micalang/mica repository overview

⁠Mica

The official image of Mica⁠, compute you can see through: a statically typed systems language with its own optimizing native backend and a built-in numeric and machine-learning surface, from tensors with compile-time shapes to an autograd tape and GPU-resident training. From formula to GPU in one language.

One pull is a complete working environment: the compiler, the GNU assembler and linker it drives, and gdb ready for the DWARF debug information Mica emits. Nothing else to install.

docker run -it micalang/mica

The container opens a shell, and the compiler answers:

mica --version

⁠Try it in one minute

Inside the container, open a new file in the editor:

nano Hello.mica

Type or paste this program:

program Hello;

imp
    WriteLn : std;

begin
    WriteLn("hello from mica");
end.

Save it with Ctrl+O and Enter, and leave the editor with Ctrl+X. Then build and run it:

mica --compile --link --optimize release,full_optimizations --source Hello.mica --build build
./build/Hello

hello from mica

The compiler builds for the machine it runs on, x86-64 or ARM64, whichever Docker gave you. The result is one native binary: no runtime to install, no interpreter, nothing but the C library every Linux system has. Naming the other architecture, as in --platform linux,arm64,utf-8, builds for that one.

⁠The full development setup

The image is also a ready-made VS Code dev container. Clone the starter repository and reopen it in the container:

mkdir -p ~/source/repos
cd ~/source/repos
git clone https://gitlab.com/mica-lang/mica-container.git
code mica-container      # then: Reopen in Container, and choose Mica Developer

The container opens on ~/source/repos, the folder that holds the starter, and make -C mica-container/examples/Hello run in its terminal builds and runs the first example.

With the Mica extension⁠ you get diagnostics, navigation, hovers, and notebooks, and F5 compiles and debugs the file you have open under gdb. The repository carries the whole example suite — from first programs to a GPT trained in pure Mica — plus the benchmarks and the verified notebooks, all working out of the box.

⁠What is in the image

The image is deliberately lean: the C toolchains and the SDK install in seconds when asked for, so the pull stays small.

⁠Tags and platforms

• latest — the current release
• <version> (for example 7.6.0) — that release, kept permanently, so a pinned tag keeps building forever

Every tag is a multi-platform image for linux/amd64 and linux/arm64; docker pull selects your architecture automatically. During the build, each platform installs the released package and runs the compiler before its layer is kept.

⁠Learn more

• mica-dev.com⁠ — the language in one page, the facts, and the reference
• Start⁠ — the four roads to a running Mica: the dev container, this image, the packages, a machine in the cloud
• Learn AI with Mica⁠ — twelve chapters from what learning is, before any mathematics, to a GPT you train yourself
• Dragon⁠ — the compiler's backend as an SDK, and the course that builds a compiler on it
• The learn area⁠ — from an empty folder to your first debugged program in fifteen minutes, and onward to the twelve-chapter AI course
• Installing Mica natively⁠ — the .deb packages, if you would rather not use Docker
• The starter repository⁠ — this image's Dockerfile, the examples, benchmarks, and notebooks

⁠License

The Mica compiler is free for any use, including commercial work, under the Mica Compiler License 1.0. The Dragon SDK installed by mica-provision sdk is a separate package under its own license. Details: mica-dev.com/company/licensing⁠.