Changelog

the site is very unstable right now, and you may run into slow loading, errors, or interruptions.

we’re upgrading servers and making infrastructure changes while working to restore stability. the work is still in progress.

sorry for the disruption, and thanks for bearing with us.

We’re aware of an issue with chat branching on private characters and are currently working on a fix.

our search servers are under heavy load, so searches may be slow or occasionally time out.

we’re adding more capacity to improve performance. that work is still in progress, so you may continue to see slowdowns for now.

thanks for your patience while we work on this.

quick one, but annoying if it happened to you.

sometimes a chat message never actually reached our servers — network blip, whatever — but it still showed up like a normal sent bubble, cleared what you typed, then vanished when you reloaded. so it looked like you sent something you didn’t.

that’s fixed on web and in the app. if the save fails, the message gets rolled back and your draft comes back so you can try again.

The character search service is having downtime, our team is investigating the cause and should have a fix in production soon.

quick follow up on yesterdays memory post since reports kept coming in.

turns out there was one more piece of the experiment still live: if you deleted a message that your memory covered, it cleared the memory too. that hit around 15,000 chats between aug 18 and this morning. deleting messages doesn’t touch your memory anymore, ever, unless you specifically choose to clear it.

the good news is the bug was hiding memories, not destroying them, so we were able to restore them on those ~15,000 chats. if yours disappeared in the last two days go check, it should be back. yesterday we said lost memory wasn’t recoverable, we were wrong about that in the right direction. the only ones we couldn’t bring back are memories made after aug 18 that got overwritten, sorry about those.

also, branching was never eating memories. branches copy your memory over like they always have, people were just branching chats the delete bug had already emptied. and we fixed a thing where really long memories would fail to summarize with a “context exceeds the limit” error.

last notes: the toggle from yesterday is web only until the next app update, and it doesn’t show on proxy chats because none of this applies there. chats and messages themselves were never touched.

last week we started testing a chat memory change with a small group. that test leaked out of its group and ended up affecting everyone, and some of you lost memory text or watched bots forget things they should have remembered. sorry. here’s what went wrong and what we changed.

memory was replacing your old messages. the experiment was about saving context: once you had a memory saved, the messages it covered stopped being sent with the prompt. that was only ever meant for the test group, but the behavior shipped to everyone. for people with a memory saved, bots suddenly had less of the actual conversation to work with and replies got vaguer or forgot details.

this is now off for everyone, and it stays off unless you turn it on. there’s a new memory replaces old messages toggle in the chat memory panel (web and app). off is the old behavior: your memory is added on top of the conversation and nothing is removed. on is the context-saving behavior, if you want it.

a failed summary could wipe the memory you already had. if a summary generation errored, was cancelled, or produced nothing, the memory field could end up blank or holding half a summary — one save away from replacing good memory with nothing. now your previous memory stays on screen until real text arrives, gets restored if the run fails, and we refuse to save a blank memory at all.

reasoning text could end up in your memory. with some models the model’s thinking was streamed straight into the memory field. it’s filtered out now on web and in the app.

handwritten memory had no safe update path. if you’d typed your own memory instead of generating one, the “summarize since last update” option was hidden, so the only button you could reach was the one that replaced everything. that option now shows for handwritten memory too, and regenerating over existing text asks first.

summaries were being generated on their own. auto-summarizing was part of the same experiment. it can only run when you’ve turned on memory replacing old messages, and it has its own off switch next to it.

if a chat’s memory got emptied during this, we’re sorry — that text isn’t recoverable from our side. everything else, including chats and messages, was untouched. you don’t need to change any settings or restart your chats; the default is the way it worked before any of this.

some plus subscribers had much slower generations than expected for part of today. this is fixed now and response times should be returning to normal.

we were moving plus traffic across multiple cloud gpu providers when instability on one of the routes caused requests to fall back onto a smaller pool. generations were still completing, but that fallback pool ended up carrying way more traffic than intended and replies became slower during busy periods.

we corrected the affected route, rebalanced traffic across the available gpu pools, and verified that requests are reaching the right models again. no chats, account data, or subscriptions were affected.

sorry to everyone on plus who got slower replies while this was happening. you should not need to change anything or restart your chats.

deleting your account no longer erases it on the spot. it now starts a 30-day countdown, and you can sign back in and restore everything before that window closes.

during those 30 days your account is deactivated: you’re signed out everywhere, your profile is hidden, and your characters go private. if you change your mind, sign in again and you’ll get a restore screen with the deletion date plus restore / log out.

after 30 days the deletion is permanent and can’t be undone. if you meant to leave, you don’t need to do anything else.

jllm was having a bunch of errors earlier today where messages would fail or time out. this is fixed now, no chats or account data were affected.

what happened was our hosting provider did emergency maintenance on one of the servers in the gpu cluster. when it came back online some of the traffic routing didnt come back properly, so most requests still worked but some got sent to routes that werent accepting connections.

the error rate went from its normal ~0.3% to around 8% between about 11:30pm pdt on august 7 and 11:25am pdt on august 8. both normal jllm and short responses were affected.

we restored the traffic routing on every gateway and tested both model paths, everything is back to normal now.

sorry if your messages failed or you had to retry them!!

