Skip to content
The tiniest PaaS you've ever seen. Piku allows you to do git push deployments to your own servers.
Python Shell
Branch: master
Clone or download

piku logo

The tiniest Heroku/CloudFoundry-like PaaS you've ever seen.

piku, inspired by dokku, allows you do git push deployments to your own servers.


Using piku

piku supports a Heroku-like workflow, like so:

  • Create a git SSH remote pointing to your piku server with the app name as repo name. git remote add piku piku@yourserver:appname.
  • Push your code: git push piku master.
  • piku determines the runtime and installs the dependencies for your app (building whatever's required).
    • For Python, it segregates each app's dependencies into a virtualenv.
    • For Go, it defines a separate GOPATH for each app.
    • For Node, it installs whatever is in package.json into node_modules.
    • For Java, it builds your app depending on either pom.xml or build.gradle file.
  • It then looks at a Procfile and starts the relevant workers using uWSGI as a generic process manager.
  • You can optionally also specify a release worker which is run once when the app is deployed.
  • You can then remotely change application settings (config:set) or scale up/down worker processes (ps:scale).
  • You can also bake application settings into a file called ENV which is documented here.


To use piku you need a VPS, Raspberry Pi, or other server bootstrapped with piku's requirements. You can use a single server to run multiple piku apps.

Warning: You should use a fresh server or VPS instance without anything important running on it already, as piku-bootstrap will make changes to configuration files, running services, etc.

Once you've got a fresh server, download the piku-bootstrap shell script onto your local machine and run it:

curl > piku-bootstrap && chmod 755 piku-bootstrap

The first time it is run piku-bootstrap will install itself into ~/.piku-bootstrap on your local machine and set up a virtualenv there with the dependencies it requires. It will only need to do this once.

The script will display a usage message and you can then bootstrap your server:


If you put the piku-bootstrap script on your PATH somewhere, you can use it again to provision other servers in the future.

See below for instructions on installing other custom dependencies that your apps might need like a database etc.

piku client

To make life easier you can also download the piku helper shell script and install it on your local.

curl > piku && chmod 755 piku

This shell script makes working with piku remotes a bit simpler. If you have a git remote called piku in the current folder it will infer the remote server and app name and insert those into the remote piku commands. This allows you do execute commands like the following on your running remote app:

$ piku logs
$ piku config:set MYVAR=12
$ piku stop
$ piku deploy
$ piku destroy
$ piku # <- will show help for the remote app

If you put this piku script on your PATH you can use the piku command across multiple apps on your local.

Installing other dependencies

piku-bootstrap uses Ansible internally and it comes with several extra built-in playbooks which you can use to bootstrap common components onto your piku server.

Use piku-bootstrap list-playbooks to show a list of built-in playbooks, and then to install one add it as an argument to the bootstrap command.

For example, to deploy nodeenv onto your server:

piku-bootstrap nodeenv.yml

You can also use piku-bootstrap to run your own Ansible playbooks like this:

piku-bootstrap ./myplaybook.yml


You can find examples for deploying various kinds of apps into a piku server in the Examples folder.


I kept finding myself wanting an Heroku/CloudFoundry-like way to deploy stuff on a few remote ARM boards and my Raspberry Pi cluster, but since dokku didn't work on ARM at the time and even docker can be overkill sometimes, I decided to roll my own.

Project Status/To Do:

This is currently being used for production deployments of my website and a few other projects of mine that run on Azure and other IaaS providers. Regardless, there is still room for improvement:

From the bottom up:

  • Prebuilt Raspbian image with everything baked in
  • chroot/namespace isolation (tentative)
  • Relay commands to other nodes
  • Proxy deployments to other nodes (build on one box, deploy to many)
  • Support Clojure/Java deployments through boot or lein
  • Sample Go app
  • Support Go deployments (in progress)
  • nginx SSL optimization/cypher suites, own certificates
  • Review deployment messages
  • WIP: Review docs/CLI command documentation (short descriptions done, need help <cmd> and better descriptions)
  • Lua/WSAPI support
  • Support for Java Apps with maven/gradle (in progress through jwsgi, by @matrixjnr)
  • Django and Wisp examples (by @chr15m)
  • Project logo (by @chr15m)
  • Various release/deployment improvements (by @chr15m)
  • Support Node deployments (by @chr15m)
  • Let's Encrypt support (by @chr15m)
  • Allow setting nginx IP bindings in ENV file (NGINX_IPV4_ADDRESS and NGINX_IPV6_ADDRESS)
  • Cleanups to remove 2.7 syntax internally
  • Change to Python 3 runtime as default, with PYTHON_VERSION = 2 as fallback
  • Run in Python 3 only
  • (experimental) REPL in feature/repl
  • Python 3 support through PYTHON_VERSION = 3
  • static URL mapping to arbitrary paths (hat tip to @carlosefr for nginx tuning)
  • remote CLI (requires ssh -t)
  • saner uWSGI logging
  • gevent activated when UWSGI_GEVENT = <integer>
  • enable CloudFlare ACL when NGINX_CLOUDFLARE_ACL = True
  • Autodetect SPDY/HTTPv2 support and activate it
  • Basic nginx SSL config with self-signed certificates and UNIX domain socket connection
  • nginx support - creates an nginx config file if NGINX_SERVER_NAME is defined
  • Testing with pre-packaged uWSGI versions on Debian Jessie (yes, it was painful)
  • Support barebones binary deployments
  • Complete installation instructions (see, which also has a draft of Go installation steps)
  • Installation helper/SSH key setup
  • Worker scaling
  • Remote CLI commands for changing/viewing applied/live settings
  • Remote tailing of all logfiles for a single application
  • HTTP port selection (and per-app environment variables)
  • Sample Python app
  • Procfile support (wsgi and worker processes for now, web processes being tested)
  • Basic CLI commands to manage apps
  • virtualenv isolation
  • Support Python deployments
  • Repo creation upon first push
  • Basic understanding of how dokku works


This is an illustrated example of how piku works for a Python deployment:

Supported Platforms

piku is intended to work in any POSIX-like environment where you have Python, uWSGI and SSH, i.e.: Linux, FreeBSD, Cygwin and the Windows Subsystem for Linux.

As a baseline, it began its development on an original, 256MB Rasbperry Pi Model B, and still runs reliably on it.

Since I have an ODROID-U2, a bunch of Pi 2s and a few more ARM boards on the way, it is often tested on a number of places where running x64 binaries is unfeasible.

But there are already a few folk using piku on vanilla x64 Linux without any issues whatsoever, so yes, you can use it as a micro-PaaS for 'real' stuff. Your mileage may vary.

Supported Runtimes

piku currently supports deploying apps (and dependencies) written in Python, with Go, Clojure (Java) and Node (see above) in the works. But if it can be invoked from a shell, it can be run inside piku.


Q: Why piku?

A: Partly because it's supposed to run on a Pi, because it's Japanese onomatopeia for 'twitch' or 'jolt', and because I know the name will annoy some of my friends.

Q: Why Python/why not Go?

A: I actually thought about doing this in Go right off the bat, but click is so cool and I needed to have uWSGI running anyway, so I caved in. But I'm very likely to take something like suture and port this across, doing away with uWSGI altogether.

Go also (at the time) did not have a way to vendor dependencies that I was comfortable with, and that is also why Go support fell behind. Hopefully that will change soon.

Q: Does it run under Python 3?

A: Right now, it only runs on Python 3, even though it can deploy apps written in both major versions. It began its development using 2.7 and usingclick for abstracting the simpler stuff, and I eventually switched over to 3.5 once it was supported in Debian Stretch and Raspbian since I wanted to make installing it on the Raspberry Pi as simple as possible.

Q: Why not just use dokku?

A: I used dokku daily for most of my personal stuff for a good while. But it relied on a number of x64 containers that needed to be completely rebuilt for ARM, and when I decided I needed something like this (March 2016) that was barely possible - docker itself was not fully baked for ARM yet, and people were at the time trying to get herokuish and buildstep to build on ARM.

You can’t perform that action at this time.