hud-python 0.4.53__tar.gz → 0.4.54__tar.gz

{hud_python-0.4.53 → hud_python-0.4.54}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: hud-python
-Version: 0.4.53
+Version: 0.4.54
 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
@@ -42,6 +42,7 @@ 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==2.3.20
+Requires-Dist: langchain==0.3.27
 Requires-Dist: numpy>=1.24.0
 Requires-Dist: openai
 Requires-Dist: opentelemetry-api>=1.34.1
@@ -247,8 +248,8 @@ The above example let's the agent play 2048 ([See replay](https://hud.so/trace/6
 RL using GRPO a Qwen2.5-VL model on any hud dataset:
 
 ```bash
-hud get hud-evals/basic-2048 # from HF
-hud rl basic-2048.json
+hud get hud-evals/2048-basic # from HF
+hud rl 2048-basic.json
 ```
 
 > See [agent training docs](https://docs.hud.so/train-agents/quickstart)
@@ -439,14 +440,14 @@ Train with the new interactive `hud rl` flow:
 uv tool install hud-python
 
 # Option A: Run directly from a HuggingFace dataset
-hud rl hud-evals/basic-2048
+hud rl hud-evals/2048-basic
 
 # Option B: Download first, modify, then train
-hud get hud-evals/basic-2048
-hud rl basic-2048.json
+hud get hud-evals/2048-basic
+hud rl 2048-basic.json
 
 # Optional: baseline evaluation
-hud eval basic-2048.json
+hud eval 2048-basic.json
 ```
 
 Supports multi‑turn RL for both:

{hud_python-0.4.53 → hud_python-0.4.54}/README.md

@@ -109,8 +109,8 @@ The above example let's the agent play 2048 ([See replay](https://hud.so/trace/6
 RL using GRPO a Qwen2.5-VL model on any hud dataset:
 
 ```bash
-hud get hud-evals/basic-2048 # from HF
-hud rl basic-2048.json
+hud get hud-evals/2048-basic # from HF
+hud rl 2048-basic.json
 ```
 
 > See [agent training docs](https://docs.hud.so/train-agents/quickstart)
@@ -301,14 +301,14 @@ Train with the new interactive `hud rl` flow:
 uv tool install hud-python
 
 # Option A: Run directly from a HuggingFace dataset
-hud rl hud-evals/basic-2048
+hud rl hud-evals/2048-basic
 
 # Option B: Download first, modify, then train
-hud get hud-evals/basic-2048
-hud rl basic-2048.json
+hud get hud-evals/2048-basic
+hud rl 2048-basic.json
 
 # Optional: baseline evaluation
-hud eval basic-2048.json
+hud eval 2048-basic.json
 ```
 
 Supports multi‑turn RL for both:

