hud.cli.utils.metadata.console")
     async def test_analyze_not_found(self, mock_console, mock_hud_console, mock_fetch, mock_check):
         """Test when environment not found anywhere."""

hud/cli/utils/tasks.py

@@ -0,0 +1,26 @@
+from __future__ import annotations
+
+from pathlib import Path
+
+from hud.utils.hud_console import hud_console
+
+
+def find_tasks_file(tasks_file: str | None, msg: str = "Select a tasks file") -> str:
+    """Find tasks file."""
+    if tasks_file:
+        return tasks_file
+
+    # Get current directory and find all .json and .jsonl files
+    current_dir = Path.cwd()
+    all_files = list(current_dir.glob("*.json")) + list(current_dir.glob("*.jsonl"))
+    all_files = [
+        str(file).replace(str(current_dir), "").lstrip("/").lstrip("\\") for file in all_files
+    ]
+    all_files = [file for file in all_files if file[0] != "."]  # Remove all config files
+
+    if len(all_files) == 1:
+        return str(all_files[0])
+
+    else:
+        # Prompt user to select a file
+        return hud_console.select(msg, choices=all_files)

hud/clients/base.py

@@ -11,18 +11,16 @@ from mcp.types import Implementation
 
 from hud.shared.exceptions import HudAuthenticationError, HudException
 from hud.types import MCPToolCall, MCPToolResult
+from hud.utils.hud_console import HUDConsole
 from hud.utils.mcp import setup_hud_telemetry
 from hud.version import __version__ as hud_version
 
 if TYPE_CHECKING:
     import mcp.types as types
-
-else:
-    pass
-
-
 logger = logging.getLogger(__name__)
 
+hud_console = HUDConsole(logger=logger)
+
 
 @runtime_checkable
 class AgentMCPClient(Protocol):
@@ -113,7 +111,7 @@ class BaseHUDClient(AgentMCPClient):
     async def initialize(self, mcp_config: dict[str, dict[str, Any]] | None = None) -> None:
         """Initialize connection and fetch tools."""
         if self._initialized:
-            logger.warning(
+            hud_console.warning(
                 "Client already connected, if you want to reconnect or change the configuration, "
                 "call shutdown() first. This is especially important if you are using an agent."
             )
@@ -130,7 +128,7 @@ class BaseHUDClient(AgentMCPClient):
 
         self._auto_trace_cm = setup_hud_telemetry(self._mcp_config, auto_trace=self._auto_trace)
 
-        logger.debug("Initializing MCP client...")
+        hud_console.debug("Initializing MCP client...")
 
         try:
             # Check if API key is set for HUD API
@@ -155,7 +153,6 @@ class BaseHUDClient(AgentMCPClient):
         await self._fetch_telemetry()
 
         self._initialized = True
-        logger.info("Client initialized")
 
     async def shutdown(self) -> None:
         """Disconnect from the MCP server."""
@@ -163,9 +160,9 @@ class BaseHUDClient(AgentMCPClient):
         if self._auto_trace_cm:
             try:
                 self._auto_trace_cm.__exit__(None, None, None)
-                logger.info("Closed auto-created trace")
+                hud_console.info("Closed auto-created trace")
             except Exception as e:
-                logger.warning("Failed to close auto-created trace: %s", e)
+                hud_console.warning(f"Failed to close auto-created trace: {e}")
             finally:
                 self._auto_trace_cm = None
 
@@ -173,9 +170,9 @@ class BaseHUDClient(AgentMCPClient):
         if self._initialized:
             await self._disconnect()
             self._initialized = False
-            logger.info("Client disconnected")
+            hud_console.info("Shutdown completed")
         else:
-            logger.debug("Client was not initialized, skipping disconnect")
+            hud_console.debug("Client was not initialized, skipping disconnect")
 
