Adapters API Reference#

LLM provider adapters for automatic observability.

AdapterConfig#

AdapterConfig `dataclass` #

AdapterConfig(
    log_requests: bool = True,
    log_tool_calls: bool = True,
    log_streams: bool = True,
    log_stream_chunks: bool = False,
    chunk_log_interval: int = 10,
    on_error: Optional[
        Callable[[Exception, dict[str, Any]], None]
    ] = None,
    on_stream_start: Optional[
        Callable[[str, str, str], None]
    ] = None,
    on_token: Optional[Callable[[str, str], None]] = None,
    on_stream_end: Optional[
        Callable[[str, str, int], None]
    ] = None,
    on_stream_error: Optional[
        Callable[[Exception, str], None]
    ] = None,
    metadata: dict[str, Any] = dict(),
)

Configuration for provider adapters.

Attributes:

Name	Type	Description
`log_requests`	`bool`	Whether to log LLM request events. Default True.
`log_tool_calls`	`bool`	Whether to log tool call events. Default True.
`log_streams`	`bool`	Whether to log streaming events. Default True.
`log_stream_chunks`	`bool`	Whether to log individual stream chunks. Default False to reduce noise.
`chunk_log_interval`	`int`	If log_stream_chunks is True, log every Nth chunk. Default 10.
`on_error`	`Optional[Callable[[Exception, dict[str, Any]], None]]`	Optional callback for adapter errors. Signature: (error: Exception, context: dict) -> None
`on_stream_start`	`Optional[Callable[[str, str, str], None]]`	Optional callback invoked when a stream begins. Signature: (stream_id: str, model: str, provider: str) -> None
`on_token`	`Optional[Callable[[str, str], None]]`	Optional callback invoked for each content token during streaming. Signature: (token: str, stream_id: str) -> None
`on_stream_end`	`Optional[Callable[[str, str, int], None]]`	Optional callback invoked when a stream completes. Signature: (stream_id: str, content: str, chunks: int) -> None
`on_stream_error`	`Optional[Callable[[Exception, str], None]]`	Optional callback invoked when a stream fails. Signature: (error: Exception, stream_id: str) -> None

BaseAdapter#

BaseAdapter #

BaseAdapter(
    openintent_client: OpenIntentClient,
    intent_id: str,
    config: Optional[AdapterConfig] = None,
)

Base class for provider adapters.

Subclasses implement provider-specific wrapping logic while this base class provides common utilities and configuration handling.

Initialize the adapter.

Parameters:

Name	Type	Description	Default
`openintent_client`	`OpenIntentClient`	The OpenIntent client for logging events.	required
`intent_id`	`str`	The intent ID to associate events with.	required
`config`	`Optional[AdapterConfig]`	Optional adapter configuration. Uses defaults if not provided.	`None`

client `property` #

client: OpenIntentClient

The OpenIntent client.

config `property` #

config: AdapterConfig

The adapter configuration.

intent_id `property` #

intent_id: str

The intent ID for event logging.

OpenAI Adapter#

OpenAIAdapter #

OpenAIAdapter(
    openai_client: Any,
    openintent_client: OpenIntentClient,
    intent_id: str,
    config: Optional[AdapterConfig] = None,
)

Bases: BaseAdapter

Adapter for the OpenAI Python client.

Wraps an OpenAI client instance to automatically log OpenIntent events for all chat completions, tool calls, and streaming responses.

The adapter exposes the same interface as the OpenAI client, so you can use it as a drop-in replacement:

adapter = OpenAIAdapter(openai_client, openintent, intent_id)
response = adapter.chat.completions.create(...)

Events logged: - LLM_REQUEST_STARTED: When a completion request begins - LLM_REQUEST_COMPLETED: When a completion finishes successfully - LLM_REQUEST_FAILED: When a completion fails - TOOL_CALL_STARTED: When the model calls a tool - TOOL_CALL_COMPLETED: When tool execution completes (if tracked) - STREAM_STARTED: When a streaming response begins - STREAM_CHUNK: Periodically during streaming (if configured) - STREAM_COMPLETED: When streaming finishes - STREAM_CANCELLED: If streaming is interrupted

Initialize the OpenAI adapter.

