TL;DR: Kiri is a container runtime written in Rust that implements the Docker Engine API, allowing it to drop in as a replacement for Docker Desktop. It manages containers via a small Linux VM that sleeps when idle, keeping the daemon footprint around 15MB.
Best for: Developers on macOS, Windows, or Linux who find Docker Desktop too resource-heavy.
Repository snapshot: ⭐ 11 685 stars · 🍴 547 forks · Rust · Apache-2.0
Daily picks: Follow GitHub Open Source on Telegram
What it is
Kiri is an open-source container runtime written in Rust that provides a lightweight drop-in replacement for Docker Desktop. Built to solve the problem of high idle resource consumption, it operates with a tiny 15MB daemon footprint and achieves container start times of around 2.5ms.
How it works
Under the hood, Kiri manages containers via a small Linux virtual machine that automatically goes to sleep after 60 seconds of inactivity. It implements the standard Docker Engine API, meaning you can continue using your familiar CLI commands, Docker Compose files, and BuildKit workflows without modifying your existing scripts.
Key features
- Starts containers in roughly 2.5ms via the Docker API
- Maintains a 15MB daemon footprint with automatic idle suspension
- Provides automatic local domains with built-in HTTPS support
- Includes a single-node k3s Kubernetes cluster managed through the CLI
- Supports image and volume migration from existing setups via simple commands
Getting started
To try Kiri on your local machine, clone the repository and install it according to the documentation provided in the project source.
git clone https://github.com/libracoder/kiri
When not to use it
While Kiri is performant and suitable for daily development work, you should verify your specific edge-case volume mounts and specialized network configurations before adopting it in production or complex staging pipelines.
Discuss
Are you sticking with Docker Desktop, or have you already switched to a lighter runtime like Kiri or OrbStack?
Curated from the repository README and metadata. No hands-on test is claimed; evaluate it in a sandbox before production use.
Top comments (0)