Choosing a model

memcode runs on one model at a time, and you choose it. It never swaps models mid-session or picks a different one per task.

The first time you run memcode it starts on a sensible default so you can get going without a setup step. After that it remembers what you picked.

Pick a model

In a session, open the picker:

/model

Each row shows the model's name, its context window, and whether one of your own keys covers it. Pick one and the rest of the session runs on it.

If you already know what you want, /model takes the name directly:

/model opus

Your choice is remembered per project, and also as your personal default, so a new repo starts on the model you last chose instead of asking again.

For a single run without changing what's remembered:

memcode run --model opus "..."

When a model can't do the job

Models differ in what they accept. If you attach an image to a model that can't read images, or a PDF to one without PDF support, or send more than fits in its context window, memcode tells you and stops rather than quietly running the turn on some other model. Switch with /model, or run /compact to shrink the conversation, and send it again.

If a provider itself fails mid-request — an outage, a timeout — memcode retries on a declared backup model and tells you which one answered. That's the only case where something other than your chosen model runs.

Control how hard it thinks

Separately from the model, /effort sets how much the model reasons before it acts. It takes off, medium, high, or auto. Auto is the default, and the setting resets when you restart the session.

/effort high

Using your own keys

If you've added your own provider keys, memcode uses your key for any model it covers, so you spend your own quota rather than credits. See Authentication for how to add keys.