Support for anthropic models

.env.example

@@ -1 +1,2 @@
-OPENAI_API_KEY=sk-proj-***************-***********************-***********************
+OPENAI_API_KEY=sk-proj-***************-***********************-***********************
+ANTHROPIC_API_KEY=sk-ant-****-***************-***********************-***********************

bluebox/agents/docs_digger_agent.py

@@ -32,7 +32,7 @@
     PendingToolInvocation,
     ToolInvocationStatus,
 )
-from bluebox.data_models.llms.vendors import OpenAIModel
+from bluebox.data_models.llms.vendors import LLMModel, OpenAIModel
 from bluebox.llms.llm_client import LLMClient
 from bluebox.llms.infra.documentation_data_store import (
     DocumentationDataStore,
@@ -185,7 +185,7 @@ def __init__(
         persist_chat_callable: Callable[[Chat], Chat] | None = None,
         persist_chat_thread_callable: Callable[[ChatThread], ChatThread] | None = None,
         stream_chunk_callable: Callable[[str], None] | None = None,
-        llm_model: OpenAIModel = OpenAIModel.GPT_5_1,
+        llm_model: LLMModel = OpenAIModel.GPT_5_1,
         chat_thread: ChatThread | None = None,
         existing_chats: list[Chat] | None = None,
     ) -> None:

bluebox/agents/network_spy_agent.py

@@ -33,7 +33,7 @@
     PendingToolInvocation,
     ToolInvocationStatus,
 )
-from bluebox.data_models.llms.vendors import OpenAIModel
+from bluebox.data_models.llms.vendors import LLMModel, OpenAIModel
 from bluebox.llms.llm_client import LLMClient
 from bluebox.llms.infra.network_data_store import NetworkDataStore
 from bluebox.utils.code_execution_sandbox import execute_python_sandboxed
@@ -192,7 +192,7 @@ def __init__(
         persist_chat_callable: Callable[[Chat], Chat] | None = None,
         persist_chat_thread_callable: Callable[[ChatThread], ChatThread] | None = None,
         stream_chunk_callable: Callable[[str], None] | None = None,
-        llm_model: OpenAIModel = OpenAIModel.GPT_5_1,
+        llm_model: LLMModel = OpenAIModel.GPT_5_1,
         chat_thread: ChatThread | None = None,
         existing_chats: list[Chat] | None = None,
     ) -> None:

bluebox/agents/trace_hound_agent.py

@@ -32,7 +32,7 @@
     PendingToolInvocation,
     ToolInvocationStatus,
 )
-from bluebox.data_models.llms.vendors import OpenAIModel
+from bluebox.data_models.llms.vendors import LLMModel, OpenAIModel
 from bluebox.llms.llm_client import LLMClient
 from bluebox.llms.infra.network_data_store import NetworkDataStore
 from bluebox.llms.infra.storage_data_store import StorageDataStore
@@ -184,7 +184,7 @@ def __init__(
         persist_chat_callable: Callable[[Chat], Chat] | None = None,
         persist_chat_thread_callable: Callable[[ChatThread], ChatThread] | None = None,
         stream_chunk_callable: Callable[[str], None] | None = None,
-        llm_model: OpenAIModel = OpenAIModel.GPT_5_1,
+        llm_model: LLMModel = OpenAIModel.GPT_5_1,
         chat_thread: ChatThread | None = None,
         existing_chats: list[Chat] | None = None,
     ) -> None:

bluebox/config.py

@@ -35,6 +35,7 @@ class Config():
 
     # API keys
     OPENAI_API_KEY: str | None = os.getenv("OPENAI_API_KEY")
+    ANTHROPIC_API_KEY: str | None = os.getenv("ANTHROPIC_API_KEY")
 
     # Code execution sandbox configuration
     # Mode: "docker" (require Docker), "blocklist" (no Docker), "auto" (Docker if available)

bluebox/data_models/llms/vendors.py

@@ -5,57 +5,73 @@
 """
 
 from enum import StrEnum
+from typing import ClassVar
 
 
 class LLMVendor(StrEnum):
     """
     Represents the vendor of an LLM.
     """
     OPENAI = "openai"
-    #TODO:ANTHROPIC = "anthropic"
+    ANTHROPIC = "anthropic"
 
