cera-client

Asynchronous, strongly typed Rust client for OpenAI and OpenRouter endpoints.

Part of the Cera inference workspace.

Features

• Multi-Provider Support: First-class support for OpenAI, OpenRouter, and custom OpenAI-compatible servers (vLLM, Ollama, LocalAI, etc.).
• Chat Completions: Typed request and response models with tool calling, JSON schema output, and custom sampling parameters.
• Streaming Responses: Real-time Server-Sent Events (SSE) stream parsing yielding chunk deltas.
• Embeddings: Generate vector embeddings for text with batch input support.
• Model Discovery: Fetch available model catalogs from providers.
• Environment Auto-Discovery: Automatically resolve OPENAI_API_KEY or OPENROUTER_API_KEY when available.

Usage

use cera_client::{ChatMessage, ChatCompletionRequest, Client};

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    // Connects to OpenRouter or OpenAI using environment variables
    let client = Client::from_env()?;

    let request = ChatCompletionRequest::new(
        "gpt-4o-mini",
        vec![
            ChatMessage::system("You are a concise assistant."),
            ChatMessage::user("Explain Rust ownership in two sentences."),
        ],
    )
    .temperature(0.7)
    .max_tokens(150);

    let response = client.chat(request).await?;
    if let Some(choice) = response.choices.first() {
        if let Some(content) = &choice.message.content {
            println!("{content}");
        }
    }

    Ok(())
}