Without --tag the image is named morloc-<env>:<morloc version>. The image is the environment’s own image with the rest of the environment added: the Morloc runtime, the toolchain installed from the environment’s pixi.lock, and every installed program with its source directory. If a required part is missing, for example because the environment was never fully built, freeze names it and stops.

The result is a tag in the engine’s image store; nothing is written to your working directory. When the build finishes, freeze prints commands to try the image: open a shell in it, list its programs, and run the first one.

Each installed program travels as the copy of its project directory that mim install made, so whatever else sat in that directory goes along. Before building, freeze lists every program with its size, and flags large files and tool state:

--force answers yes to all of these, tool state included.

The image’s default command serves the environment’s views, so the mim view declarations from Serving and Eval travel with it. Inside the container the server always listens on port 8080:

Morloc moves data between languages through shared memory, and container engines give a container only 64 MB of it by default; --shm-size raises that. The size the environment used is recorded on the image as the label morloc.suggested-shm-size.

Inside a container the server has to listen on every interface, or a published port could not reach it. So the image does not decide who can reach it; you do, with the -p mapping (127.0.0.1:8080:8080 publishes on the host’s loopback only) and the network you attach it to. With no token the image serves openly and says so in its log; set MORLOC_MCP_TOKEN to require one, exactly as with mim start.

Eval is the exception. It stays locked without a token. To serve it without one, because something in front of the container already checks callers, set MORLOC_EVAL_ALLOW_NO_AUTH=1.

The launchers of the installed programs are on the image’s PATH, so you can also run a program directly, with no server involved:

An environment with no views gives an image with no default command; it holds the programs and runs them only when named, as above.

To carry the image as a file, add --save. This writes the engine’s own image archive, which includes every layer and loads on any machine with the same engine:

To share it through a registry, tag and push it with the engine:

Once it leaves mim, nothing tracks the image. It describes itself in labels instead: the Morloc version, the environment it came from, its programs, and which modules answer on each adapter.

mim status, logs, and stop act on servers started with mim start, not on containers you run from an image.

The image above is the full environment, compilers included. That is what lets it evaluate expressions, which compile code at request time. When the programs are all you need, --slim builds an image without the compilers, the Morloc compiler, pixi, or build headers, keeping every interpreter and package the programs use:

After building a slim image, freeze checks that every program still starts and that the runtime and every compiled pool still find their libraries. A slim image cannot evaluate, so an environment with eval enabled is refused: turn it off with mim view eval --off, or freeze without --slim. A program that compiles code while it runs, such as a Python package built on first import, also needs the full image.