{hud_python-0.4.53 → hud_python-0.4.54}/environments/README.md

@@ -804,9 +804,9 @@ class TodoCompleted:
 @problem("todo_basic", description="Complete two todo items", difficulty="easy")
 class TodoBasic:
     def get_setup(self):
-        return {"function": "todo_seed", "args": {"num_items": 5}}
+        return {"name": "todo_seed", "arguments": {"num_items": 5}}
     def get_evaluation(self):
-        return {"function": "todo_completed", "args": {"expected_count": 2}}
+        return {"name": "todo_completed", "arguments": {"expected_count": 2}}
 ```
 
 Decorators keep registration *next to the implementation* and avoid manual bookkeeping.  The server simply exposes the combined metadata through an MCP **resource**.  Follow `environments/browser/src/hud_controller/problems/registry.py` as a template and expose the JSON with `@mcp.resource("problems://registry")`.

{hud_python-0.4.53 → hud_python-0.4.54}/environments/blank/server/pyproject.toml

@@ -4,7 +4,7 @@ version = "0.1.0"
 description = "MCP server for blank environment"
 requires-python = ">=3.11"
 dependencies = [
-    "hud-python>=0.4.53",
+    "hud-python>=0.4.54",
     "httpx>=0.28.1",
 ]
 

{hud_python-0.4.53 → hud_python-0.4.54}/environments/browser/environment/todo/README.md

@@ -47,8 +47,8 @@ await setup({"name": "todo_basic_usage"})
 await evaluate({"name": "todo_basic_usage"})
 
 # Direct function calls
-await setup({"function": "todo_reset", "args": {}})
-await evaluate({"function": "todo_completion_rate", "args": {"min_rate": 0.5}})
+await setup({"name": "todo_reset", "arguments": {}})
+await evaluate({"name": "todo_completion_rate", "arguments": {"min_rate": 0.5}})
 
 # MCP resource discovery
 todo_evaluators = await client.read_resource("evaluators://todo")

{hud_python-0.4.53 → hud_python-0.4.54}/environments/browser/server/pyproject.toml

@@ -4,7 +4,7 @@ version = "0.1.0"
 description = "HUD Browser MCP Server"
 requires-python = ">=3.11,<3.14"
 dependencies = [
-    "hud-python>=0.4.53",
+    "hud-python>=0.4.54",
     "httpx",
     "playwright",
     "pyautogui",

{hud_python-0.4.53 → hud_python-0.4.54}/environments/deepresearch/server/pyproject.toml

@@ -4,7 +4,7 @@ version = "0.1.0"
 description = "MCP server for DeepResearch environment"
 requires-python = ">=3.11"
 dependencies = [
-    "hud-python>=0.4.53",
+    "hud-python>=0.4.54",
     "httpx>=0.24.0",
 ]
 

{hud_python-0.4.53 → hud_python-0.4.54}/hud/cli/__init__.py

@@ -12,6 +12,8 @@ from rich.console import Console
 from rich.panel import Panel
 from rich.table import Table
 
+from hud.types import AgentType
+
 from . import list_func as list_module
 from .analyze import (
     analyze_environment,
@@ -847,7 +849,7 @@ def eval(
     hud_console = HUDConsole()
 
     if integration_test:
-        agent = "integration_test"
+        agent = AgentType.INTEGRATION_TEST
 
     # If no source provided, reuse RL helper to find a tasks file interactively
     if source is None:
@@ -894,17 +896,17 @@ def eval(
         # Add standard agent choices
         choices.extend(
             [
-                {"name": "Claude 4 Sonnet", "value": "claude"},
-                {"name": "OpenAI Computer Use", "value": "openai"},
-                {"name": "vLLM (Local Server)", "value": "vllm"},
-                {"name": "LiteLLM (Multi-provider)", "value": "litellm"},
+                {"name": "Claude 4 Sonnet", "value": AgentType.CLAUDE},
+                {"name": "OpenAI Computer Use", "value": AgentType.OPENAI},
+                {"name": "vLLM (Local Server)", "value": AgentType.VLLM},
+                {"name": "LiteLLM (Multi-provider)", "value": AgentType.LITELLM},
             ]
         )
 
         agent = hud_console.select("Select an agent to use:", choices=choices, default=0)
 
     # Handle HUD model selection
-    if agent and agent not in ["claude", "openai", "vllm", "litellm", "integration_test"]:
+    if agent and agent not in [e.value for e in AgentType]:
         # Find remote model name
         model = agent
         if not vllm_base_url:
@@ -921,20 +923,23 @@ def eval(
             hud_console.error(f"Model {model} not found")
             raise typer.Exit(1)
