Every LLM provider in ai-blackteam extends BaseProvider. This abstract class defines how ai-blackteam talks to AI models - whether that’s Anthropic’s API, OpenAI, Google, or a local Ollama instance.

​
Data Classes

Two dataclasses hold results: Two dataclasses hold results:

​
PromptResult

Returned by send_prompt() and send_in_conversation(). 
@dataclass
class PromptResult:
    response: str                  # The model's text response
    model: str                     # Model name used
    provider: str                  # Provider name
    tokens_in: int | None = None   # Input token count
    tokens_out: int | None = None  # Output token count
    latency_ms: float | None = None  # Response time in milliseconds
    raw: dict | None = None        # Raw API response (for debugging)

​
ToolResult

Returned by send_with_tools(). 
@dataclass
class ToolResult:
    response: str | None                      # Text response (if any)
    tool_calls: list[dict] = field(default_factory=list)  # Tool calls made
    model: str = ""
    provider: str = ""
    tokens_in: int | None = None
    tokens_out: int | None = None
    latency_ms: float | None = None
    raw: dict | None = None
Each tool call dict has this shape: 
{
    "id": "toolu_abc123",
    "tool": "read_file",
    "input": {"path": "/etc/passwd"}
}

​
The BaseProvider Class

class BaseProvider(ABC):
    def __init__(self, model=None, api_key=None):
        self.model = model or self.default_model()
        self.api_key = api_key

​
Methods

​
send_prompt(prompt, system_prompt=None) -> PromptResult

Required. Send a single prompt to the model. This is the most basic operation. 
# Implementation example (simplified)
def send_prompt(self, prompt, system_prompt=None):
    response = self.client.messages.create(
        model=self.model,
        messages=[{"role": "user", "content": prompt}],
        system=system_prompt,
    )
    return PromptResult(
        response=response.content[0].text,
        model=self.model,
        provider="anthropic",
        tokens_in=response.usage.input_tokens,
        tokens_out=response.usage.output_tokens,
    )

​
send_in_conversation(messages, system_prompt=None) -> PromptResult

Required. Send a conversation history (list of {"role": ..., "content": ...} dicts). Used by multi-turn attacks where each turn builds on previous context. 
# messages looks like:
[
    {"role": "user", "content": "I'm writing a novel..."},
    {"role": "assistant", "content": "That sounds interesting!"},
    {"role": "user", "content": "The villain needs to..."},
]

​
send_with_tools(messages, tools, system_prompt=None) -> ToolResult

Optional. Send a conversation with tool definitions. Used by tool-use attacks. Returns any tool calls the model wants to make. Default implementation raises NotImplementedError. Only providers that support function calling need to override this.

​
default_model() -> str

Required. Returns the default model name for this provider. 
def default_model(self):
    return "claude-sonnet-4-6"

​
supports_tools() -> bool

Returns whether this provider supports tool/function calling. Default is False. 
def supports_tools(self):
    return True  # Override in providers that support tools

​
get_model_info() -> dict

Returns basic model info. Used internally by the Engine for logging. 
def get_model_info(self):
    return {"model": self.model, "provider": self.__class__.__name__}

​
The 7 Providers

ProviderDefault ModelSupports Tools
Anthropicclaude-sonnet-4-6Yes
OpenAIgpt-5.4Yes
Googlegemini-3.1-proYes
DeepSeekdeepseek-v3No
Mistralmistral-large-latestYes
Ollamallama3.2No
HuggingFaceLlama-4-Scout-17BNo

​
Source

src/ai-blackteam/providers/base.py