We just finished migrating our chat message database to new infrastructure. All your chats and messages came along — nothing was lost in the move.

The migration surfaced one bug: some of you got “An error occurred when deleting a message” when trying to delete messages sent after the switch. Same root cause could also stop forking from (or reacting to) those newest messages. A fix is deploying right now — once it lands, no action needed on your side, everything just works again.

Older messages were never affected. Thanks for the quick reports!

Janitor is currently experiencing downtime across core services, our team is investigating and we aim to have the issue resolved shortly.

fixed the all chats modal so it shows all your chats with a bot again, not just 10 — plus personas, portraits, message previews, sorting, and a proper total count.

Janitor+ subscribers can now add GIFs and stickers to comments on web.

Open the emoji picker in character reviews, public chat comments, or script comments, then switch to the new GIFs or Stickers tab. You can search the KLIPY library, browse what is trending, or quickly find media you have used recently.

Emoji, GIF, and sticker options now live together in the same picker, including when replying to another comment.

In response to feedback, we’ve increased the character limit for saved custom prompts to 500K (vs 10K previously).

We have rolled out a targeted update to JanitorLLM and Janitor+ models to reduce cases where a reply gets stuck repeating the same phrase or repeatedly returns to the same answer.

• detected loops fell from 6/64 generations to 2/64
• the highest-failure test fell from 5/8 loops to 0/8
• results across six general-purpose benchmarks stayed within -1 to +1.6 percentage points of the unedited model

This is also not a complete cure for every kind of repetition. So, while this is a big improvement, we will keep monitoring real chats and evaluating further improvements.

hey we had issues with moderation api and churned through a bunch for better results, working on fixes, should be more stable now tho

Janitor+ subscribers now have a new Profile Customization page in Settings.

You can choose how your public profile appears across Janitor, including:

• showing or hiding your Janitor+ badge
• enabling the shiny username style
• picking a custom username color
• showing or hiding your “Member since” date

Open Settings on web (mobile update soon) → Profile Customization to try it.

https://janitorai.com/profile-settings?tab=customization

the plus llm and free llm is being redeployed tonight with some updates to help repetition and stuttering

We’ve rolled out an early-access preview for janitor+ users of our new beta model, jllm-short. This model has notably shorter replies for a faster-paced chat experience.

It can be enabled inside the janitor+ settings. Please let us know what you think!

(Currently, the jllm-short preview is only available on web, it will be coming to the app shortly.)

We are aware of an issue causing text on web to show up weirdly, we have pushed a fix and this should be resolved in the next 10 minutes, apologies for the inconvenience.

all janitor+ subscribers now get access to an emoji library for comment reactions (currently only on web) and a shiny golden username across the site

We’ve removed the 100-persona limit on web, you’re now free to create as many personas as you need.

We’ve made it easier to duplicate personas from the My Personas page.

You can now quickly create a copy of an existing persona, making it easier to experiment with variations without having to start from scratch.

We’ve shipped some improvements to the web version of the personas page, this should be a better for power users with easier management and grouping!

Search now has a Creators tab. Discover recommends creators for you based on who the creators you follow also follow, tuned to highlight smaller creators, updated daily. You can now search creators by username (if they have 1 public character or more), with previews of each creator’s top public characters. Blocked creators won’t show up.

We’ve pushed a small fix to prevent long usernames from breaking profile cards.

If you’re using custom CSS on your profile, please take a moment to verify that everything still looks correct.

We are aware of issues with the trending page and are working on a fix.

Age verification has now gone into effect for UK users, as per the previous announcement. Users accessing Janitor from IP addresses located in the UK will need to verify their age using k-ID. This follows the same process as required in Australia and Brazil, more details available here: https://help.janitorai.com/en/article/age-verification-for-brazil-and-australia-t4opjk/.

Some profiles were unintentionally affected after we introduced additional security restrictions. Those issues have now been resolved, and affected profiles should be back to their original appearance.