         model = base_model
-        agent = "vllm"  # Use vLLM backend for HUD models
+        agent = AgentType.VLLM  # Use vLLM backend for HUD models
         hud_console.info(f"Using HUD model: {model} (trained on {base_model})")
 
     # Validate agent choice
-    valid_agents = ["claude", "openai", "vllm", "litellm", "integration_test"]
+    valid_agents = [e.value for e in AgentType]
     if agent not in valid_agents:
         hud_console.error(f"Invalid agent: {agent}. Must be one of: {', '.join(valid_agents)}")
         raise typer.Exit(1)
 
+    # Type narrowing: agent is now guaranteed to be an AgentType value after validation
+    agent = AgentType(agent)
+
     # Run the command
     eval_command(
         source=source,
         full=full,
-        agent=agent,  # type: ignore
+        agent=agent,
         model=model,
         allowed_tools=allowed_tools,
         max_concurrent=max_concurrent,

{hud_python-0.4.53 → hud_python-0.4.54}/hud/cli/dev.py

@@ -238,9 +238,9 @@ async def run_mcp_module(
         if env_dir.exists() and (env_dir / "server.py").exists():
             hud_console.info("")
             hud_console.info(
-                f"{hud_console.sym.FLOW} Don't forget to start the environment backend:"
+                f"{hud_console.sym.FLOW} Don't forget to start the environment backend in another terminal:"
             )
-            hud_console.info("   cd ../environment && uvicorn server:app --reload")
+            hud_console.info("   cd environment && uv run python uvicorn server:app --reload")
 
         # Launch inspector if requested (first run only)
         if inspector and transport == "http":

{hud_python-0.4.53 → hud_python-0.4.54}/hud/cli/eval.py

@@ -5,13 +5,14 @@ from __future__ import annotations
 import asyncio
 import logging
 from pathlib import Path
-from typing import TYPE_CHECKING, Any, Literal
+from typing import TYPE_CHECKING, Any
 
 import typer
 
 import hud
 from hud.cli.utils.env_check import ensure_built, find_environment_dir
 from hud.settings import settings
+from hud.types import AgentType
 from hud.utils.group_eval import display_group_statistics, run_tasks_grouped
 from hud.utils.hud_console import HUDConsole
 
@@ -113,7 +114,7 @@ def _build_vllm_config(
 
 
 def build_agent(
-    agent_type: Literal["claude", "openai", "vllm", "litellm", "integration_test"],
+    agent_type: AgentType,
     *,
     model: str | None = None,
     allowed_tools: list[str] | None = None,
@@ -123,11 +124,11 @@ def build_agent(
     """Create and return the requested agent type."""
 
     # Import agents lazily to avoid dependency issues
-    if agent_type == "integration_test":
+    if agent_type == AgentType.INTEGRATION_TEST:
         from hud.agents.misc.integration_test_agent import IntegrationTestRunner
 
         return IntegrationTestRunner(verbose=verbose)
-    elif agent_type == "vllm":
+    elif agent_type == AgentType.VLLM:
         # Create a generic OpenAI agent for vLLM server
         try:
             from hud.agents.openai_chat_generic import GenericOpenAIChatAgent
@@ -147,7 +148,7 @@ def build_agent(
         )
         return GenericOpenAIChatAgent(**config)
 
-    elif agent_type == "openai":
+    elif agent_type == AgentType.OPENAI:
         try:
             from hud.agents import OperatorAgent
         except ImportError as e:
@@ -165,7 +166,7 @@ def build_agent(
         else:
             return OperatorAgent(verbose=verbose)
 
-    elif agent_type == "litellm":
+    elif agent_type == AgentType.LITELLM:
         try:
             from hud.agents.lite_llm import LiteAgent
         except ImportError as e:
@@ -209,7 +210,7 @@ def build_agent(
 async def run_single_task(
     source: str,
     *,
-    agent_type: Literal["claude", "openai", "vllm", "litellm", "integration_test"] = "claude",
+    agent_type: AgentType = AgentType.CLAUDE,
     model: str | None = None,
     allowed_tools: list[str] | None = None,
     max_steps: int = 10,
@@ -268,14 +269,14 @@ async def run_single_task(
 
     # Use grouped evaluation if group_size > 1
     agent_config: dict[str, Any] = {}
-    if agent_type == "integration_test":
