Passthrough Chat

This module contains passthrough views for interacting directly with the LLM.

provider backend API.

class smarter.apps.prompt.api.v1.views.passthrough.PassthroughChatViewSet(**kwargs)

Bases: SmarterAuthenticatedNeverCachedWebView

Handle POST requests to the passthrough endpoint for direct LLM provider API access.

path: /api/v1/prompts/passthrough/{provider_name}/

This endpoint allows authenticated users to send arbitrary prompt dicts to the underlying LLM provider (such as OpenAI). The request body should be a JSON object containing any valid parameters accepted by the provider’s prompt completion API.

Parameters:

• request (rest_framework.request.ASGIRequest) – The HTTP request object, expected to have a JSON body with prompt completion parameters.

• args – Additional positional arguments (unused).

• kwargs – Additional keyword arguments. May include ‘provider’ to select the LLM provider.

Returns:

A JSON response containing the provider’s prompt completion result, or an error message.

Return type:

SmarterJournaledJsonResponse | SmarterJournaledJsonErrorResponse | SmarterHttpResponseBadRequest | SmarterHttpResponseForbidden | SmarterHttpResponseNotFound

Signals:

• chat_started: Sent before the prompt completion request is made.

• chat_completion_request: Sent with the prompt data before calling the provider.

• chat_completion_response: Sent after a successful response from the provider.

• chat_finished: Sent after the prompt completion process is finished.

• chat_response_failure: Sent if an exception occurs during the provider call.

Raises:

• SmarterHttpResponseForbidden – If the user is not authenticated.

• SmarterHttpResponseNotFound – If the specified provider is not found.

• SmarterHttpResponseBadRequest – If the request body is invalid.

• SmarterJournaledJsonErrorResponse – If the provider API call fails.

See also

• The OpenAI API documentation for prompt completions: https://developers.openai.com/api/reference/overview/prompt/create

• openai.types.prompt.chat_completion.ChatCompletion

handler: OpenAICompatiblePassthroughProtocol

post(request, *args, **kwargs)

Handle POST requests to the passthrough endpoint for direct LLM.

provider API access.

Return type:

SmarterJournaledJsonResponse | SmarterJournaledJsonErrorResponse | SmarterHttpErrorResponse | SmarterHttpResponseForbidden

provider_name: str

setup(request, *args, **kwargs)

Set the provider_name and handler based on the URL kwargs.

The handler can be any function that implements the OpenAICompatiblePassthroughProtocol interface.

See also

• OpenAICompatiblePassthroughProtocol

smarter.apps.prompt.api.v1.views.passthrough.should_log(level)

Check if logging should be done based on the waffle switch.