Please be aware that all images used on profiles should be uploaded to the Media Library. Images loaded from external domains will be hidden in a future update, so we recommend migrating them as soon as possible.

plz follow us 🩷 @janitorapp, https://www.tiktok.com/@janitorapp

The text overflow issue on iOS should now be resolved.

If you’re still experiencing this bug, please fully close and reopen the app to ensure the latest fixes are applied.

If the issue persists, please contact Support so we can investigate further.

In order to ensure compliance with UK regulation, age verification will go into effect for all UK users later this week. This will follow existing procedures for users in Brazil and Australia using k-ID, about which more details are available here: https://help.janitorai.com/en/article/age-verification-for-brazil-and-australia-t4opjk/

Chutes and NanoGPT can once again be used directly as proxy providers without relying on third-party services.

If you use another service that’s not working as expected, please contact support.

We have identified the cause of a bug that was allowing chat messages to continue outside of the screen. A fix will be live in the next app update.

Users can now toggle autoscroll on or off from the Customize configuration modal. We’ve also moved the Text Streaming setting there, making it easier to find both chat behavior options in one place.

When autoscroll is enabled, the chat will continue to automatically follow new content, but only if you’re already at the bottom of the conversation. When disabled, the chat will only scroll when new messages arrive, giving you more control while the character is responding.

Additionally, background images no longer interfere with the scrollbar, making it easier to grab and drag it for faster navigation through long conversations.

Note: We’ll also be bringing this autoscroll option to the mobile app on future releases.

We’ve made more adjustments to the chat experience, including improved scroll handling and smoother autoscroll behavior, which should now feel less sticky overall.

We’ve also added a copy-to-clipboard button to the top-right corner of each message, making it much easier to copy content without fighting against the scroll.

Some other small adjustments were also made, and more improvements are coming over the next hours.

web chat page scrolling issues have mostly been fixed, will be making it pixel perfect over the next few days, thank you for your patience

we are aware of this bug and working on it, seems to have been caused by background updates in lots of major browsers yesterday

We’re aware some users have been experiencing weird scrolling behavior on the chat screen. More adjustments have been rolled out to improve the experience and address those issues.

Following the new design system we’ve been rolling out across pages like “My Characters” and “Settings”, the My Scripts page has received a small visual refresh. This update is purely cosmetic — no changes were made to the structure or existing features.

hey we had an issue with our database connection pooler and it was causing queries to fail, servers have been restarted and everything should be running smoothly again, sorry about the inconvenience caused

seems like browser updates broke chat page scrolling, a fix has been released and is deploying now

Some cosmetic updates just landed on the My Characters and Edit Character pages to make it easier to build your characters. The overall structure remains mostly the same.

Additionally, creators can now use “Save as new” to quickly create different versions of the same character. To use it, click the chevron next to the Update button. This will open a “New character” form with all data from the current one pre-filled.

We have identified and resolved the cause of a bug that was causing some mobile app users to need to close the app between each message, the fix will be released with the next app update.

API Settings, Generation Settings, and Generation Rules are now unified under a single “Model Settings” menu on mobile.

We have identified the cause of a bug with the continue message feature on our mobile app and have pushed a fix to resolve the issue.

We have reverted the new CSS editor due to infrastructure issues with custom CSS sanitization causing a small portion of custom CSS profiles to break. The feature will be back once this issue is addressed after additional testing.

We rolled out some improvements to JLLM that seem to have substantially improved performance across our benchmarks and user ratings, thank you all for the continued feedback and we’ll continue working on this to keep making it better.

account settings got a security tab. you can now see every active session with its device, location, ip, and last-active time, and sign out individual sessions or every other device at once.

connected sign-in methods (google, apple, discord, x.com, plus email) are also listed there, so you can link extra providers or disconnect ones you don’t use, as long as you keep at least one method on the account.

two-factor authentication is next — authenticator app and sms support is coming soon so you can require a code on sign-in.

sorry — one of our gpu hosts had been dead for a while and we were still routing traffic to it, which caused failed and slow generations for some of you. it’s been removed and traffic is healthy again.

public script pages got a full cleanup. they now have bigger cover images, cleaner creator details, created/updated dates, script stats, and better actions for cloning or reporting.

the page is also better at showing what a script actually is. the about section has more room for lore notes and creator context, while the source section stays separate so you can inspect the lorebook or code without the page feeling like one big editor.

we also added a proper used-by section with character cards, so scripts that power characters feel more connected to the worlds and creators using them.