Parameters:

Name	Type	Description	Default
`openai_client`	`Any`	The OpenAI client instance to wrap.	required
`openintent_client`	`OpenIntentClient`	The OpenIntent client for logging events.	required
`intent_id`	`str`	The intent ID to associate events with.	required
`config`	`Optional[AdapterConfig]`	Optional adapter configuration.	`None`

Raises:

Type	Description
`ImportError`	If the openai package is not installed.

Anthropic Adapter#

AnthropicAdapter #

AnthropicAdapter(
    anthropic_client: Any,
    openintent_client: OpenIntentClient,
    intent_id: str,
    config: Optional[AdapterConfig] = None,
)

Bases: BaseAdapter

Adapter for the Anthropic Python client.

Wraps an Anthropic client instance to automatically log OpenIntent events for all message creations, tool use, and streaming responses.

The adapter exposes the same interface as the Anthropic client, so you can use it as a drop-in replacement:

adapter = AnthropicAdapter(anthropic_client, openintent, intent_id)
message = adapter.messages.create(...)

Events logged: - LLM_REQUEST_STARTED: When a message request begins - LLM_REQUEST_COMPLETED: When a message finishes successfully - LLM_REQUEST_FAILED: When a message request fails - TOOL_CALL_STARTED: When the model uses a tool - STREAM_STARTED: When a streaming response begins - STREAM_CHUNK: Periodically during streaming (if configured) - STREAM_COMPLETED: When streaming finishes - STREAM_CANCELLED: If streaming is interrupted

Initialize the Anthropic adapter.

Parameters:

Name	Type	Description	Default
`anthropic_client`	`Any`	The Anthropic client instance to wrap.	required
`openintent_client`	`OpenIntentClient`	The OpenIntent client for logging events.	required
`intent_id`	`str`	The intent ID to associate events with.	required
`config`	`Optional[AdapterConfig]`	Optional adapter configuration.	`None`

Raises:

Type	Description
`ImportError`	If the anthropic package is not installed.

Azure OpenAI Adapter#

AzureOpenAIAdapter #

AzureOpenAIAdapter(
    azure_client: Any,
    openintent_client: OpenIntentClient,
    intent_id: str,
    config: Optional[AdapterConfig] = None,
)

Bases: BaseAdapter

Adapter for the Azure OpenAI Python client.

Wraps an AzureOpenAI client instance to automatically log OpenIntent events for all chat completions, tool calls, and streaming responses.

The adapter exposes the same interface as the OpenAI client, so you can use it as a drop-in replacement:

adapter = AzureOpenAIAdapter(azure_client, openintent, intent_id)
response = adapter.chat.completions.create(...)

Events logged: - LLM_REQUEST_STARTED: When a completion request begins - LLM_REQUEST_COMPLETED: When a completion finishes successfully - LLM_REQUEST_FAILED: When a completion fails - TOOL_CALL_STARTED: When the model calls a tool - STREAM_STARTED: When a streaming response begins - STREAM_CHUNK: Periodically during streaming (if configured) - STREAM_COMPLETED: When streaming finishes - STREAM_CANCELLED: If streaming is interrupted

Streaming hooks (on_stream_start, on_token, on_stream_end, on_stream_error) are fully supported via AdapterConfig.

Initialize the Azure OpenAI adapter.

Parameters:

Name	Type	Description	Default
`azure_client`	`Any`	The AzureOpenAI client instance to wrap.	required
`openintent_client`	`OpenIntentClient`	The OpenIntent client for logging events.	required
`intent_id`	`str`	The intent ID to associate events with.	required
`config`	`Optional[AdapterConfig]`	Optional adapter configuration.	`None`

Raises:

Type	Description
`ImportError`	If the openai package is not installed.