+    if agent_type == AgentType.INTEGRATION_TEST:
         from hud.agents.misc.integration_test_agent import IntegrationTestRunner
 
         agent_class = IntegrationTestRunner
         agent_config = {"verbose": verbose}
         if allowed_tools:
             agent_config["allowed_tools"] = allowed_tools
-    elif agent_type == "vllm":
+    elif agent_type == AgentType.VLLM:
         # Special handling for vLLM
         from hud.agents.openai_chat_generic import GenericOpenAIChatAgent
 
@@ -288,14 +289,14 @@ async def run_single_task(
             allowed_tools=allowed_tools,
             verbose=verbose,
         )
-    elif agent_type == "openai":
+    elif agent_type == AgentType.OPENAI:
         from hud.agents import OperatorAgent
 
         agent_class = OperatorAgent
         agent_config = {"verbose": verbose}
         if allowed_tools:
             agent_config["allowed_tools"] = allowed_tools
-    elif agent_type == "litellm":
+    elif agent_type == AgentType.LITELLM:
         from hud.agents.lite_llm import LiteAgent
 
         agent_class = LiteAgent
@@ -305,7 +306,7 @@ async def run_single_task(
         }
         if allowed_tools:
             agent_config["allowed_tools"] = allowed_tools
-    elif agent_type == "claude":
+    elif agent_type == AgentType.CLAUDE:
         from hud.agents import ClaudeAgent
 
         agent_class = ClaudeAgent
@@ -353,7 +354,7 @@ async def run_single_task(
 async def run_full_dataset(
     source: str,
     *,
-    agent_type: Literal["claude", "openai", "vllm", "litellm", "integration_test"] = "claude",
+    agent_type: AgentType = AgentType.CLAUDE,
     model: str | None = None,
     allowed_tools: list[str] | None = None,
     max_concurrent: int = 30,
@@ -395,12 +396,12 @@ async def run_full_dataset(
 
     # Build agent class + config for run_dataset
     agent_config: dict[str, Any]
-    if agent_type == "integration_test":  # --integration-test mode
+    if agent_type == AgentType.INTEGRATION_TEST:  # --integration-test mode
         from hud.agents.misc.integration_test_agent import IntegrationTestRunner
 
         agent_class = IntegrationTestRunner
         agent_config = {"verbose": verbose}
-    elif agent_type == "vllm":
+    elif agent_type == AgentType.VLLM:
         try:
             from hud.agents.openai_chat_generic import GenericOpenAIChatAgent
 
@@ -419,7 +420,7 @@ async def run_full_dataset(
             allowed_tools=allowed_tools,
             verbose=verbose,
         )
-    elif agent_type == "openai":
+    elif agent_type == AgentType.OPENAI:
         try:
             from hud.agents import OperatorAgent
 
@@ -435,7 +436,7 @@ async def run_full_dataset(
         if allowed_tools:
             agent_config["allowed_tools"] = allowed_tools
 
-    elif agent_type == "litellm":
+    elif agent_type == AgentType.LITELLM:
         try:
             from hud.agents.lite_llm import LiteAgent
 
@@ -539,8 +540,8 @@ def eval_command(
         "--full",
         help="Run the entire dataset (omit for single-task debug mode)",
     ),
-    agent: Literal["claude", "openai", "vllm", "litellm", "integration_test"] = typer.Option(
-        "claude",
+    agent: AgentType = typer.Option(  # noqa: B008
+        AgentType.CLAUDE,
         "--agent",
         help="Agent backend to use (claude, openai, vllm for local server, or litellm)",
     ),
@@ -648,21 +649,21 @@ def eval_command(
 
     # We pass integration_test as the agent_type
     if integration_test:
-        agent = "integration_test"
+        agent = AgentType.INTEGRATION_TEST
 
     # Check for required API keys
-    if agent == "claude":
+    if agent == AgentType.CLAUDE:
         if not settings.anthropic_api_key:
             hud_console.error("ANTHROPIC_API_KEY is required for Claude agent")
             hud_console.info(
                 "Set it in your environment or run: hud set ANTHROPIC_API_KEY=your-key-here"
             )
             raise typer.Exit(1)
-    elif agent == "openai" and not settings.openai_api_key:
+    elif agent == AgentType.OPENAI and not settings.openai_api_key:
         hud_console.error("OPENAI_API_KEY is required for OpenAI agent")
