What Monoceros is

Monoceros is a workbench for local, reproducible dev containers with AI coding tooling. You describe what should live in a container - languages, services, AI tools, repos - in one YAML file, and Monoceros materializes it. The point: to give your AI agents (and yourself) a box that holds the whole stack but can’t reach your host - so you can let an agent work without handing it your machine. Language- and stack-agnostic; no cloud, no SaaS, no built-in workflow lock-in.

New to Monoceros? Start with Installation, then build your first workbench.

How you use it

1. Describe the workbench in its yml - languages, services, AI tools, repos.
2. Build it with monoceros apply <name>: one command turns the file into a running container.
3. Work inside it - a shell, your IDE, or an AI agent, with the whole stack on hand.

Change the yml and apply again. The container is always derived from the file, so it is reproducible on any machine and disposable: tear it down, rebuild it, get the same thing back.

How it differs

Alternative	What they do differently	What Monoceros does better
GitHub Codespaces	Cloud-only, vendor lock-in, hourly cost	Local, no rental, data stays on your machine
Cursor Cloud	Cloud workspace, fixed tooling stack	Local, the tools are your choice
Plain devcontainer	Works, but you build everything yourself	Reusable yml profiles, CLI boilerplate, curated AI-tool features
Local dev setup	Full host access for every tool	Container isolation: AI tools run in a separate Linux, not on your host

The building blocks

Local, no cloud

Everything runs on your own machine and your own Docker. No account, no rental, no hourly bill; nothing leaves the device except what you choose to send.

Declarative yml model

One file describes the container; monoceros apply materializes it. Reproducible across machines, versionable, diffable.

AI tools are first-class

Claude Code, OpenCode, Atlassian Rovo Dev, the GitHub and GitLab CLIs land as devcontainer features, chosen explicitly in the yml.

Container isolation by default

Everything runs in a Linux container, not on your host. An agent cannot reach your ~/.ssh, your browser state, or files outside the workspace.

Note

Isolation is about the host, not the network. By default an agent in the container can still make network calls - containment keeps it off your machine, not off the internet. We do not claim otherwise.

What Monoceros is not

• Not a cloud service. It is your Docker, your data, no rental - the opposite of a hosted workspace.
• Not a web UI. The CLI is the interface and the container is where you work; there is no dashboard to log into.
• Not an agent or a workflow. There is no built-in plan/generate/review pipeline - you bring your own AI tool; Monoceros just gives it a good box.
• Not a fixed stack. Nothing is preset: you assemble what goes in the container, from any language to any Docker image.
• Not multi-user. Each person has their own configs under $MONOCEROS_HOME; sharing a setup is a matter of git, not shared state.