anthropic_client

Anthropic ModelClient integration with OpenAI SDK compatibility.

Functions

get_chat_completion_usage(completion)

Convert ChatCompletion.usage into our AdalFlowResponseUsage format.

Classes

AnthropicAPIClient([api_key, base_url, ...])

A component wrapper for Anthropic API using OpenAI SDK compatibility.
class AnthropicAPIClient(api_key: str | None = None, base_url: str = 'https://api.anthropic.com/v1/', non_streaming_chat_completion_parser: Callable[[ChatCompletion], Any] | None = None, streaming_chat_completion_parser: Callable[[AsyncStream], Any] | None = None)[source]

Bases: ModelClient

A component wrapper for Anthropic API using OpenAI SDK compatibility.

This client leverages Anthropic’s OpenAI SDK compatibility layer to provide ChatCompletion-based responses while maintaining AdalFlow’s GeneratorOutput structure.

Features: - Uses OpenAI SDK with Anthropic’s compatibility endpoint - Supports both streaming and non-streaming calls - Handles ModelType.LLM and ModelType.LLM_REASONING - Converts ChatCompletion responses to Response API format for compatibility - Maintains backward compatibility with existing AdalFlow parsers

Parameters:

  • api_key (Optional[str]) – Anthropic API key. Defaults to ANTHROPIC_API_KEY env var.

  • base_url (str) – Anthropic’s OpenAI compatibility endpoint.

  • non_streaming_chat_completion_parser (Callable) – Legacy parser for non-streaming ChatCompletion objects. Used for backward compatibility with existing code that depends on the original AnthropicAPI client’s parsing behavior.

  • streaming_chat_completion_parser (Callable) – Parser for streaming ChatCompletion responses. Handles conversion from ChatCompletion streams to Response API format.

Note

Requires ANTHROPIC_API_KEY environment variable or api_key parameter. Uses OpenAI SDK internally but calls Anthropic’s API via compatibility layer. The non_streaming_chat_completion_parser is provided for legacy compatibility only.

init_sync_client()[source]

Initialize synchronous OpenAI client pointing to Anthropic.

init_async_client()[source]

Initialize asynchronous OpenAI client pointing to Anthropic.

parse_chat_completion(completion: ChatCompletion | Stream[ChatCompletionChunk] | AsyncStream[ChatCompletionChunk]) → GeneratorOutput[source]

Parse ChatCompletion and convert to GeneratorOutput with Response API compatibility.

This method uses ChatCompletionToResponseConverter to transform ChatCompletion objects into text format compatible with existing Response API parsers.

Parameters:

completion – ChatCompletion object or stream from OpenAI SDK

Returns:

GeneratorOutput with converted raw_response text

track_completion_usage(completion: ChatCompletion) → CompletionUsage[source]

Track completion usage from ChatCompletion object.

convert_inputs_to_api_kwargs(input: Any | None = None, model_kwargs: Dict = {}, model_type: ModelType = ModelType.UNDEFINED) → Dict[source]

Convert AdalFlow inputs to OpenAI ChatCompletion API format.

Converts single input text to OpenAI messages format expected by chat.completions.create endpoint.

Parameters:

  • input – Text input or messages array

  • model_kwargs – Additional model parameters

  • model_type – Type of model (LLM or LLM_REASONING)

Returns:

API kwargs formatted for OpenAI chat.completions.create

Return type:

Dict

convertible with original api of Anthropic’s Message API

call(api_kwargs: Dict = {}, model_type: ModelType = ModelType.UNDEFINED)[source]

Synchronous call to Anthropic via OpenAI SDK compatibility.

Supports both LLM and LLM_REASONING model types with streaming and non-streaming.

Parameters:

  • api_kwargs – API parameters for chat.completions.create

  • model_type – ModelType.LLM or ModelType.LLM_REASONING

Returns:

ChatCompletion or Stream[ChatCompletionChunk] from Anthropic

async acall(api_kwargs: Dict = {}, model_type: ModelType = ModelType.UNDEFINED)[source]

Asynchronous call to Anthropic via OpenAI SDK compatibility.

Parameters:

  • api_kwargs – API parameters for chat.completions.create

  • model_type – ModelType.LLM or ModelType.LLM_REASONING

Returns:

ChatCompletion or AsyncStream[ChatCompletionChunk] from Anthropic