Character Tavern

Changelog

What I've been building

Every update to Character Tavern, written as it ships. Big launches, small fixes, and the occasional thing you asked for on Discord.

Written by Sal

Get pinged when something ships

Join the Discord to hear about new features and fixes as soon as they go live.

Join the Discord
  1. Latest Chat

    CAM now remembers every chat

    CT AutoMemory (CAM) is out of Experiments and on in every chat, so even a 1,000-message chat remembers how it started. If you'd rather keep writing your own summary, the new Legacy sheet turns CAM off, and Legacy Prompt Settings lives there now too.

    New

    • CAM is on in every chat with no Experiments toggle, and the first time, a long chat catches up within a minute or two of your next message.
    • Legacy Memory, in the new Legacy sheet under Extras in Settings, turns CAM off so replies read your Summary in Memory Settings again, and CAM keeps its notes for when you switch back.

    Improved

    • If you had turned CAM off in Experiments, it's on again, and Legacy Memory is where you turn it off now.
    • Legacy Prompt Settings moved from the bottom of Writing Style and Prompt Settings into the Legacy sheet.
    • Legacy takes the Discord tile's spot under Extras, and Discord is still linked in the site footer.
  2. KI

    Anyone can link Kyara Intelligence, and Spark and Ember share every perk

    Linking a Kyara Intelligence account is now open to everyone, and your plan decides what you get on Character Tavern. Both plans get the same things. I recommend linking your account, but pasted keys keep working.

    New

    • Every account can link Kyara Intelligence from KI Mode, no invite needed.

    Improved

    • Spark and Ember get the same perks on Character Tavern, Max Mode included, so Spark users can turn it on.
    • A linked account with no plan now falls back to CT Direct on its own, and everything comes back by itself once a plan does.
    • Pasting a key now checks it first, and that key follows your real plan from then on, perks and badge included. Working keys already saved here moved over too.
    • Free Quill now comes with a paying plan. A connected account with no plan uses Puni for Quill, like everyone else.
  3. Account

    Moderation now happens on the site, appeals included

    When a moderator acts on your card, comment or account, you hear about it in the notifications bell and you can answer right there. That includes disputing a decision, so you no longer have to find a moderator on Discord for it.

    New

    • A row at the top of the notifications bell shows what needs your attention, and Dismiss for 48h quiets it without hiding it.
    • The new Moderation page, in the account menu and in Account standing, lists your notices, your history and your strike count.
    • Each notice says what happened, what it means, when a strike, mute or restriction ends, and gives you one button for what to do next.
    • Appeal this decision, on a notice, a strike or a finished review, takes one box saying why. You get one appeal per decision, and the answer comes on the site and by email.
    • A strike that gets overturned is removed, and stops counting toward your limit right away.
    • A card under review stays Private while you fix it: you can edit it, talk to the moderator on its page, and press Request review when it is ready. It goes back to how it was once the review is done.
    • If a moderator replaces your profile picture or banner, or changes your display name, you get a notice saying why, and you can change it again right away.

    Improved

    • When a moderator needs to reach you about your account, it opens a conversation you can reply to, instead of an email you can only read.
    • After a warning or strike, opening the notice comes first: saving a card as Public or Unlisted asks you to read it, and saving as Private always works.
    • Moderation emails link straight to the notice. A banned account can't log in, so its email still points to Discord.
  4. Chat

    Writing Style is out of Experiments

    Writing Style now writes every chat, starting on CT Universal unless you already picked a style, and you can save your own styles and give each character one. The old prompt presets are still one switch away. For now.

