-
miin-k8s 0.5.0 Stable
miin-releases released this
2026-09-14 15:58:08 +02:00 | 0 commits to main since this releaseFeatures
-
License keys and a 30-day trial. miin k8s now runs with the full
feature set for 30 days, then keeps reading your clusters for free while
changes, shells and the assistant ask for a license key. Paste the key
under Settings → License; it is checked offline, nothing is sent anywhere.
The header shows the days left during the trial and from 30 days before a
key expires, and a single toast per start reminds you — no dialogs. -
The assistant can ask you a question instead of guessing. When a
diagnosis reaches a fork — roll the deployment back or look at the logs of
the crashed container first — it now offers the choices as buttons, and your
answer continues the same reply rather than starting a new one. Until now
the question was a sentence somewhere in a paragraph, and answering it meant
typing the answer out yourself. Your own wording is still possible; the
field for it sits next to the buttons. -
It keeps running when you close the window. The app now lives in the
menu bar: watches, log streams and alerts carry on instead of ending with
the window, and one click brings it back. Quit from the icon there. It can
also start when you log in. -
Critical alerts reach you as a system notification, with the cluster in
the title. Until now an alert waited in the window until somebody looked. -
OpenRouter is one click in the provider list. One key, and the models of
Anthropic, OpenAI, Google, Mistral and the open ones are all reachable —
useful when you want to compare what a model makes of a cluster without
opening an account everywhere. -
The model list leaves out what cannot call tools. OpenRouter offers over
four hundred models and a sixth of them have no tool support; such a model
does not answer worse here, it claims to have looked at your cluster without
ever having looked. Those are now filtered out, so the ones you can choose
are the ones that work. A model that is missing can still be typed in by
hand. -
A long model list no longer floods the settings. It shows the count and
opens on request — with four models the full list was right, with three
hundred it was a wall of buttons. -
The model list fills itself once a key is stored — no button to press
first, and no message when it works. The choice used to stay empty until you
found "Check connection", and next to it sat a field for typing a model name
by hand; that field is still there for the providers that publish no list,
but it is no longer the obvious path. -
Pick a model by searching for it. The list is a search box now, not a
drop-down to scroll: type any part of a name —opus,flash,70b— and
it narrows down, across every provider you have configured. Typing a
provider and a fragment together works too (openrouter flash). Typing a
model name by hand is still there, but only where it is actually needed:
providers that publish no model list. -
Models show what they cost. Where a provider reports prices — OpenRouter
does — the model list carries them: input and output per million tokens,
side by side with the context window. Between the cheapest and the most
expensive model lie three orders of magnitude, and that is worth knowing
before the bill, not after. Providers that report nothing show nothing;
there is no empty column.
Fixed
- The context list scrolls. With more contexts than fit on screen it
simply cut off at the bottom instead of scrolling. It took a kubeconfig
with sixty entries for anyone to notice. - The bastion error speaks your language. When a connection through a
bastion fails, the explanation — which machine could not reach what, and the
hint about a kubeconfig pointing at127.0.0.1— was always in German. It
now follows the language setting, and says what to check rather than only
what happened. - Two strings show up in English again. The number of live connections on
a port forward, and the confirmation after copying an address, were stuck in
German.
Downloads
-