ZeroLog
Private preview
EN DE

Product

2 November: instant chat switching, a model picker that opens at once, tools on by default

Switching between chats no longer waits for the server, the model picker opens with its list, new chats start with tools on, and chats that had broken on a missing tool result work again.

User impact: Daily use

Chat feels faster, new chats use tools by default, and chats that failed after a missing tool result recover on the next message. No action needed.

Several changes to the chat went live on 2 November. Most of them are about waiting less.

Speed WAIT LESS Chats switch at once Typing stays light Picker opens with list Tools DEFAULTS New chat: tools auto Off for small models Own keys: always auto Repair BROKEN CHATS Result follows its call Old chats work again Start view in motion
The three areas of the chat on 2 November.
Text version

Speed: chat history cache, light typing, model picker from cache. Tools: auto by default, off only for small and uncensored models. Repair: every tool call gets its result right behind it.

Speed

The sidebar loads a chat in advance when you hover, focus or press it, so switching shows the chat without a pause. Rename, pin, move, delete and new chat update the list at once and are confirmed by the server afterwards. While you type, only the input box redraws. The model picker used to wait for the server every time it opened. It now shows a cached first page at once and refreshes it in the background.

During review we found that a late answer from one account could have been cached for the next account on the same browser. It was fixed before the release went live, and tests now cover the account change.

Tools by default

A new chat no longer starts with tools off. Tools are on automatically for every account. They start off only for small models (for example nano, mini, flash, haiku, up to 20B) and uncensored models. With your own key they are always on.

Repair

If a call went to a tool that was not loaded, its result was missing from the history and every later message in that chat failed with a provider error. Results now follow their call, a missing one gets a placeholder that tells the model it was not executed, and broken chats work again on their next message.

The chat start view also moved: glowing fibres run into the crystal, with a one-line input and four intent cards.

No prices or plans changed.

Back to Newsroom

Corrections belong here too. Feed · Docs · X · Legal & privacy