A container runtime with nothing in the way.

A container runtime is the program that turns an image into a running process: it unpacks the layers, sets up namespaces, cgroups and a network, and starts your command. Kiri does all of that in one Rust daemon, so a container is running before you'd notice the wait.

Download Kiri See the numbers

Kiri speaks the Docker Engine API, so the docker CLI and Compose drive it. Free and open source under Apache 2.0.

How it's built

One daemon instead of a stack.

The usual stack

On a typical setup, docker talks to dockerd, which talks to containerd, which starts a shim, which runs runc, which starts your process. Every hop adds time and memory.

Kiri's stack

docker talks to Kiri. Kiri sets up the namespaces, the cgroup, overlayfs and the seccomp filter itself and creates the container with one clone3. No dockerd, no containerd, no shims, no runc.

Nothing left per container

No process stays behind for each container. OrbStack keeps a shim for every one: 143 running containers held 1.13 GB in shims on our test Mac.

On Linux, no VM

The same daemon runs straight on your kernel, as a systemd service, packaged as .deb and .rpm for x86-64 and ARM.

On a Mac, one small VM

One Linux VM through Apple's Virtualization.framework serves every container. It boots in 112 ms, and requests cross one vsock channel with a 31 µs round trip.

On Windows, inside WSL2

The engine runs in a WSL distro you pick, and the Docker API reaches Windows on a named pipe. No admin rights, no service.

Measured

What a small runtime buys you.

Four engines on one Apple silicon Mac, each limited to the same 2 CPUs and 8 GB, driven by the same docker CLI with the same warm images. Medians, lower is better for times.

Measurement	Kiri	Docker Desktop	OrbStack	Colima
`docker run --rm alpine true`	22 ms	178 ms	279 ms	152 ms
`docker exec … true`	15.7 ms	47.7 ms	32.5 ms	36.8 ms
Container lifecycle through the Docker API	3.1 ms	163 ms	238 ms	88 ms
Containers started in parallel	88.9 /s	11.4 /s	8.2 /s	12.9 /s
200 MB out of a container	223 MB/s	45 MB/s	87 MB/s	63 MB/s
Bind mount read, 1 GB	1724 MB/s	1354 MB/s	3240 MB/s	5235 MB/s

2 November 2026, Kiri 0.1.13, Docker Desktop 4.90.0, OrbStack 2.2.4, Colima 0.10.3. Where another engine wins, the table says so: Colima reads bind mounts faster. On Windows Kiri starts a container in 78 ms against Docker Desktop's 491 ms. Method, every row and the caveats: benchmarks.

Side by side

Kiri against the other ways to run containers.

Feature	Kiri	Docker Desktop	OrbStack	Colima
Mac	Yes	Yes	Yes	Yes
Windows	Yes	Yes	No	No
Linux	Yes, no VM	In a VM	No	Yes
Free at work	Yes	Paid over 250 staff	Paid	Yes
Open source	Apache 2.0	No	No	MIT
Desktop app	Native	Electron	Native	CLI only
`docker`, Compose, buildx	Yes	Yes	Yes	Yes
Kubernetes	Yes	Yes	Yes	k3s
Public link to a container	Kiri Share	No	No	No
`docker run` on a Mac	22 ms	178 ms	279 ms	152 ms

Docker Desktop needs a paid subscription in companies with more than 250 employees or more than $10 million a year in revenue. OrbStack is free for personal use and paid for work. Checked November 2026; each project's own pages have the current terms.

Switching

Keep your commands. Change the engine.

Install Kiri
On a Mac, the app. On Windows, a per-user installer for WSL2. On Linux, a .deb or .rpm: curl -fsSL https://runkiri.com/install.sh | bash.
Bring your things over
Kiri moves images, volumes and containers over from Docker Desktop or OrbStack, so nothing is rebuilt. Keep the old one installed while you try it.
Run what you always run
```
$ docker compose up
```
Kiri adds a kiri docker context, so the docker CLI, Compose, buildx and your IDE talk to Kiri.

Questions

Asked often.

What is a container runtime?

A container runtime is the software that runs containers: it takes an image, unpacks its layers, isolates a process with Linux namespaces and cgroups, connects it to a network and starts it. runc, crun, containerd and Docker Engine are runtimes at different levels; Kiri does the whole job in one daemon.

What is the best container runtime for development?

For a laptop, the one that starts containers fast, stays light when idle and works with the tools you already use. Kiri starts a container in 22 ms on a Mac, sleeps its VM after a minute idle and works with the docker CLI and Compose. Every measurement is on our benchmarks page, including the ones where another engine wins.

Is Kiri an alternative to containerd or runc?

Yes. Kiri does not use containerd, runc or shims. It creates containers itself with clone3 and sets up namespaces, cgroups, overlayfs and seccomp on its own.

Is Kiri OCI compatible?

Kiri runs standard OCI and Docker images from Docker Hub and any registry, and builds images with buildx.

Which systems does Kiri run on?

macOS 14 or later, Windows 10 and 11 with WSL2, and Linux with systemd on x86-64 and ARM.