generation settings got a bigger cleanup today. the api settings modal now has one parameters section with advanced sampler chips, per-parameter toggles, custom parameters, payload counts, and json copy/paste sharing for configs.

proxy config editing is cleaner too: configs now edit inline, have import/export/recovery actions, can test connections in place, and show a reload-required banner when the active proxy url changes.

behind the scenes, proxy config saves are stricter now. invalid urls get rejected, duplicate/oversized entries are cleaned up, and selected config ids are sanitized so settings stay tidy.

mobile also picked up the parameter controls, better labels, translations, decimal rounding, and copy/paste feedback.

jllm was down for ~15 minutes earlier from a bad deploy on our end. fully recovered now. sorry to anyone who got dropped mid-chat.

we know jllm has been getting stuck repeating itself sometimes, phrases doubling, single tokens looping until the response cuts off. this is a known issue at the model level so no amount of tuning fully solves it. it’s a limitation of the model architecture and we are working on a fix.

we are training a new model to fix this properly. in the meantime if you see a “bot got stuck repeating itself” notice, just reroll generation now stops early instead of filling your reply with garbage.

Version Control & History

Your scripts now have full version control! Every change is automatically saved, and you can view, compare, and restore any previous version.

Key Features:

• Automatic Version History - Every time you stop editing, the system automatically creates a version snapshot. No manual saving needed.
• Visual Diff Viewer - See exactly what changed between versions with syntax-highlighted diff views.
• One-Click Restore - Found a bug in your latest changes? Restore any previous version instantly.
• Published Version Tracking - Versions created when you publish your script are specially marked for easy reference.

How It Works:

Look for the new History icon in the script editor header.

forbidden words: fixed a bug where removing words from your forbidden words list wasn’t actually saving — refreshing the site brought them all back. clearing the list works correctly now.

continue: pressing continue on a message no longer deletes your other swipe alternatives. they stay around so you can swipe back to them. the dropdown was removed too — tapping the button now continues in place directly.

chat ui: fixed z-index issues on the chat page. the persona switcher now properly appears above the continue button, and the scroll-to-bottom button no longer sits on top of the persona dropdown.

generation: hardened the fallback path so blank message content no longer 500s. mobile users will get this as a codepush — fully quit the app or site and reopen to pick it up.

account settings: usernames are no longer permanent. you can now edit your username from profile settings on web and mobile.

limits: username changes are available once every 30 days. settings will show when your next change is available.

validation: new usernames still need to be available, and use letters, numbers, and underscores only.

chat: long chats should feel lighter now. the message list has been upgraded so big histories render less at once, scrolling stays steadier, and streaming replies follow the bottom more reliably.

input & personas: the chat input got small layout fixes, and persona switching is cleaner on desktop and mobile. source chat personas are easier to scan, long descriptions can expand, and there is now a quicker path to manage personas.

messages: message actions are cleaned up into a footer, ratings moved with the rest of the controls, and the scroll-to-bottom button should feel more direct.

if chat scrolling or persona switching still feels off, ping support.

settings & account: account page gets a cleaner layout. following and blocks sit together under connections. main persona editing moved to my personas (removed from profile settings). avatar section updated. sign out and unfollow also now have confirmation prompts so mis-taps hurt less.

navigation: old standalone following/blocks pages are deprecated; that flow is in settings until the new profile management page ships.

if somethings missing or in the wrong place, ping support.

llm issues with long chats are fixed. long context should work reliably now instead of throwing unk errors.

ping support if anything still looks off.

starting tomorrow at 9am PT, users chatting from brazil or australia will need to verify their age through k-id to keep chatting. this is due to regulatory requirements in those regions, not something we chose.

more details here: https://janitorai.com/news/announcements/24

the comment moderation was being way too aggressive and flagging a lot of normal comments as spam. we rewrote the moderator from scratch.

your comments on character pages should go through much more reliably now. we are still blocking the stuff that actually matters (phishing links, threats, hate, csam, self-harm), just not normal chatter, rp snippets, or random promo sounding text.

if you had a comment wrongly blocked recently, try reposting it. keep the reports coming if anything still feels off.

we have switched the base model of jllm over to a new architecture, heavily tuned with high quality gemini and opus data.

this is a major under the hood upgrade designed to improve logic and coherence. that said, a new base model means the writing style, pacing, and personality might feel a bit different than what you are used to.

we are reading all your feedback. if you are seeing weird quirks or do not love the new vibe, do not worry as we are actively tweaking and tuning the model based on what you guys are telling us. right now we are mostly working on solving repetition, getting more variety in the responses, and better gender locking on characters.

