Choose the model behind compaction, checks and planning
You can now pick which model summarises long chats, checks answers, judges goals and plans runs. Without a choice, nothing changes.
Alle können jetzt unter Einstellungen → Modelle & Agenten wählen, welches Modell Nebenaufgaben wie das Verdichten langer Chats, Antwort- und Zielprüfung oder das Planen erledigt. Ohne Wahl bleibt die Vorgabe von ZeroLog. Keine Aktion nötig.
Every chat leans on smaller models for side work: one summarises long chats so they fit the context window, one checks answers and tool results, one judges whether a /goal is reached, one plans /ultracode runs. Until now ZeroLog picked these for everyone.
Under Settings → Models & Agents → Helper models you can now set a model for each of these tasks. The list shows what each one does. Leave a row empty and the ZeroLog default stays in place. Paid models are charged to your balance, like any other model.
If a model you picked is later paused or no longer available to your account, that task falls back to the default instead of stopping. Chats that run on your own API key keep using the helper model from the provider card, so they never spend platform credit.
The same section is in the Android app and in ZeroLog Desktop. We also fixed a case where chat compaction stayed stuck for accounts without a spending cap after the summariser's provider refused a request.