Container runtime for Linux and macOS

Docker runtime, cross-platform & fast.

Containers reimagined for the Agentic era.

Install Docs

$ curl -fsS https://www.coraild.com/install | bash
run --rm /bin/true
                corail  docker  podman
Linux  start      6 ms  214 ms  110 ms
       exec       2 ms   53 ms  105 ms
       50 idle   54 MB  1.2 GB  107 MB
macOS  start    5.4 ms  132 ms  152 ms
       exec     5.2 ms   33 ms  115 ms
       50 idle  196 MB  1.5 GB  1.3 GB
$  
  • 6 ms Warm start Docker: 214 ms
  • 2 ms Exec into a running container Docker: 53 ms
  • 54 MB Memory for 50 idle containers Docker: 1.2 GB
  • 250 ms First container from a stopped VM, macOS Docker Desktop: 2.6 s

Benchmarked head to head with Docker and Podman on the same hardware. See the full results

Binaries

corail runs a container. coraild keeps a group running.

corail exits when the command finishes. coraild stays up on one node.

  • corail

    Runtime

    Pulls an OCI image from a registry and runs it. The command exits. On Linux, no process stays in the background.

    Run a container

  • coraild and recify

    Control plane

    Describe a group of containers in one file and apply it. coraild probes them, restarts what fails and keeps the group running on a single node. recify is its command line.

    Run an atoll

  • sonar

    Terminal UI

    Terminal UI for coraild. It lists groups, shows a container's logs, and opens a shell with s.

    Screen and keys

Runtime

Docker commands. Rootless. Linux and macOS.

Docker commands.

run, ps, logs, exec, stop and rm follow Docker's flags. corail pulls OCI images from any registry.

All commands and flags

$ corail run -d --name web --memory 64m alpine \
    -- sh -c 'echo started; sleep 300'
web
$ corail ps
ID      PID      STATUS   COMMAND
web     47722    running  sh
$ corail logs web
started
$ corail exec web -- echo hi
hi

Rootless and confined by default.

No sudo to run a container. No daemon runs as root.

  • Containers run in user namespaces, with their own cgroups.
  • A seccomp allow-list is on by default: a system call that is not on the list is refused.
  • Image pulls are sandboxed with Landlock while corail reads what a registry sends.
  • One static binary.

Rootless prerequisites

$ id -u
1000
$ corail run --rm --memory 64m --cpus 1 --pids-limit 64 \
    -v /srv/data:/data:ro alpine -- cat /data/greeting.txt
hello
$ corail run --rm alpine -- sh -c 'exit 3'; echo $?
3

Linux and macOS.

On Linux, corail starts containers on the host. On macOS, one Linux VM per store.

Introduction

Linux

Namespaces and cgroups on the host. x86_64 and aarch64.

6 mswarm start
54 MBfor 50 idle containers

macOS

One Linux VM per store. It reserves 2 GB; Docker Desktop's default is 8 GB.

5.4 mswarm start
250 msfrom a stopped VM

coraild restarts a container that fails.

A reef.json describes the group. recify apply starts it. coraild runs the probes.

  • 6 ms from apply to running.
  • 8 ms from a killed container to a recovered one.
  • 3.6 MB of daemon memory with 50 groups.

reef.json

$ recify apply -f examples/web/reef.json
atoll/web applied
$ recify get atolls
NAME  READY    STATUS      RESTARTS  AGE
web   2/2      Running     0         6s
$ recify exec web -c sidecar -- echo hi
hi

Benchmarks

Measured against Docker and Podman.

Same machine, same image, runs interleaved. Rows where corail does not win are shown too.

Warm start

Warm start

Warm start

Exec into a running container

Exec into a running container

Exec into a running container

Cold pull, alpine

Cold pull, busybox

Cold pull, debian

CPU with 50 idle workloads

CPU with 50 idle workloads

CPU with 50 idle workloads

Memory with 50 idle workloads

Memory with 50 idle workloads

Memory with 50 idle workloads

Warm start

Warm start

Warm start

Exec into a running container

Exec into a running container

Exec into a running container

From a stopped VM

From a stopped VM

Not measured for busybox.

From a stopped VM

Not measured for debian.

First pull, alpine

First pull, busybox

Not measured for busybox.

First pull, debian

Not measured for debian.

CPU reserved

Memory with 50 idle workloads

Memory with 50 idle workloads

Memory with 50 idle workloads

Memory reserved

Hover a bar for the command and date behind it. Every figure, the machines and the method are on the numbers page.

Install

Install, run, apply.

  1. Install

    Static binaries for Linux. No dependencies.

    $ curl -fsS https://www.coraild.com/install | bash
  2. Run a container

    The first run pulls the image. No sudo.

    $ corail run --rm alpine -- echo hi
  3. Run a group

    Start coraild, then apply a reef.json.

    $ recify apply -f reef.json
  • Linux runs it. x86_64 and aarch64.
  • macOS runs it in a VM. Intel and Apple silicon.
  • Windows is not planned.
  • Full install guide

Containers reimagined for the Agentic era.

Install Numbers