OpenRouter Adapter#

OpenRouterAdapter #

OpenRouterAdapter(
    openrouter_client: Any,
    openintent_client: OpenIntentClient,
    intent_id: str,
    config: Optional[AdapterConfig] = None,
)

Bases: BaseAdapter

Adapter for the OpenRouter API (OpenAI-compatible client).

Wraps an OpenAI client configured for OpenRouter's API to automatically log OpenIntent events for all chat completions, tool calls, and streaming.

OpenRouter provides access to 200+ models from multiple providers through a single unified API. The adapter tracks which underlying model is used.

The adapter exposes the same interface as the OpenAI client:

adapter = OpenRouterAdapter(openrouter_client, openintent, intent_id)
response = adapter.chat.completions.create(...)

Events logged: - LLM_REQUEST_STARTED: When a completion request begins - LLM_REQUEST_COMPLETED: When a completion finishes successfully - LLM_REQUEST_FAILED: When a completion fails - TOOL_CALL_STARTED: When the model calls a tool - STREAM_STARTED: When a streaming response begins - STREAM_CHUNK: Periodically during streaming (if configured) - STREAM_COMPLETED: When streaming finishes - STREAM_CANCELLED: If streaming is interrupted

Streaming hooks (on_stream_start, on_token, on_stream_end, on_stream_error) are fully supported via AdapterConfig.

Initialize the OpenRouter adapter.

Parameters:

Name	Type	Description	Default
`openrouter_client`	`Any`	The OpenAI client configured for OpenRouter's API.	required
`openintent_client`	`OpenIntentClient`	The OpenIntent client for logging events.	required
`intent_id`	`str`	The intent ID to associate events with.	required
`config`	`Optional[AdapterConfig]`	Optional adapter configuration.	`None`

Raises:

Type	Description
`ImportError`	If the openai package is not installed.

Gemini Adapter#

GeminiAdapter #

GeminiAdapter(
    gemini_model: Any,
    openintent_client: OpenIntentClient,
    intent_id: str,
    config: Optional[AdapterConfig] = None,
)

Bases: BaseAdapter

Adapter for the Google Generative AI (Gemini) Python client.

Wraps a GenerativeModel instance to automatically log OpenIntent events for all content generation, tool calls, and streaming responses.

The adapter exposes the same interface as the GenerativeModel, so you can use it as a drop-in replacement:

adapter = GeminiAdapter(model, openintent, intent_id)
response = adapter.generate_content("Hello")

Events logged: - LLM_REQUEST_STARTED: When a generation request begins - LLM_REQUEST_COMPLETED: When generation finishes successfully - LLM_REQUEST_FAILED: When generation fails - TOOL_CALL_STARTED: When the model calls a function - STREAM_STARTED: When a streaming response begins - STREAM_CHUNK: Periodically during streaming (if configured) - STREAM_COMPLETED: When streaming finishes - STREAM_CANCELLED: If streaming is interrupted

Initialize the Gemini adapter.

Parameters:

Name	Type	Description	Default
`gemini_model`	`Any`	The GenerativeModel instance to wrap.	required
`openintent_client`	`OpenIntentClient`	The OpenIntent client for logging events.	required
`intent_id`	`str`	The intent ID to associate events with.	required
`config`	`Optional[AdapterConfig]`	Optional adapter configuration.	`None`

Raises:

Type	Description
`ImportError`	If the google-generativeai package is not installed.

model `property` #

model: Any

The wrapped GenerativeModel.

generate_content #

generate_content(
    contents: Any, *, stream: bool = False, **kwargs: Any
) -> Any

Generate content with automatic event logging.

Parameters:

Name	Type	Description	Default
`contents`	`Any`	The prompt or conversation contents.	required
`stream`	`bool`	Whether to stream the response.	`False`
`**kwargs`	`Any`	Additional arguments passed to generate_content.	`{}`

Returns:

Type	Description
`Any`	GenerateContentResponse or iterator of chunks if streaming.

start_chat #

start_chat(**kwargs: Any) -> GeminiChatSession

Start a chat session with automatic event logging.

Returns a wrapped chat session that logs events for each message.

