`mod request`#

module request#

LLM request codec types and trait.

This module defines the AnnotatedLlmRequest type system for structured LLM request representation and the crate::codec::traits::LlmCodec trait for bidirectional translation between opaque crate::api::llm::LlmRequest payloads and typed form.

Enums

enum ContentPart#

A single content part within a multimodal message.

v1 supports text only. Future versions may add ImageUrl, Audio, etc.

Text#

A text content part.

text: String#: The text content.

enum Message#

A single message in a conversation, tagged by role.

System#

A system instruction message.

content: MessageContent#: The message content.

name: Option<String>#: Optional sender name.

User#

A user message.

content: MessageContent#: The message content.

name: Option<String>#: Optional sender name.

Assistant#

An assistant response, optionally containing tool calls.

content: Option<MessageContent>#: The message content (optional — may be absent when tool calls are present).

tool_calls: Option<Vec<ToolCall>>#: Tool calls requested by the assistant.

name: Option<String>#: Optional sender name.

Tool#

A tool result message.

content: MessageContent#: The tool execution result.

tool_call_id: String#: The ID of the tool call this result corresponds to.

enum MessageContent#

Message content: either a plain string or multimodal parts array.

Text(String)#: Plain text content.

Parts(Vec<ContentPart>)#: Multimodal content parts.

enum ToolChoice#

Tool choice control: how the model should use available tools.

Auto#: Let the model decide whether to call a tool.

None#: Do not call any tools.

Required#: The model must call at least one tool.

Specific(ToolChoiceFunction)#: Force a specific function by name.

Structs and Unions

struct AnnotatedLlmRequest#

Structured view of an LLM request, produced by a Codec from opaque LlmRequest content.

The extra field captures any provider-specific keys not modeled by the known fields, ensuring lossless round-trip through decode/encode.

messages: Vec<Message>#: Parsed conversation messages.

model: Option<String>#: Model identifier (e.g., "gpt-4", "claude-sonnet-4-20250514").

params: Option<GenerationParams>#: Common generation parameters, normalized.

tools: Option<Vec<ToolDefinition>>#: Tool definitions (function schemas) available to the model.

tool_choice: Option<ToolChoice>#: Tool choice control.

extra: serde_json::Map<String, Json>#: Extensible key-value pairs for unmodeled provider-specific fields. Merged back into the request body during encode via serde(flatten).

Implementations

impl AnnotatedLlmRequest#

Functions

fn has_tool_calls(&self) -> bool#: Check if any assistant message in the conversation contains tool calls.

Returns true if at least one Message::Assistant variant has a non-empty tool_calls field.

fn last_user_message(&self) -> Option<&str>#: Get the text content of the last user message, if any.

Searches messages in reverse order and returns the first user message found. For MessageContent::Parts, returns the text of the first ContentPart::Text part.

fn system_prompt(&self) -> Option<&str>#: Extract the text content of the first system message, if any.

For MessageContent::Text, returns the string directly. For MessageContent::Parts, returns the text of the first ContentPart::Text part.

struct FunctionCall#

A function call within a tool call.

name: String#: The name of the function to call.

arguments: String#: The function arguments as a JSON string (per OpenAI convention).

struct FunctionDefinition#

A function definition within a tool definition.

name: String#: The name of the function.

description: Option<String>#: A description of what the function does.

parameters: Option<Json>#: The JSON Schema for the function parameters.

struct GenerationParams#

Normalized generation parameters across providers.

temperature: Option<f64>#: Sampling temperature.

max_tokens: Option<u64>#: Maximum number of tokens to generate.

top_p: Option<f64>#: Nucleus sampling probability.

stop: Option<Vec<String>>#: Stop sequences.

struct ToolCall#

A tool call requested by the assistant.

id: String#: Unique identifier for this tool call.

call_type: String#: The type of tool call (typically "function").

function: FunctionCall#: The function to call.

struct ToolChoiceFunction#

A specific tool choice that forces a named function.

choice_type: String#: The type (typically "function").

function: ToolChoiceFunctionName#: The function to call.

struct ToolChoiceFunctionName#

The name component of a specific tool choice.

name: String#: The function name.

struct ToolDefinition#

A tool definition (function schema) available to the model.

tool_type: String#: The type of tool (typically "function").

function: FunctionDefinition#: The function definition.

mod request#

`mod request`#