     @overload
     async def call_tool(self, tool_call: MCPToolCall, /) -> MCPToolResult: ...
@@ -280,27 +277,28 @@ class BaseHUDClient(AgentMCPClient):
                 telemetry_data = json.loads(result.contents[0].text)  # type: ignore
                 self._telemetry_data = telemetry_data
 
-                logger.info("📡 Telemetry data fetched:")
                 if "live_url" in telemetry_data:
-                    logger.info("   🖥️  Live URL: %s", telemetry_data["live_url"])
+                    hud_console.info(f"   🖥️  Live URL: {telemetry_data['live_url']}")
                 if "vnc_url" in telemetry_data:
-                    logger.info("   🖥️  VNC URL: %s", telemetry_data["vnc_url"])
+                    hud_console.info(f"   🖥️  VNC URL: {telemetry_data['vnc_url']}")
                 if "cdp_url" in telemetry_data:
-                    logger.info("   🦾  CDP URL: %s", telemetry_data["cdp_url"])
+                    hud_console.info(f"   🦾  CDP URL: {telemetry_data['cdp_url']}")
                 if "status" in telemetry_data:
-                    logger.info("   📊 Status: %s", telemetry_data["status"])
+                    hud_console.debug(f"   📊 Status: {telemetry_data['status']}")
                 if "services" in telemetry_data:
-                    logger.debug("   📋 Services:")
+                    hud_console.debug("   📋 Services:")
                     for service, status in telemetry_data["services"].items():
                         status_icon = "✅" if status == "running" else "❌"
-                        logger.debug("      %s %s: %s", status_icon, service, status)
+                        hud_console.debug(f"      {status_icon} {service}: {status}")
 
                 if self.verbose:
-                    logger.debug("Full telemetry data:\n%s", json.dumps(telemetry_data, indent=2))
+                    hud_console.debug(
+                        f"Full telemetry data:\n{json.dumps(telemetry_data, indent=2)}"
+                    )
         except Exception as e:
             # Telemetry is optional
             if self.verbose:
-                logger.debug("No telemetry available: %s", e)
+                hud_console.debug(f"No telemetry available: {e}")
 
     async def analyze_environment(self) -> dict[str, Any]:
         """Complete analysis of the MCP environment.
@@ -363,7 +361,7 @@ class BaseHUDClient(AgentMCPClient):
                 analysis["resources"].append(resource_info)
         except Exception as e:
             if self.verbose:
-                logger.debug("Could not list resources: %s", e)
+                hud_console.debug(f"Could not list resources: {e}")
 
         return analysis
 
@@ -387,5 +385,5 @@ class BaseHUDClient(AgentMCPClient):
                 return functions
         except Exception as e:
             if self.verbose:
-                logger.debug("Could not read hub functions for '%s': %s", hub_name, e)
+                hud_console.debug(f"Could not read hub functions for '{hub_name}': {e}")
         return []

hud/clients/mcp_use.py

@@ -3,6 +3,7 @@
 from __future__ import annotations
 
 import logging
+import traceback
 from typing import Any
 
 from mcp import Implementation, types
@@ -12,12 +13,14 @@ from mcp_use.session import MCPSession as MCPUseSession
 from pydantic import AnyUrl
 
 from hud.types import MCPToolCall, MCPToolResult
+from hud.utils.hud_console import HUDConsole
 from hud.version import __version__ as hud_version
 
 from .base import BaseHUDClient
 from .utils.mcp_use_retry import patch_all_sessions
 
 logger = logging.getLogger(__name__)
+hud_console = HUDConsole(logger=logger)
 
 
 class MCPUseHUDClient(BaseHUDClient):
@@ -62,11 +65,11 @@ class MCPUseHUDClient(BaseHUDClient):
         try:
             assert self._client is not None  # noqa: S101
             self._sessions = await self._client.create_all_sessions()
-            logger.info("Created %d MCP sessions", len(self._sessions))
+            hud_console.info(f"Created {len(self._sessions)} MCP sessions")
 