we should also have more optionality soon so you can pick in chat what version of the model you want to use.

thanks for bearing with us while we get it dialed in. keep the feedback coming.

the llm issue from yesterday has been fixed — we’re monitoring to make sure things stay stable.

we’re also spawning new gpus to add more capacity, and will keep an eye on them as they come online.

a few of you have reported weird spelling issues in responses. we’re aware and looking into it.

some of you are seeing slow responses, timeouts, and occasional errors again. it’s a background job queue issue — when certain batch jobs take longer than expected, the backend starts retrying them in a loop and things spiral from there.

we’ve identified the root cause and are rolling out fixes right now. expect things to stabilize over the next few hours.

sorry for the rough ride.

the intermittent errors and slow loading issues have been fixed. everything should be back to normal now.

some users may be experiencing slow page loads, timeouts, or errors when using the site. we’re aware of the issue and actively working on a fix. things should be back to normal soon — thanks for your patience.

character cards now show how many public chats are attached to them.

we were testing a CDN caching layer on character pages to speed up load times. a config issue meant some pages got served from cache incorrectly for a short window so some of you saw the wrong character page content. we caught it and rolled the change back within minutes.

your chats, api keys, account credentials etc are all served through completely separate authenticated endpoints. none of that was part of this caching layer and none of it was affected.

sorry for the scare. well be more careful rolling out stuff like this going forward.

hey everyone.

wanted to give you a clear picture of what went on with the proxy settings over the past day.

what happened

we redesigned the proxy settings panel. the old system that managed proxy configs had been built up over a long time with a lot of messy legacy code underneath. when we rebuilt the UI on top of it, some of that old plumbing broke in a way that caused config data to get sent as empty values when you hit save. that is what wiped people’s proxy URLs, models and prompts.

the underlying issue was old technical debt that should have been cleaned up a long time ago. the panel redesign just exposed it. this also wasnt caught in testing because the bug only manifested in production builds.

were api keys leaked to other users?

no. we understand the concern and we looked into this thoroughly. every database query and API response is scoped to your user ID from your login session. there is no code path that returns another user’s data to you.

what people saw was their own old config data coming back from local browser storage after the server-side data got cleared by the bug. the new panel formats things differently so it looked unfamiliar.

what we fixed

• resolved the legacy data handling issue that was silently blanking configs on save
• saves now always go through to the server properly
• editing a config and clicking save now correctly captures everything you typed
• added a Recover button in proxy settings to restore previous configs from recent account history
• added backend protections so empty config data cannot overwrite existing configs

we are continuing to clean up the old proxy settings code and will keep you updated as we ship improvements.

We know many of you lost proxy configurations due to the recent bug and we’re sorry about that. To help, we’ve added a Recover button in the proxy settings panel that lets you restore your previous configs from recent history with one click. You will need to re-enter your API keys after restoring as they are never stored in history for security reasons.

We’ve also made additional reliability improvements to prevent saves from being lost going forward.

An update on the proxy settings bug. If you saved your settings while the bug was active, your proxy URL, model, and API key may have been overwritten with blank values. Unfortunately we are not able to recover those automatically — you will need to re-enter them. We’re really sorry about this.

We are also aware of reports claiming that API keys were leaked to other users. We have thoroughly investigated this and can confirm that it is not possible for your API keys to have been exposed to anyone else. Each user’s settings are stored separately and are only ever loaded for your own account. The bug was strictly in how the new settings panel read and wrote your own config — at no point was any data shared or visible across accounts. If you see an unfamiliar key in your settings, it is your own key displayed in the new proxy config format. That said, if you’re uncomfortable, rotating your key with your provider is always a safe precaution.

The issue is now fully patched and everything saves correctly going forward. Again, we apologize for the trouble.

Proxy settings issue is now fixed. We identified and resolved a bug introduced with the new API Settings panel that could cause your proxy configurations (URL, model, API key, prompt) to appear blank or get cleared when saving settings. Your existing configs were not deleted — the issue was in how the settings panel loaded and saved them, which made it look like data was missing. Everything should now save and load correctly. Sorry for the scare!

API Settings panel update has a big bug fix incoming..

Update

• Janitor LLM is being fixed - Sorry about the bad responses right now. This should be resolved in about 15 minutes.

Improvement

• Profile viewing reliability — Fixed intermittent failures when viewing profiles that contain large amounts of formatted text or HTML. The sanitization service has been upgraded so profile pages should load consistently instead of failing behind the scenes.

Improvement

• Dismiss button on viewed chats — You can now dismiss individual public chats from the “Viewed Chats” section on the homepage. Viewing the chat again will bring it back.

