Skip to main content
This is an advanced option. Most people should install the desktop app instead.
Docker runs AppBlips inside a container: a sealed-off box with everything AppBlips needs already in it. You don’t install Node.js or anything else on your computer. One command builds the container and starts it, and you use AppBlips in your web browser. It’s the same app as the browser version from source: your projects are saved in your browser, not in the container. Your API key is saved by the container in a Docker volume named appblips-data, not in your browser.

​
Before you start

  • Install Docker. On Windows and macOS, install Docker Desktop and open it. On Linux, install Docker Engine with the Compose plugin. On Ubuntu that’s sudo apt install docker.io docker-compose-v2; don’t install the package called just docker-compose, which is an old version that no longer works.
  • Get the AppBlips code. With Git:
    Without Git, open the GitHub page, click Code → Download ZIP, unzip it, and open a terminal in that folder.

​
Start AppBlips

1

Build and start the container

In the AppBlips folder, run:
The first time, Docker downloads what it needs and builds AppBlips, including the voice-input speech model (about 76 MB, from Hugging Face). That takes a few minutes. Later starts take seconds.
2

Open AppBlips

Go to http://localhost:3000 in your browser.
3

Set up your AI provider

A first screen asks for your name, and a second for a chat avatar; click Continue on each. Then a setup screen asks for your AI provider, API key and model. Click Start building when you’re done. You can change them later in Settings → AI. If your .env file already sets a provider, this screen is skipped.
4

Build your first app

Click a studio, such as Mobile & Web App, type what you want in plain English, and press Enter. See Build your first app.
Type docker compose, with a space. docker-compose with a hyphen is an old, retired tool: on newer Linux systems it fails with No module named 'distutils', and it can’t read AppBlips’ docker-compose.yml. If docker compose says it’s an unknown command, install the Compose plugin (on Ubuntu: sudo apt install docker-compose-v2).
To see what AppBlips is printing (for example, if the page doesn’t load), run docker compose logs -f. Press Ctrl+C to stop watching; AppBlips keeps running. The container restarts by itself when your computer or Docker restarts, until you stop it.

​
Stop and update

To stop AppBlips, run this in the AppBlips folder:
To update to a newer version:
  1. Get the new code: git pull (or download the ZIP again and unzip it over the old folder).
  2. Rebuild and restart:
Your projects stay in your browser, and your API key stays in the appblips-data volume, through updates and restarts. docker compose down -v deletes that volume, so you would enter your key again.

​
Use another port

AppBlips uses port 3000 on your computer. To use another one:
  1. In the AppBlips folder, create a file named .env (or copy .env.example to .env).
  2. Add this line, with the port you want:
  3. Run docker compose up -d again, then open http://localhost:8080.
Projects are saved per address, so moving to a new port starts with an empty project list. Export your apps first from Settings → Data, and import them at the new address.

​
Give the server an AI provider (optional)

Instead of entering your provider in the app, you can put it in the same .env file. This works with DeepSeek or another OpenAI-compatible service, not Anthropic: for Anthropic, enter your key in the app under Settings → AI. For example, with a DeepSeek key:
Run docker compose up -d after changing .env so the container picks up the new values. See the Configuration reference for every setting.

​
Local models (LM Studio and Ollama)

AppBlips can use LM Studio or Ollama running on your computer. Choose it in Settings → AI as usual and leave the server URL as it is: the container reaches your computer through the name host.docker.internal by itself.
  • Windows and macOS (Docker Desktop): this works as it is.
  • Linux: the local server has to accept connections from the container, not only from your own computer:
    • Ollama: start it with OLLAMA_HOST=0.0.0.0. For the system service, run sudo systemctl edit ollama, add Environment="OLLAMA_HOST=0.0.0.0" under [Service], save, and run sudo systemctl restart ollama.
    • LM Studio: in the Developer tab, turn on Serve on Local Network.
    • If it still can’t connect, your firewall may be blocking the container. Running from source avoids this.
“Serve on local network” and OLLAMA_HOST=0.0.0.0 also let other devices on your network use your local model. Only do this on a network you trust.

​
Keep it on your computer

docker-compose.yml publishes AppBlips as 127.0.0.1:3000, so only your own computer can open it.
Don’t remove the 127.0.0.1: part. AppBlips has no sign-in, so anyone who can reach it can spend the API key you gave the server. See Security.
The container runs a small Node.js server bundled into a single file, with no node_modules folder, and it runs as a non-root user.