A high-performance inference engine for AI models
use std::io::IsTerminal;
use iocraft::prelude::*;
use shoji::types::basic::ReasoningEffort;
use uzu::engine::{Engine, EngineConfig, EngineError};
use crate::interactive::components::Application;
#[derive(Debug, Clone, PartialEq, thiserror::Error)]
#[non_exhaustive]
pub enum CliError {
#[error(transparent)]
Engine(#[from] EngineError),
#[error("Rendering error: {message}")]
RenderingError {
message: String,
},
}
#[derive(Clone)]
pub struct CliApplication {
engine: Engine,
}
impl CliApplication {
pub async fn create(config: EngineConfig) -> Result<Self, CliError> {
let engine = Engine::new(config).await?;
Ok(Self::new(engine))
}
pub fn new(engine: Engine) -> Self {
Self {
engine,
}
}
pub async fn run_with_model(
&self,
model: Option<String>,
reasoning_effort: Option<ReasoningEffort>,
seed: Option<i64>,
no_tools: bool,
) -> Result<(), CliError> {
if !std::io::stdout().is_terminal() {
return Err(CliError::RenderingError {
message: "stdout is not a terminal".to_string(),
});
}
element! {
Application(
engine: Some(self.engine.clone()),
model,
reasoning_effort,
seed,
no_tools,
)
}
.render_loop()
.await
.map_err(|error| CliError::RenderingError {
message: error.to_string(),
})?;
Ok(())
}
}