logo

furnace – Pure Rust inference server with Burn (zero‑Python, single binary)

50 points Asleep_Site_3731 22 comments
View on Reddit ↗

Hi Rustaceans! 🦀

I've built Furnace, a blazing-fast inference server written entirely in Rust, powered by the Burn framework.

It’s designed to be:

  • 🧊 Zero-dependency: no Python runtime, single 2.3MB binary
  • ⚡ Fast: sub-millisecond inference (~0.5ms tested on MNIST-like)
  • 🌐 Production-ready: REST API, CORS, error handling, CLI-based

🚀 Quick Start

git clone https://github.com/Gilfeather/furnace
cd furnace
cargo build --release
./target/release/furnace --model-path ./sample_model --port 3000

curl -X POST http://localhost:3000/predict \-H "Content-Type: application/json" \ -d "{"input": $(python3 -c 'import json; print(json.dumps([0.1] * 784))')}"

📊 Performance

Metric Value
Binary Size 2.3 MB
Inference Time ~0.5 ms
Memory Usage < 50 MB
Startup Time < 100 ms

🔧 Use Cases

  • Lightweight edge inference (IoT, WASM-ready)
  • Serverless ML without Python images
  • Embedded Rust systems needing local ML

🧪 GitHub Repo

https://github.com/Gilfeather/furnace

I'd love to hear your thoughts!
PRs, issues, stars, or architectural feedback are all welcome 😊

(Built with Rust 1.70+ and Burn, CLI-first using Axum and Tokio)