Improvement

• Recent chats on the home page — Your recently opened public chats now surface on the homepage so you can jump back in without hunting through history.
• Public chat cards — redesigned
• Simpler modal — Search was removed from the published-chats modal so loading and sorting stay the focus.

Fix

• Persona no longer overridden after forking — When sending new messages in a forked chat, your own persona now displays correctly instead of showing the original publisher’s persona.

Status

• Moderation API — Some of you may see errors when publishing chats, saving characters, or other flows that run through automated moderation. We’re working on fixes on our side. Thanks for your patience while we iron this out.

Improvement

• Fork attribution chain - Forked chats now show the full fork history on the fork-point message, so you can click back through each previous fork all the way to the original public chat.
• Fork any message - You can now fork from both bot and user messages in private chats, not just bot messages.
• Better persona handling - Forked chats now correctly display the original publisher’s persona name and avatar on historical messages instead of your own.
• Public chat card layout - Chat titles can now wrap to two lines, and the stats ribbon has been moved to the bottom-right to prevent overlap.

Fix

• Fork banner on correct message - The “forked from” banner now appears on the actual message you forked from, not the first message in the chat.

Improvement

• Better image resizing - Images in the character description editor can now be resized by dragging the corner handle. The resize behavior is smoother and preserves aspect ratio automatically.
• Cleaner image controls - When you select an image, a small alignment toolbar appears above it so you can quickly left-align, center, or right-align without hunting through menus.
• Images no longer appear squished - Fixed an issue where images in character descriptions could render with distorted proportions. Images now display at their correct aspect ratio.

Fix

• Editor no longer flashes on focus - The description editor border no longer changes color when you click into it, reducing visual noise while editing.
• Image alignment works correctly - Previously, centering or right-aligning an image could break after saving. Alignment is now preserved reliably.

Improvement

• Comments update in real time - Public chat comments now appear faster and stay in sync more reliably, so it’s easier to follow along as conversations happen.
• More responsive public chats - Likes, favorites, and other public chat activity now feel smoother and more consistent while you’re viewing a chat.

Fix

• Better mobile comments panel behavior - On smaller screens, opening and closing the comments panel now feels more natural and behaves more consistently.

Fix

• Image resizing & alignment – Fixed image resizing and alignment in the character bio editor. Images can now be resized by dragging handles and aligned left, center, or right via a toolbar that appears on click. Alignment and size are preserved after saving.

Improvement

• Resizable editor – The character bio text area can now be vertically resized by dragging the bottom edge.

Improvement

• Published chats panel – The published chats section on character pages now displays as a fixed-height scrollable panel instead of expanding the full page. Cards are lazy-rendered so only what’s visible is loaded into the DOM.
• Smoother sort & search – Switching between Newest / Longest / Most Discussed and typing in the search box no longer causes the list to flash blank. Previous results stay visible while new results load in.
• Better search – Search now uses fuzzy matching so typos still find the right results, prefix matching for partial words, and exact username boosting so searching a username surfaces their chats first. Searches across title, description, tags, and publisher username.

New

• Published chats on all characters – You can now allow or disable public chats for every character you own in one place under Account settings, instead of only toggling each character on its own page. This was a highly requested quality-of-life improvement.

Bug fix

• Chat on mobile – After turning on “NSFW on Mobile” in web profile settings, some users still saw “content not available” in the app. The mobile session could keep an older login token that didn’t include the updated setting. Generation now checks your saved profile setting on the server when needed, so mobile chat matches what you enabled on the web.

Bug fix

• Character data caching – Fixed an issue where characters were being cached and served to other users incorrectly due to CDN caching behavior. Character data should now display correctly for each user.

Service notice

• Image moderation – Moderation is back online. Thank you for your patience.

Service notice

• Image moderation – The moderation API is currently unavailable. We are working on restoring it. We apologize for the inconvenience.

bug fix

• My Chats – message counts per chat were showing as 0 instead of the real number; a recent refactor to how numbers are formatted stopped accepting the values the backend was sending. counts are accurate again.