    New

    • Every chat uses Writing Style, with no Experiments toggle needed, and you start on CT Universal if you never picked a style.
    • Legacy Prompt Settings, at the bottom of Writing Style, brings back your preset, Custom Prompt, and Prompt Helpers as you left them, until I retire it.
    • Each character can have its own style: pick it at the top of Writing Style, and the characters you never set stay on Default.
    • Save as new style keeps your current setup under a name, custom prompt included, ready for any character.
    • Rename and Delete sit in the same menu, and deleting a style puts its characters back on Default.
    • If you still want a hard limit, Token cap under Response Length in AI Settings stops Writing Style replies at your Max tokens, mid-sentence if it has to.
    • Presets come in two rows: Play style has CT Universal, Chat, Cinematic, Expressive, Game Master, and Novel, and Mood has Slow Burn, Cozy, Grimdark, and Horror.
    • CT Universal brings the classic CT prompt to Writing Style: a detailed world that doesn't care about you, and loud dialogue.
    • Game Master gives you a dry narrator who runs the world and remarks on your choices.
    • Voice: Expressive writes speech the way it sounds, with shouting in caps, stretched and stuttered words, and sounds spelled out.
    • Retry, on Quill's latest turn, sends the same message again with whichever model you have picked.

    Improved

    • Writing Style's Length setting decides how long replies are, and Response Length and Smart Limit apply again when Legacy Prompt Settings is on.
    • Picking a preset sets Difficulty and Story drive too, so Grimdark starts on Harsh and Living world, and Cozy on Forgiving and You lead.
    • Changing Difficulty or Story drive after picking a preset keeps the preset.
    • Dialogue First is called Chat now, and your settings come along.
    • The Auto and First Person presets are gone: your settings stay as they were and show as Custom, and Perspective still offers First person.
    • When something you do lands on a character, they react to it before the scene moves on.
    • Non-human characters keep their paws, fur, tails, and snouts, written as they are.
    • Before replying, Quill proofreads its changes against your request and catches passages a rewrite dropped by mistake.
    • Undo this shows on every Quill turn, including ones that only chatted, and refills an empty message box with what you sent.
    • Amane is retired, and if you had it picked, your next chat switches to Michi.

    Fixed

    • Quill keeps its edits to what you asked for and offers other ideas in chat, after you flagged its uninvited rewrites on Discord.
    • When Quill thinks a passage you wrote needs work, it says so in chat instead of cutting or softening it.
  5. Chat

    Replies survive my updates now

    Replies no longer cut off halfway when I ship an update, and I ship most days. Quill got steadier too: it picks its own model, and when it stops partway it keeps what it finished.

    New

    • When a reply is written but can't be saved, Save reply saves it instead of writing a new one.
    • Quill starts on Auto, which picks the model for you and moves to another one when the first can't take your request.

    Improved

    • A tab you left open during an update loads the new version on your next click, no refresh needed.
    • Memory Settings asks before discarding unsaved Memory Snippet edits, like it already did for the Summary.
    • Mei 2 is now obsolete, so the Model Explorer hides it unless you check Show obsolete models.
    • When Quill stops partway, it keeps the edits it finished, and saying "continue" picks up the rest.
    • A Quill message that fails, or that you stop, doesn't use Puni.
    • When a model can't take your request at all, Quill says why and offers to send it to another model.
    • The card editor locks while Quill works, so you never have to sort out clashing edits afterwards.
    • In a long conversation, Quill makes room by letting go of its oldest messages and tells you when it does, while the card keeps everything.
    • A card name now needs at least 3 characters.

    Fixed

    • A reply still being written when I ship an update finishes instead of stopping halfway.
    • Fixed my bug where Experiments like CAM and Writing Style switched themselves off after 30 days.
    • Fixed my bug where a custom prompt that leaves the summary out made CAM forget the story while still saying it was up to date.
    • If your first chat with a card hits an error while CT copies the card, the next visit tries again, so your copy never ends up missing its greetings or lorebook.
    • The Smart Prompt Disabled warning only shows when your Main Prompt really leaves the summary out.
    • A card named only in Chinese, Japanese, Cyrillic, or other non-Latin letters saves normally, and the pages that list such cards load again.
  6. Chat

    Michi keeps up with new models so you don’t have to

    Michi runs the best free model under her own name and moves to a better one the day it arrives. Pick her once, and keeping up with new models becomes my problem.

