rust-project

Rust Project Architecture

Core Principles

• Ownership-first — Embrace borrow checker, no unnecessary clones
• Zero-cost abstractions — Newtype, iterators, async/await
• Workspace for scale — Use Cargo workspace for multi-crate projects
• Error precision — thiserror for libs, anyhow for apps
• Async with Tokio — Tokio runtime + tracing for observability
• No backwards compatibility — Delete, don't deprecate. Change directly
• LiteLLM for LLM APIs — Use LiteLLM proxy for all LLM integrations

No Backwards Compatibility

Delete unused code. Change directly. No compatibility layers.

// ❌ BAD: Deprecated attribute kept around
#[deprecated(since = "0.2.0", note = "Use new_function instead")]
pub fn old_function() { ... }

// ❌ BAD: Type alias for renamed types
pub type OldName = NewName; // "for backwards compatibility"

// ❌ BAD: Unused parameters
fn process(_legacy: &str, data: &Data) { ... }

// ❌ BAD: Feature flags for old behavior
#[cfg(feature = "legacy")]
fn old_impl() { ... }

// ✅ GOOD: Just delete and update all usages
pub fn new_function() { ... }
// Then: Find & replace all old_function → new_function

// ✅ GOOD: Remove unused parameters entirely
fn process(data: &Data) { ... }

LiteLLM for LLM APIs

Use LiteLLM proxy. Don't call provider APIs directly.

// src/llm.rs
use async_openai::{Client, config::OpenAIConfig};

pub fn create_client(base_url: &str, api_key: &str) -> Client<OpenAIConfig> {
    let config = OpenAIConfig::new()
        .with_api_base(base_url)  // LiteLLM proxy URL
        .with_api_key(api_key);
    Client::with_config(config)
}

// Usage: connect to LiteLLM, use any model
let client = create_client("http://localhost:4000", &api_key);
let request = CreateChatCompletionRequestArgs::default()
    .model("gpt-4o")  // or "claude-3-opus", "gemini-pro", etc.
    .messages(vec![...])
    .build()?;

Quick Start

1. Initialize Project

# Simple project
cargo new myapp
cd myapp

# Workspace project
mkdir myapp && cd myapp
cargo init --name app