 
-class OpenAIAPIType(StrEnum):
-    """
-    OpenAI API type.
-    """
-    CHAT_COMPLETIONS = "chat_completions"
-    RESPONSES = "responses"
+class VendorModel(StrEnum):
+    """Base for model enums. Subclasses must define _vendor."""
+    _vendor: ClassVar[LLMVendor]
 
+    def __init_subclass__(cls, **kwargs: object) -> None:
+        super().__init_subclass__(**kwargs)
+        if not hasattr(cls, '_vendor'):
+            raise TypeError(f"{cls.__name__} must define _vendor class attribute")
+
+    @property
+    def vendor(self) -> LLMVendor:
+        return self.__class__._vendor
+
+
+class OpenAIModel(VendorModel):
+    """OpenAI models."""
+    _vendor = LLMVendor.OPENAI
 
-class OpenAIModel(StrEnum):
-    """
-    OpenAI models.
-    """
     GPT_5 = "gpt-5"
     GPT_5_1 = "gpt-5.1"
     GPT_5_2 = "gpt-5.2"
     GPT_5_MINI = "gpt-5-mini"
     GPT_5_NANO = "gpt-5-nano"
 
 
-# class AnthropicModel(StrEnum):
-#     """Anthropic models."""
-#     CLAUDE_OPUS_4_5 = "claude-opus-4-5-20251101"
-#     CLAUDE_SONNET_4_5 = "claude-sonnet-4-5-20250929"
-#     CLAUDE_HAIKU_4_5 = "claude-haiku-4-5-20251001"
+class AnthropicModel(VendorModel):
+    """Anthropic models."""
+    _vendor = LLMVendor.ANTHROPIC
 
+    CLAUDE_OPUS_4_5 = "claude-opus-4-5"
+    CLAUDE_SONNET_4_5 = "claude-sonnet-4-5"
+    CLAUDE_HAIKU_4_5 = "claude-haiku-4-5"
 
-# LLMModel is only OpenAI models for now
-LLMModel = OpenAIModel
 
+# LLMModel type: union of all vendor models
+LLMModel = OpenAIModel | AnthropicModel
 
-# Build model to vendor lookup from OpenAI models only
-_model_to_vendor: dict[str, LLMVendor] = {}
-_all_models: dict[str, str] = {}
 
-for model in OpenAIModel:
-    _model_to_vendor[model.value] = LLMVendor.OPENAI
-    _all_models[model.name] = model.value
+def get_model_by_value(model_value: str) -> LLMModel | None:
+    """
+    Get model enum by value string.
 
+    Args:
+        model_value: The model value string (e.g., "gpt-5.1", "claude-opus-4-5").
 
-def get_model_vendor(model: LLMModel) -> LLMVendor:
-    """
-    Returns the vendor of the LLM model.
+    Returns:
+        LLMModel if found, None otherwise. Use model.vendor to get the vendor.
     """
-    return _model_to_vendor[model.value]
+    for model_cls in VendorModel.__subclasses__():
+        try:
+            return model_cls(model_value)
+        except ValueError:
+            continue
+    return None
+
+
+def get_all_model_values() -> list[str]:
+    """Get all available model value strings."""
+    return [m.value for cls in VendorModel.__subclasses__() for m in cls]

bluebox/llms/abstract_llm_vendor_client.py

@@ -11,7 +11,7 @@
 from pydantic import BaseModel
 
 from bluebox.data_models.llms.interaction import LLMChatResponse
-from bluebox.data_models.llms.vendors import OpenAIModel
+from bluebox.data_models.llms.vendors import LLMModel, LLMVendor
 
 
 T = TypeVar("T", bound=BaseModel)
@@ -27,13 +27,27 @@ class AbstractLLMVendorClient(ABC):
 
     # Class attributes ____________________________________________________________________________________________________
 
+    _vendor: ClassVar[LLMVendor]
     DEFAULT_MAX_TOKENS: ClassVar[int] = 4_096
     DEFAULT_TEMPERATURE: ClassVar[float] = 0.7
     DEFAULT_STRUCTURED_TEMPERATURE: ClassVar[float] = 0.0
 