    New

    • Michi joins the Standard models: she runs the best free model and switches by herself when a better one arrives.
    • Michi is the default for new accounts, and her card in the Model Explorer shows which model she's running now.
    • When a reply hits its length limit, a notice offers Continue to pick up where it stopped.

    Improved

    • Nyx 7 Mini moves from Premium to the Standard models and doesn't use Puni anymore.
    • Mei 3 Flash's free run has ended, and it moves to the Premium models.
    • Aria and Nyx 3 Mini are now obsolete.
    • The default Response Length is now Detailed, so fewer replies stop mid-sentence.
    • The Model Explorer is simpler, with search and filters in one row and the models right below.

    Fixed

    • Fixed my bug where a double tap on Continue stopped the reply it had just started.
  7. Chat

    Long chats remember how they started

    CT AutoMemory (CAM) is in Experiments: it keeps track of your story on its own, so even a 1,000-message chat remembers how it began. I tested it on chats as long as a novel.

    New

    • CT AutoMemory (CAM) in Experiments keeps your character on top of names, promises, and old scenes long after they scroll out of view.
    • CAM works in the background, is free, and doesn't use Puni.
    • A small note above the message box shows CAM catching up on a long chat, and says when it's up to date.

    Improved

    • Editing or deleting an older message updates what CAM remembers.
    • Memory Settings is now in sections you open and close: CT AutoMemory (CAM), Memory Snippet, and Summary. While CAM is on, Summary is grayed out.
    • AI Settings is now in sections too: Presets, Model, Reasoning, Response Length, and Advanced Parameters, each showing its current choice while closed.
    • Changes in AI Settings save as you make them, so the Save buttons are gone.
    • A model you star shows up in Favorites right away, and Manage Favorites lets you drag them into order.
    • In Writing Style, Length now counts paragraphs: Short is 1 to 2, Medium 2 to 4, and Long several, so replies stop padding to reach a word count.
    • Writing Style keeps characters from reacting to what only you narrated, and answers an out-of-character question directly, without a scene after it.
    • On a phone, settings panels drop the line under their title, so their content gets more room.

    Fixed

    • Saving Advanced Parameters no longer resets your Reasoning Effort.
  8. Chat

    Your stories have a difficulty setting now

    Writing Style now sets how often what you try works out, and whether the world waits for your next move. Nightmare is accurately named.

    New

    • Writing Style has a Difficulty setting with four steps: Forgiving, Fair, Harsh, and Nightmare.
    • Story drive offers You lead, where the world waits for your moves, or Living world, where people and events move without you.
    • Under Content warnings in the Studio, an advisory AI suggestion shows which warnings your definition seems to need.

    Improved

    • What you try in a story succeeds as often as your Difficulty allows.
    • Quill writes and labels cards by the Prohibited Content Guide.
    • The Prohibited Content Guide has a new layout that reads well on a phone, with each rule split into Prohibited and Allowed.
    • The Strike System section says each strike expires one year after it's given.
  9. Chat

    Summaries got a proper editor

    You can edit a chat summary right in Memory Settings, and pop it out fullscreen when it gets long. Mine always get long.

    New

    • The expand button next to Summary opens it in a fullscreen editor.

    Improved

    • A Summary Not in Your Prompt notice tells you when your custom prompt leaves the summary out, and how to fix it.
    • Memory Settings asks before closing with unsaved changes.
  10. Chat

    Thinking models stop ghosting you

    Some models spent the whole reply budget thinking and answered with nothing, which I find relatable. It happens less now, and when it does, you get an explanation instead of a blank reply, and it doesn't use Puni.

    Improved

    • A reply that comes back empty doesn't use Puni.

    Fixed

    • When thinking uses up the reply budget, the chat says so and suggests raising Max Tokens, turning on Smart Limit, or setting Reasoning Effort to None.
    • Reasoning Effort on Auto leaves room for thinking, so replies get cut short less often.
    • Improve Message only fixes typos now, instead of rewriting what you wrote.