             # Patch all sessions with retry logic
             patch_all_sessions(self._sessions)
-            logger.debug("Applied retry logic to all MCP sessions")
+            hud_console.debug("Applied retry logic to all MCP sessions")
 
             # Configure validation for all sessions based on client setting
             try:
@@ -86,21 +89,21 @@ class MCPUseHUDClient(BaseHUDClient):
             # Log session details in verbose mode
             if self.verbose and self._sessions:
                 for name, session in self._sessions.items():
-                    logger.debug("  - %s: %s", name, type(session).__name__)
+                    hud_console.debug(f"  - {name}: {type(session).__name__}")
 
         except McpError as e:
             # Protocol error - the server is reachable but rejecting our request
-            logger.error("MCP protocol error: %s", e)
-            logger.error("This typically means:")
-            logger.error("- Invalid or missing initialization parameters")
-            logger.error("- Incompatible protocol version")
-            logger.error("- Server-side configuration issues")
+            hud_console.warning(f"MCP protocol error: {e}")
+            hud_console.warning("This typically means:")
+            hud_console.warning("- Invalid or missing initialization parameters")
+            hud_console.warning("- Incompatible protocol version")
+            hud_console.warning("- Server-side configuration issues")
             raise
         except Exception as e:
             # Transport or other errors
-            logger.error("Failed to create sessions: %s", e)
+            hud_console.error(f"Failed to create sessions: {e}")
             if self.verbose:
-                logger.info("Check that the MCP server is running and accessible")
+                hud_console.info("Check that the MCP server is running and accessible")
             raise
 
         # Populate tool map during initialization
@@ -129,17 +132,14 @@ class MCPUseHUDClient(BaseHUDClient):
                     await session.initialize()
 
                 if session.connector.client_session is None:
-                    logger.warning("Client session not initialized for %s", server_name)
+                    hud_console.warning(f"Client session not initialized for {server_name}")
                     continue
 
                 # List tools (retry logic is handled at transport level)
                 tools_result = await session.connector.client_session.list_tools()
 
-                logger.info(
-                    "Discovered %d tools from '%s': %s",
-                    len(tools_result.tools),
-                    server_name,
-                    [tool.name for tool in tools_result.tools],
+                hud_console.info(
+                    f"Discovered {len(tools_result.tools)} tools from '{server_name}': {', '.join([tool.name for tool in tools_result.tools])}",  # noqa: E501
                 )
 
                 # Add to collections with optional prefix
@@ -167,16 +167,15 @@ class MCPUseHUDClient(BaseHUDClient):
                 if self.verbose:
                     for tool in tools_result.tools:
                         description = tool.description or ""
-                        logger.debug(
-                            "  Tool '%s': %s",
-                            tool.name,
-                            description[:100] + "..." if len(description) > 100 else description,
+                        hud_console.debug(
+                            f"  Tool '{tool.name}': {description[:100] + '...' if len(description) > 100 else description}",  # noqa: E501
                         )
 
             except Exception as e:
-                logger.error("Error discovering tools from '%s': %s", server_name, e)
+                hud_console.error(f"Error discovering tools from '{server_name}': {e}")
                 if self.verbose:
-                    logger.exception("Full error details:")
+                    hud_console.error("Full error details:")
+                    traceback.print_exc()
 
         return all_tools
 
@@ -196,12 +195,8 @@ class MCPUseHUDClient(BaseHUDClient):
         session = self._sessions[server_name]
 
         if self.verbose:
-            logger.debug(
-                "Calling tool '%s' (original: '%s') on server '%s' with arguments: %s",
-                tool_call.name,
-                original_tool.name,
-                server_name,
-                tool_call.arguments,
+            hud_console.debug(
+                f"Calling tool '{tool_call.name}' (original: '{original_tool.name}') on server '{server_name}' with arguments: {tool_call.arguments}"  # noqa: E501
             )
 
         if session.connector.client_session is None:
@@ -214,7 +209,7 @@ class MCPUseHUDClient(BaseHUDClient):
         )
 
         if self.verbose:
-            logger.debug("Tool '%s' result: %s", tool_call.name, result)
+            hud_console.debug(f"Tool '{tool_call.name}' result: {result}")
 
