Module llm_utils 

Structs

APIInputMessage

Common message structure used for LLM API requests

APIOutputMessage

Common message structure used for LLM API requests

APITool

AzureCompletionRequest

Simple completion-focused LLM request for Azure OpenAI Note: In Azure OpenAI, the model is specified in the URL, not in the request body

AzureCompletionRequestRef 🔒

The wire shape of AzureCompletionRequest, borrowing its conversation instead of owning it.

LLMResponse

Response from LLM for simple completions

MessageContentItem

RefusalContentItem

StoreDisabled

The store: false every request sends, as a type rather than a bool so that no construction site can send the other value.

Enums

MessageContent

Functions

azure_chatbot_configuration

The Azure chatbot (Foundry) section of the application configuration, or an error naming what’s missing.

azure_configuration

The Azure section of the application configuration, or an error naming what’s missing.

azure_search_configuration

The Azure AI Search section of the application configuration, or an error naming what’s missing.

azure_search_request

A request to the Azure AI Search management API, carrying the headers every one of its endpoints wants. The search API key is exposed only here.

build_llm_headers

Builds common headers for LLM requests

estimate_tokens

Estimate the number of tokens in a given text.

get_params_for_model

llm_http_error 🔒

Builds the error for a failed LLM HTTP response, parsing error_text as an Azure error body when possible and attaching it as the error’s Azure source.

make_blocking_llm_request

Makes a non-streaming request to an LLM using application configuration

make_llm_request 🔒

Makes a non-streaming request to an LLM

make_streaming_llm_request

Makes a streaming request to an LLM

model_is_thinking

Checks if the model_type is a thinking model type. This function defines which model types are thinking (reasoning)

parse_text_completion

Collects all the completion choices to a string. Assumes the completion has only text message content, no tool calls or tool output.

process_llm_response 🔒

Process a non-streaming LLM response

reasoning_include 🔒

The include a reasoning request asks for, or None for a model that produces no reasoning.

request_structured_json

Sends input as a one-shot structured-JSON request and deserializes the reply as T.

stored_summary 🔒

A stored summary text as the summary parts that go back to Azure.

string_list_response_format

The structured output format for a feature whose whole answer is a list of strings.

summarize_input_for_log 🔒

Logs the shape and order of a request’s input items when Azure rejects it.

summary_as_stored 🔒

Summary parts as the database round trip spells them.

summary_text 🔒

The single summary text that ChatbotConversationMessageReasoning::summary can hold, or None for an item with no summary. Azure streams a summary in parts; a row keeps one.