Maintenance notice

• Moderation – Automated moderation is being updated; checks may be briefly delayed or inconsistent for a short window. Everything should be back to normal in a moment. Thanks for your patience.

service update

• forking chats creating a new branch from a message works again; a recent backend change was querying the wrong place for message data.
• thanks for the quick reports and logs; they pointed straight at the failure

service update

• trending we repaired search index data so the trending tab reflects real recent activity again instead of surfacing old high-traffic characters by mistake
• what went wrong some character records had incorrect publish timestamps in search, which broke the trending filters
• thanks for the reports while we tracked this down

service update

• trending we’re aware the trending list isn’t displaying correctly and are actively working on a fix
• what’s happening some users are seeing a different set of characters than expected on the trending tab
• fix incoming we’ve identified the issue and a fix is in progress

service update

• status the performance issues from yesterday are probably fixed but were still monitoring to make sure everything stays stable
• if you still see slowness please let us know and well dig in. thanks for hanging in there

service update

• whats happening were aware that parts of the site have been slower than usual including login, registration, and general navigation. this is not the experience you should be having and we take it seriously
• what were doing our team is actively investigating and rolling back recent infrastructure changes that may be contributing to the slowdown. were deploying fixes as we identify them
• current status performance should begin improving shortly as fixes go out. well update this post once everything is fully stable
• we appreciate your patience, we know how frustrating slowdowns are especially when theyre across the whole app. fixing this is our top priority right now

improvements

• thinking vs writing when using enhance my draft or write something for me with reasoning models, thinking content no longer appears in the input. instead youll see “thinking…” above the input while the model reasons, then “writing…” once real content starts streaming
• new command button the slash command button now uses a glass-style sparkle icon that lights up on hover or when the command palette is open
• available on both web and mobile

new feature

• slash commands press the / button in the chat input to open the command palette
• enhance my draft rewrites what youve typed to sound more natural and in-character
• write something for me generates a brand new message from scratch using chat context, no draft needed
• review flow after generating you can keep, regenerate, or discard the result before sending
• available on both web and mobile

service update

• what happened one of our databases ran out of ram, failing to return chat counts
• status the issue has been resolved and all systems are back to normal
• we apologize for the inconvenience and appreciate your patience while we worked through it

service update

• issue identified some users were experiencing errors when sending messages with janitorLLM
• fix weve identified the root cause and a fix is rolling out now. things should stabilize shortly

maintenance notice

• model response time The model is slower than normal while we rebuild servers
• status this is temporary and speed should improve as rebuild work finishes in the next couple of hours

were now routing about 1 in 5 requests to a new inference setup where the model thinks before it responds. it reasons through your message internally before writing anything back

you dont need to do anything. if you get routed to it, responses might take a beat longer to start but should feel noticeably better.

more to come

Valentine’s Day 2026

• Valentine’s Day 2026 Badge – all creators who participated in the Valentine’s Day 2026 event by tagging their character with TheValentine have been awarded an exclusive profile badge. thank you for participating!
• Event Banner Removed – the Valentine’s Day 2026 event promotion has been removed from the character creation form as the event has concluded.

Enhancements

• Unified Login & Register – we are rolling out a new combined login and registration screen. instead of separate pages, you now enter your email first and we detect whether you have an account. if you do, you will see your available sign-in methods (password, Google, etc). if not, we will create an account for you automatically
• OAuth provider visibility – the new flow now shows you which provider (Google, Discord, etc) you originally signed up with. this should cut down on the “i can not log in” issues we have been getting tons of support tickets about — usually caused by people trying to use email/password when they signed up with Google, or vice versa
• OTP by default, password still available – the default sign-in method is now a one-time code sent to your email. this helps protect against bot attacks and unauthorized logins. you can still sign in with your password — just tap “use password instead” on the code screen

this is an experiment gated via statsig, so it will roll out gradually. some users may still see the old login flow while we monitor feedback and stability.

Heads Up

• Enhance Message – we’re updating the backend for the AI message enhancement feature. you may experience unexpected behaviour until the new version fully rolls out. thanks for your patience!

Enhancements

• Onboarding – we improved the onboarding flow for new users to make it smoother and easier to follow. This is an experiment, so only some users will see the change while we collect feedback.

Incident Report

• DDoS Attack – we experienced a bit of a downtime, took a bit longer to restore as i was afk. ALso happy valentines day everyone!!!

New

• Moderation Emails – you will now receive an email for each moderation action taken on your account including bans, mutes, content removals, and verification changes. emails include the action type, reason, duration, and a link to open a support ticket if you think something was a mistake.

Enhancements

• Login & Register – we’re testing new sign-in and account-creation flows (UI only). some users will see the new layout, and it may switch on or off across devices while we run the experiment.