         # MCP-use already returns the correct type, but we need to ensure it's MCPToolResult
         return MCPToolResult(
@@ -246,7 +241,7 @@ class MCPUseHUDClient(BaseHUDClient):
                 return resources.resources
             except Exception as e:
                 if self.verbose:
-                    logger.debug("Could not list resources from server '%s': %s", server_name, e)
+                    hud_console.debug(f"Could not list resources from server '{server_name}': {e}")
                 continue
         return []
 
@@ -276,8 +271,8 @@ class MCPUseHUDClient(BaseHUDClient):
                     raise AttributeError("read_resource not available")
 
                 if self.verbose:
-                    logger.debug(
-                        "Successfully read resource '%s' from server '%s'", uri, server_name
+                    hud_console.debug(
+                        f"Successfully read resource '{uri}' from server '{server_name}'"
                     )
 
                 return result
@@ -285,24 +280,21 @@ class MCPUseHUDClient(BaseHUDClient):
             except McpError as e:
                 # McpError is expected for unsupported resources
                 if "telemetry://" in str(uri):
-                    logger.debug(
-                        "Telemetry resource not supported by server '%s': %s", server_name, e
+                    hud_console.debug(
+                        f"Telemetry resource not supported by server '{server_name}': {e}"
                     )
                 elif self.verbose:
-                    logger.debug(
-                        "MCP resource error for '%s' from server '%s': %s", uri, server_name, e
+                    hud_console.debug(
+                        f"MCP resource error for '{uri}' from server '{server_name}': {e}"
                     )
                 continue
             except Exception as e:
                 # Other errors might be more serious
                 if "telemetry://" in str(uri):
-                    logger.debug("Failed to fetch telemetry from server '%s': %s", server_name, e)
+                    hud_console.debug(f"Failed to fetch telemetry from server '{server_name}': {e}")
                 else:
-                    logger.warning(
-                        "Unexpected error reading resource '%s' from server '%s': %s",
-                        uri,
-                        server_name,
-                        e,
+                    hud_console.warning(
+                        f"Unexpected error reading resource '{uri}' from server '{server_name}': {e}"  # noqa: E501
                     )
                 continue
 
@@ -311,14 +303,14 @@ class MCPUseHUDClient(BaseHUDClient):
     async def _disconnect(self) -> None:
         """Close all active sessions."""
         if self._client is None:
-            logger.warning("Client is not connected, cannot close")
+            hud_console.warning("Client is not connected, cannot close")
             return
 
         await self._client.close_all_sessions()
         self._sessions = {}
         self._tool_map = {}
         self._initialized = False
-        logger.debug("MCP-use client disconnected")
+        hud_console.debug("MCP-use client disconnected")
 
     # Legacy compatibility methods (limited; tests should not rely on these)
     def get_sessions(self) -> dict[str, Any]:

hud/clients/tests/test_mcp_use_retry.py

@@ -36,20 +36,20 @@ class TestRetrySession:
 
         # Check adapter configuration
         adapter = session.adapters["http://"]
-        assert adapter.max_retries.total == 5
-        assert 500 in adapter.max_retries.status_forcelist
-        assert 502 in adapter.max_retries.status_forcelist
-        assert adapter.max_retries.backoff_factor == 2.0
+        assert hasattr(adapter, "max_retries") and adapter.max_retries.total == 5  # type: ignore
+        assert 500 in adapter.max_retries.status_forcelist  # type: ignore
+        assert 502 in adapter.max_retries.status_forcelist  # type: ignore
+        assert adapter.max_retries.backoff_factor == 2.0  # type: ignore
 