Grok Adapter#

GrokAdapter #

GrokAdapter(
    grok_client: Any,
    openintent_client: OpenIntentClient,
    intent_id: str,
    config: Optional[AdapterConfig] = None,
)

Bases: BaseAdapter

Adapter for the xAI Grok API (OpenAI-compatible client).

Wraps an OpenAI client configured for xAI's API to automatically log OpenIntent events for all chat completions, tool calls, and streaming.

The adapter exposes the same interface as the OpenAI client:

adapter = GrokAdapter(grok_client, openintent, intent_id)
response = adapter.chat.completions.create(...)

Events logged: - LLM_REQUEST_STARTED: When a completion request begins - LLM_REQUEST_COMPLETED: When a completion finishes successfully - LLM_REQUEST_FAILED: When a completion fails - TOOL_CALL_STARTED: When the model calls a tool - STREAM_STARTED: When a streaming response begins - STREAM_CHUNK: Periodically during streaming (if configured) - STREAM_COMPLETED: When streaming finishes - STREAM_CANCELLED: If streaming is interrupted

Initialize the Grok adapter.

Parameters:

Name	Type	Description	Default
`grok_client`	`Any`	The OpenAI client configured for xAI's API.	required
`openintent_client`	`OpenIntentClient`	The OpenIntent client for logging events.	required
`intent_id`	`str`	The intent ID to associate events with.	required
`config`	`Optional[AdapterConfig]`	Optional adapter configuration.	`None`

Raises:

Type	Description
`ImportError`	If the openai package is not installed.

grok `property` #

grok: Any

The wrapped Grok client.

DeepSeek Adapter#

DeepSeekAdapter #

DeepSeekAdapter(
    deepseek_client: Any,
    openintent_client: OpenIntentClient,
    intent_id: str,
    config: Optional[AdapterConfig] = None,
)

Bases: BaseAdapter

Adapter for the DeepSeek API (OpenAI-compatible client).

Wraps an OpenAI client configured for DeepSeek's API to automatically log OpenIntent events for all chat completions, tool calls, and streaming.

The adapter exposes the same interface as the OpenAI client:

adapter = DeepSeekAdapter(deepseek_client, openintent, intent_id)
response = adapter.chat.completions.create(...)

Events logged: - LLM_REQUEST_STARTED: When a completion request begins - LLM_REQUEST_COMPLETED: When a completion finishes successfully - LLM_REQUEST_FAILED: When a completion fails - TOOL_CALL_STARTED: When the model calls a tool - STREAM_STARTED: When a streaming response begins - STREAM_CHUNK: Periodically during streaming (if configured) - STREAM_COMPLETED: When streaming finishes - STREAM_CANCELLED: If streaming is interrupted

Initialize the DeepSeek adapter.

Parameters:

Name	Type	Description	Default
`deepseek_client`	`Any`	The OpenAI client configured for DeepSeek's API.	required
`openintent_client`	`OpenIntentClient`	The OpenIntent client for logging events.	required
`intent_id`	`str`	The intent ID to associate events with.	required
`config`	`Optional[AdapterConfig]`	Optional adapter configuration.	`None`

Raises:

Type	Description
`ImportError`	If the openai package is not installed.

deepseek `property` #

deepseek: Any

The wrapped DeepSeek client.

Adapters API Reference#

AdapterConfig#

AdapterConfig dataclass #

BaseAdapter#

BaseAdapter #

client property #

config property #

intent_id property #

OpenAI Adapter#

OpenAIAdapter #

Anthropic Adapter#

AnthropicAdapter #

Azure OpenAI Adapter#

AzureOpenAIAdapter #

OpenRouter Adapter#

OpenRouterAdapter #

Gemini Adapter#

GeminiAdapter #

model property #

generate_content #

start_chat #

Grok Adapter#

GrokAdapter #

grok property #

DeepSeek Adapter#

DeepSeekAdapter #

deepseek property #

AdapterConfig `dataclass` #

client `property` #

config `property` #

intent_id `property` #

model `property` #

grok `property` #

deepseek `property` #