+    def __init_subclass__(cls, **kwargs: object) -> None:
+        super().__init_subclass__(**kwargs)
+        if not hasattr(cls, '_vendor'):
+            raise TypeError(f"{cls.__name__} must define _vendor class attribute")
+
+    @classmethod
+    def get_llm_vendor_client(cls, model: LLMModel) -> "AbstractLLMVendorClient":
+        """Create the appropriate vendor client for the given model."""
+        for subclass in cls.__subclasses__():
+            if subclass._vendor == model.vendor:
+                return subclass(model=model)
+        raise ValueError(f"No client found for vendor: {model.vendor}")
+
     # Magic methods ________________________________________________________________________________________________________
 
-    def __init__(self, model: OpenAIModel) -> None:
+    def __init__(self, model: LLMModel) -> None:
         """
         Initialize the vendor client.
 
@@ -43,6 +57,22 @@ def __init__(self, model: OpenAIModel) -> None:
         self.model = model
         self._tools: list[dict[str, Any]] = []
 
+    # Protected methods ____________________________________________________________________________________________________
+
+    def _resolve_max_tokens(self, max_tokens: int | None) -> int:
+        """Resolve max_tokens, using default if None."""
+        return max_tokens if max_tokens is not None else self.DEFAULT_MAX_TOKENS
+
+    def _resolve_temperature(
+        self,
+        temperature: float | None,
+        structured: bool = False,
+    ) -> float:
+        """Resolve temperature, using appropriate default if None."""
+        if temperature is not None:
+            return temperature
+        return self.DEFAULT_STRUCTURED_TEMPERATURE if structured else self.DEFAULT_TEMPERATURE
+
     # Tool management ______________________________________________________________________________________________________
 
     @abstractmethod
@@ -83,8 +113,8 @@ def call_sync(
         temperature: float | None = None,
         response_model: type[T] | None = None,
         extended_reasoning: bool = False,
-        stateful: bool = False,
         previous_response_id: str | None = None,
+        tool_choice: str | None = None,
     ) -> LLMChatResponse | T:
         """
         Unified sync call to the LLM.
@@ -97,8 +127,8 @@ def call_sync(
             temperature: Sampling temperature (0.0-1.0).
             response_model: Pydantic model class for structured response.
             extended_reasoning: Enable extended reasoning (if supported).
-            stateful: Enable stateful conversation (if supported).
             previous_response_id: Previous response ID for chaining (if supported).
+            tool_choice: Tool selection mode ("auto", "none", "required", or specific tool).
 
         Returns:
             LLMChatResponse or parsed Pydantic model if response_model is provided.
@@ -115,8 +145,8 @@ async def call_async(
         temperature: float | None = None,
         response_model: type[T] | None = None,
         extended_reasoning: bool = False,
-        stateful: bool = False,
         previous_response_id: str | None = None,
+        tool_choice: str | None = None,
     ) -> LLMChatResponse | T:
         """
         Unified async call to the LLM.
@@ -129,8 +159,8 @@ async def call_async(
             temperature: Sampling temperature (0.0-1.0).
             response_model: Pydantic model class for structured response.
             extended_reasoning: Enable extended reasoning (if supported).
-            stateful: Enable stateful conversation (if supported).
             previous_response_id: Previous response ID for chaining (if supported).
+            tool_choice: Tool selection mode ("auto", "none", "required", or specific tool).
 
         Returns:
             LLMChatResponse or parsed Pydantic model if response_model is provided.
@@ -146,8 +176,8 @@ def call_stream_sync(
         max_tokens: int | None = None,
         temperature: float | None = None,
         extended_reasoning: bool = False,
-        stateful: bool = False,
         previous_response_id: str | None = None,
+        tool_choice: str | None = None,
     ) -> Generator[str | LLMChatResponse, None, None]:
         """
         Unified streaming call to the LLM.
@@ -161,8 +191,8 @@ def call_stream_sync(
             max_tokens: Maximum tokens in the response.
             temperature: Sampling temperature (0.0-1.0).
             extended_reasoning: Enable extended reasoning (if supported).
-            stateful: Enable stateful conversation (if supported).
             previous_response_id: Previous response ID for chaining (if supported).
+            tool_choice: Tool selection mode ("auto", "none", "required", or specific tool).
 
         Yields:
             str: Text chunks as they arrive.