     def test_retry_session_default_values(self):
         """Test retry session with default values."""
         session = create_retry_session()
 
         adapter = session.adapters["https://"]
-        assert adapter.max_retries.total == 3
-        assert 502 in adapter.max_retries.status_forcelist
-        assert 503 in adapter.max_retries.status_forcelist
-        assert 504 in adapter.max_retries.status_forcelist
+        assert adapter.max_retries.total == 3  # type: ignore
+        assert 502 in adapter.max_retries.status_forcelist  # type: ignore
+        assert 503 in adapter.max_retries.status_forcelist  # type: ignore
+        assert 504 in adapter.max_retries.status_forcelist  # type: ignore
 
 
 class TestAsyncRetryWrapper:
@@ -316,7 +316,7 @@ class TestMCPUseClientRetry:
             # Verify retry worked
             assert call_count == 2  # Failed once, then succeeded
             assert not result.isError
-            assert result.content[0].text == "Success"
+            assert result.content[0].text == "Success"  # type: ignore
 
     @pytest.mark.asyncio
     async def test_resource_read_with_retry(self):
@@ -371,7 +371,7 @@ class TestMCPUseClientRetry:
             # Verify retry worked
             assert call_count == 2  # Failed once, then succeeded
             assert result is not None
-            assert result.contents[0].text == '{"status": "ok"}'
+            assert result.contents[0].text == '{"status": "ok"}'  # type: ignore
 
 
 if __name__ == "__main__":

hud/datasets/__init__.py

@@ -7,13 +7,14 @@ Provides data models, utilities, and execution functions for working with HUD da
 # Execution functions
 from __future__ import annotations
 
-from .execution import (
+from hud.types import Task
+
+from .parallel import (
     calculate_optimal_workers,
-    run_dataset,
     run_dataset_parallel,
     run_dataset_parallel_manual,
 )
-from .task import Task
+from .runner import run_dataset
 
 # Utilities
 from .utils import fetch_system_prompt_from_dataset, save_tasks

hud/datasets/{execution/parallel.py → parallel.py}

@@ -65,8 +65,8 @@ def _process_worker(
 
     import hud
     from hud.agents.misc.response_agent import ResponseAgent
-    from hud.datasets.task import Task
     from hud.otel import configure_telemetry
+    from hud.types import Task
 
     # Ensure stdout is not buffered for immediate output
     try:

hud/datasets/{execution/runner.py → runner.py}

@@ -9,7 +9,7 @@ from typing import TYPE_CHECKING, Any, cast
 from datasets import Dataset, load_dataset
 
 from hud.agents.misc import ResponseAgent
-from hud.datasets.task import Task
+from hud.types import Task
 
 if TYPE_CHECKING:
     from hud.agents import MCPAgent

hud/datasets/utils.py

@@ -8,7 +8,7 @@ from typing import Any
 
 from datasets import Dataset
 
-from .task import Task
+from hud.types import Task
 
 logger = logging.getLogger("hud.datasets")
 

hud/native/tests/test_native_init.py

@@ -41,7 +41,7 @@ class TestNativeInit:
         assert "compare" in tool_names
 
         # Should have the submit tool
-        assert "submit" in tool_names
+        assert "response" in tool_names
 
         # Should have all the alias tools
         expected_aliases = [

hud/otel/config.py

@@ -111,7 +111,7 @@ def configure_telemetry(
         # Error if no exporters are configured
         raise ValueError(
             "No telemetry backend configured. Either:\n"
-            "1. Set HUD_API_KEY environment variable for HUD telemetry\n"
+            "1. Set HUD_API_KEY environment variable for HUD telemetry (https://app.hud.so)\n"
             "2. Use enable_otlp=True with configure_telemetry() for alternative backends (e.g., Jaeger)\n"  # noqa: E501
         )
     elif not settings.telemetry_enabled: