site news

🚧 still cooking — here's what we're actively building

a few things work but aren't fully baked yet, bestie. here's the honest status so nothing surprises you 💅

🧠 long-chat memory — right now, when a chat outgrows your model's context, we quietly drop the oldest lines without warning you. we're building automatic memory and lore book support so you won't have to manage that yourself — and we're tuning it on real long chats, so the more you ramble, the more you shape it
🎛️ model & prompt recommendations — we're putting together simple house prompts and a shortlist of models we like for SFW and spicy chats. SillyTavern prompts work, but we don't support toggling their features on and off yet. we might add that if enough people ask, but for now we'll mostly steer you toward the simpler prompts

found a bug, or want something added? drop it in the suggestion box or message Zumi 💌

changing your key used to mean deleting the whole provider and rebuilding your model list from scratch, bestie ✦

🔁 every connected provider has a "replace key" button now — paste the new one right on its card and your model picks, their order, and your provider's spot in line all stay exactly where you left them
🔓 a fresh key starts with a clean slate — if your old key got rejected, the new one used to sit in the same 30-minute timeout anyway. swapping it in (or passing the test button) lifts that timeout on the spot
🛟 a bad paste can't wipe your old key — if the new one gets turned down, we tell you your old key is still saved and still working
🗑️ delete works even while the robots are still typing — hit delete mid-batch and it goes through right away. replies that hadn't finished get tossed, and takes that already landed stay put
✍️ your edits stop vanishing mid-sentence — while replies came in, the chat refreshed every couple of seconds and wiped whatever you were typing in an edit box. it waits for you now, and saving an edit while replies are still cooking actually saves
💬 a rejected edit keeps your words — too long or blank, the box stays open with your text still in it and tells you why

go rotate those keys like the security queen you are 💅

one expensive model turning you down was benching your entire provider, bestie ✦

🔑 one model saying "not enough balance" no longer benches your whole key — it used to lock every model on that provider for a full 30 minutes, including the cheap ones that were answering fine seconds earlier. now only the model that got turned down takes a breather, and it's about a minute
💬 the error finally tells you what actually happened — it used to insist every model you configured was failing, which was never true and never actionable. now it names the model, says whether it looks like your balance or a rate limit, points you at topping up or picking something cheaper, and gives you the real wait instead of a vague "give it a minute"
🖼️ picture safety checks stopped grounding your key too — a turned-down image check was quietly doing the same 30-minute lockout on a completely separate code path. it plays by the same rules now

go top that balance up and get back in the group chat 💅

the chat room was dressed for a giant desktop monitor and squishing everyone else, bestie ✦

💻 laptops finally get a chat layout of their own — the bar on top and the typing box on the bottom were sized for a tall monitor and eating over half your screen, so a laptop showed about two messages at a time. now the chat spreads out wider, the chrome slims down, and you see roughly twice as much of the actual conversation
👥 group chats stopped hiding the "who answers next" buttons — they were rendering underneath the typing box where nobody could click them. they sit where you can actually reach them now, on every screen
📱 turning your phone sideways shows the chat instead of burying it — the top and bottom bars used to overlap and cover the whole conversation. on a short screen they scroll out of the way now

go open a chat on the small laptop and actually read it 💅