emixai

emixai bundles feature-gated helpers for audio, imaging, language, and vision workloads on top of the EssentialMix core libraries. Each module stays optional so you only pull in the tooling your project needs.

Features

• language (default): OpenAI and Kalosm-backed chat helpers, prompt flows, and streaming responses.
• audio: Whisper-based transcription utilities and playback helpers.
• imaging: Image generation and manipulation helpers.
• vision: Shared computer-vision helpers.
• full: Convenience flag that enables every module.
• Hardware flags (cuda, metal, mkl) align with Kalosm’s acceleration backends.

Enable the crate with the features you need in your Cargo.toml:

[dependencies]
emixai = { path = "../../crates/ai", features = ["language"] }
tokio = { version = "1", features = ["full"] }
config = "0.14"

Quick Example (ChatGPT Stream)

use config::{Config, Environment};
use emixai::language::openai::{ChatGpt, OpenAiSource};
use emixai::SourceSize;

#[tokio::main]
async fn main() -> emixai::Result<()> {
    // Expect OPENAI__API_KEY in the environment.
    let settings = Config::builder()
        .add_source(Environment::with_prefix("OPENAI"))
        .build()?;

    // Use the default stream buffer size and a small model footprint.
    let chat = ChatGpt::from_size(settings, SourceSize::Small, None);
    chat.prompt("Describe EssentialMix in one sentence.").await?;

    Ok(())
}

• Swap SourceSize at runtime to move between inference cost tiers.

• Prefer ChatGpt::subscribe() to stream deltas into your own channel.

• See tests/ for more end-to-end usage covering audio transcription, Whisper, Wuerstchen imaging, and LLaMA language flows.