         hud_console.info("Set it in your environment or run: hud set OPENAI_API_KEY=your-key-here")
         raise typer.Exit(1)
-    elif agent == "vllm":
+    elif agent == AgentType.VLLM:
         if model:
             hud_console.info(f"Using vLLM with model: {model}")
         else:

{hud_python-0.4.53 → hud_python-0.4.54}/hud/cli/tests/test_eval.py

@@ -11,7 +11,7 @@ from hud.cli.eval import (
     build_agent,
     run_single_task,
 )
-from hud.types import Task, Trace
+from hud.types import AgentType, Task, Trace
 
 
 class TestBuildAgent:
@@ -26,7 +26,7 @@ class TestBuildAgent:
             mock_runner.return_value = mock_instance
 
             # Test with verbose=False
-            result = build_agent("integration_test", verbose=False)
+            result = build_agent(AgentType.INTEGRATION_TEST, verbose=False)
 
             mock_runner.assert_called_once_with(verbose=False)
             assert result == mock_instance
@@ -40,7 +40,7 @@ class TestBuildAgent:
             mock_runner.return_value = mock_instance
 
             # Test with verbose=False
-            result = build_agent("claude", verbose=False)
+            result = build_agent(AgentType.CLAUDE, verbose=False)
 
             mock_runner.assert_called_once_with(model="claude-sonnet-4-20250514", verbose=False)
             assert result == mock_instance
@@ -55,7 +55,7 @@ class TestBuildAgent:
 
             # Test with verbose=False
             result = build_agent(
-                "claude",
+                AgentType.CLAUDE,
                 model="claude-sonnet-4-20250514",
                 allowed_tools=["act"],
                 verbose=True,
@@ -97,7 +97,7 @@ class TestRunSingleTask:
             patch("hud.cli.eval.find_environment_dir", return_value=None),
             patch("hud.cli.eval.hud.trace"),
         ):
-            await run_single_task("test.json", agent_type="integration_test", max_steps=10)
+            await run_single_task("test.json", agent_type=AgentType.INTEGRATION_TEST, max_steps=10)
 
             # Verify agent.run was called with the task containing agent_config
             mock_agent.run.assert_called_once()
@@ -119,7 +119,7 @@ class TestRunSingleTask:
             mock_grouped.return_value = [{"task": mock_task, "rewards": [1.0, 0.5]}]
 
             await run_single_task(
-                "test.json", agent_type="integration_test", group_size=3, max_steps=10
+                "test.json", agent_type=AgentType.INTEGRATION_TEST, group_size=3, max_steps=10
             )
 
             # Verify run_tasks_grouped was called with correct group_size

{hud_python-0.4.53 → hud_python-0.4.54}/hud/types.py

@@ -5,6 +5,7 @@ import json
 import logging
 import uuid
 from collections import defaultdict
+from enum import Enum
 from string import Template
 from typing import Any, Literal
 
@@ -21,6 +22,14 @@ logger = logging.getLogger(__name__)
 _missing_api_key_error_logged: bool = False
 
 
+class AgentType(str, Enum):
+    CLAUDE = "claude"
+    OPENAI = "openai"
+    VLLM = "vllm"
+    LITELLM = "litellm"
+    INTEGRATION_TEST = "integration_test"
+
+
 class Task(BaseModel):
     """
     A task configuration that can be used to create a task.
@@ -325,6 +334,7 @@ class Trace(BaseModel):
 
 __all__ = [
     "AgentResponse",
+    "AgentType",
     "MCPToolCall",
     "MCPToolResult",
     "Trace",

{hud_python-0.4.53 → hud_python-0.4.54}/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__ == "0.4.53"
+    assert hud.__version__ == "0.4.54"

{hud_python-0.4.53 → hud_python-0.4.54}/hud/version.py

@@ -4,4 +4,4 @@ Version information for the HUD SDK.
 
 from __future__ import annotations
 
-__version__ = "0.4.53"
+__version__ = "0.4.54"

{hud_python-0.4.53 → hud_python-0.4.54}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "hud-python"
-version = "0.4.53"
+version = "0.4.54"
 description = "SDK for the HUD platform."
 readme = "README.md"
 requires-python = ">=3.11, <3.13"
@@ -18,6 +18,7 @@ dependencies = [
     "hud-mcp-python-sdk>=3.13.2",
     "hud-fastmcp-python-sdk>=0.1.2",
     "hud-mcp-use-python-sdk==2.3.20",
+    "langchain==0.3.27",
     "pathspec>=0.12.1",
     "wrapt>=1.14.0",
     # CLI dependencies