Crate cortexai_llm_client 

LLM Client - Shared Logic

Runtime-agnostic LLM client logic for building requests and parsing responses. This crate has NO runtime dependencies (no async, no HTTP client).

Supported Providers

• OpenAI (GPT-4, GPT-3.5, etc.)
• Anthropic (Claude 3, etc.)
• OpenRouter (100+ models)

Usage

use cortexai_llm_client::{
    Provider, Message, RequestBuilder, ResponseParser,
};

// Build a request
let messages = vec![
    Message::system("You are a helpful assistant."),
    Message::user("Hello!"),
];

let request = RequestBuilder::new(Provider::OpenAI)
    .model("gpt-4o-mini")
    .messages(&messages)
    .api_key("sk-...")
    .temperature(0.7)
    .max_tokens(1024)
    .stream(false)
    .build()
    .unwrap();

// Use your runtime's HTTP client to send request.url, request.headers, request.body
// Then parse the response:

let response_json = r#"{"choices":[{"message":{"content":"Hello!"}}]}"#;
let response = ResponseParser::parse(Provider::OpenAI, response_json).unwrap();
println!("{}", response.content);

Structs

HttpRequest

An HTTP request ready to be sent.

LlmResponse

Parsed LLM response

Message

A message in a conversation.

RequestBuilder

Builder for constructing LLM API requests.

ResponseParser

Response parser for different providers

StreamChunk

Streaming chunk from SSE response

ToolCall

Tool call requested by the model

ToolCallChunk

Partial tool call information from streaming

Usage

Token usage information

Enums

LlmClientError

Errors that can occur in the LLM client.

Provider

Supported LLM providers.

Role

Role in a conversation.

Type Aliases

Result

Result type for LLM client operations.