A modular, Docker-based development environment you assemble like LEGO bricks. Pick the bricks you want, snap them together, and reach your stack from the browser, a web terminal, or SSH.

Landing page & demo · Builder · GitHub

DevLEGO is a single Docker image that you build from a set of interchangeable bricks (components). Choose the IDEs, notebooks, terminals, and tooling you want, and DevLEGO bakes them into one container you can reach three ways:

• In your browser — VS Code (Code Server), JupyterLab, Jupyter Notebook
• In a web terminal — a full shell in a browser tab (ttyd)
• Over SSH — remote shell and tunneling, always available

It's GPU-ready out of the box: the image is built on the NVIDIA base image nvcr.io/nvidia/mxnet:24.06-py3, and docker-compose.yml reserves your GPU for the container, so CUDA-capable workloads just work.

Like LEGO, the value is in the modularity: every brick is one small shell script in component/, and turning the browser-facing services on or off is a single environment variable. New in 0.2.0: the Claude Code and Codex agentic coding CLIs come pre-installed in every terminal.

Every brick that ships with DevLEGO:

Also baked into every image: git, git-lfs, vim, nano, htop, ncdu, cmake, build-essential, g++, ffmpeg, wget, curl, screen.

The four service bricks (Code Server, JupyterLab, Jupyter Notebook, ttyd) default to ON. The AI CLIs have no port and no toggle — they're always installed and ready in any terminal.

You need Docker with the Compose V2 plugin (note: docker compose, two words).

# 1. Clone the repo and step in
git clone https://github.com/voidful/DevLEGO.git && cd DevLEGO

# 2. Create the Docker network
docker network create devlego

# 3. Copy the env template (add API keys if you want the AI CLIs authenticated — optional)
cp .env.example .env

# 4. Build and start everything in the background
docker compose up -d --build

Then open Code Server at http://localhost:38001 — default password: changeme.

Default credentials come from docker-compose.yml: USERNAME=devlego, PASSWORD=changeme. Change these before exposing DevLEGO anywhere beyond your local machine.

API keys are optional — unset keys are simply omitted, so the container still builds and runs without them. To authenticate Claude Code or Codex, provide keys via a .env file next to docker-compose.yml (Compose auto-loads it) or via your host shell. Start by copying the template:

cp .env.example .env

Then fill in the keys you need. The following variables are passed through to the container:

Compose uses the bare - VAR passthrough form, so any variable you leave unset is omitted from the container entirely (not set to an empty string). See .env.example for the full template.

Codex's base URL is configured via its config.toml (openai_base_url), not an environment variable, so there is no OPENAI_BASE_URL passthrough.

The four browser-facing service bricks are on by default. Disable any of them by setting its ENABLE_* variable to false (in .env or your host shell, picked up by docker-compose.yml):

ENABLE_CODE_SERVER=false
ENABLE_JUPYTER_LAB=false
ENABLE_JUPYTER_NOTEBOOK=false
ENABLE_TTYD=false

The SSH server always runs, and the AI CLIs (Claude Code, Codex) are always installed — neither has a toggle.

The Builder is a small companion web tool (docs/builder.html, linked from the landing page) that generates a ready-to-paste docker run command from the components and ports you select. Pick your bricks visually, copy the command, and go.

Adding a component is deliberately LEGO-simple:

1. Drop a shell script into component/ (for example component/my-brick.sh) that installs and sets up your tool.

2. Wire two lines into the Dockerfile — one COPY and one RUN:

   COPY component/my-brick.sh .
   RUN bash ./my-brick.sh

3. Rebuild with docker compose up -d --build.

That's the whole contract. If your brick exposes a port, add it to ports.sh; if it pins a version, add it to versions.sh.

Ports are defined in ports.sh — adjust them there if any collide with services on your host.

DevLEGO is inspired by these excellent projects:

• pojntfx/pojde — browser-based development environments over SSH
• p208p2002/docker-for-ai-dev — Docker setups for AI development

Built by voidful (Eric Lam).

Released under the MIT License — Copyright (c) 2024 Eric Lam.