App Update v1.2.3

• Proxy Fixes – resolved issues with proxy connections that were affecting some users
• New Onboarding – refreshed onboarding experience for new users
• Chat Backgrounds – you can now customize your chat backgrounds
• Copy to Clipboard – easily copy messages from chat
• Bug Fixes – various bug fixes and performance improvements

Resolved

• Emails – the issue causing some users to not receive password reset and login emails has been resolved, if you were affected please try again

Known Issue

• Emails – some users may not be receiving transactional emails like password resets and login confirmations, we are aware of the issue and actively working on a fix

Post

• Mobile App Bug Reporting – we’re working as hard as we can to address the mobile bugs we’re seeing reports of, if you run into any mobile-app specific bugs please submit them on the help site or at support@janitorai.com, we’re keeping a close eye on that inbox as we identify new bugs

Bug Fix

• Mobile NSFW Toggle – fixed an issue where toggling “Enable NSFW on Mobile” from the web settings wouldn’t take effect on the mobile app until the session token expired, the setting now applies immediately

New Feature

• Notification Settings – added a toggle to disable scheduled character publish countdown notifications, you can find it in your notification settings on both web and mobile

Incident Update

• DDoS Attack – the site was taken down briefly today due to a DDoS attack, everything has been resolved and things should be back to normal. you may see more Cloudflare captchas than usual for the time being as we keep protections elevated

Bug Fixes

• Scheduled Characters – push notifications for scheduled countdown releases now show the countdown placeholder instead of leaking the character avatar sorry about that, completely missed it

Community Guidelines Update

• Increased Transparency – we’ve published updates to our Community Guidelines with a new policy structure and more guidance about what is and isn’t allowed on the site.

Moderation Update

• Improved Comment Filtering – we’ve tuned our comment moderation system to reduce false positives. playful or joking comments should no longer be incorrectly flagged as harassment while still catching actually harmful content

Event System Update

• Event Tag Restrictions – event tags like #TheValentine can now only be added to characters created after the event started. this prevents older popular bots from dominating event pages and gives newer creators a fair chance to shine

New Event

• #TheValentine Event – the valentine event is now live! add the #TheValentine tag to your characters to feature them in the event with exclusive badges and themed styling

Backend Updates

• Chat History Timeouts – fixing timeout errors affecting getUserChatTotals, getRecentChatsForUser, and getCharactersWithChatsForUser queries

Backend Updates

• Large Message Handling – improved how we handle very long chat histories to prevent timeouts and improve reliability
• Context Upgrades – various improvements to how context is managed for better conversation quality

Mobile Updates

• Search Screen – added a dedicated search screen to the mobile app for easier character discovery
• General Improvements – various performance and stability fixes across the mobile app
• app Releases – pushed latest updates to the beta testers

UI Improvements

• Profile Badges – fixed an issue where having lots of badges would squish the avatar. badges now wrap into multiple rows when theres too many to fit. if you have custom css for badges you may need to update it - new classnames are .profile-badges, .profile-badge, and .profile-badge-img
• Notification Text – improved notification previews to show more than 2 lines of text so you can see more context without opening
• Username Display – usernames now consistently show the @ symbol prefix across the site for better clarity

Maintenance Complete

• Database Migration Done – the database upgrade finished successfully! if you’re seeing “unauthorized” errors or having trouble logging in, just log out and log back in. sorry for the inconvenience and thanks for your patience!

Maintenance Update

• Database Migration Rescheduled – we had to redo the migration process which caused some delays. the actual database migration will now happen around 2am PST (in about 45min). the site should go down for ideally around 5 minutes. thanks for bearing with us!

Maintenance Update

• Database Upgrade – the database upgrade process has started and is progressing smoothly. we expect the site to go down for around 10 minutes around 12:00 AM PST (8:00 AM UTC) tonight. thanks for your patience!

Analytics Improvements

• Clearer Metric Names – renamed analytics labels so its more obvious what each stat means. impressions is now “shown in feed”, clicks is “card clicks”, views is “page views”, etc

Deployment in Progress

• Full Fix Rolling Out – deploying the complete jllm fix across all machines. should be fully resolved in approximately 30 minutes as each machine updates

Issue Resolved

• JLLM Response Fixed – the no response from bot issue has been resolved. was caused by incompatible hashing function for cache. all systems back to normal

Known Issue

• JLLM Response Errors – were currently investigating issues where users are seeing no response from bot toast messages when using janitor llm. working on a fix now and will update when resolved

System Updates

• Model Optimization – rolling out a new FP8 quantized version of jllm optimized for our H20 GPUs. This update should improve memory efficiency and allows for longer context handling will be testing it at 16,384 tcntx. Will be monitoring.
• Rolling Update In Progress – The transition to the new model configuration is happening gradually across the gpus. You may experience slightly slower reply times or increased queueing until all GPU nodes have fully switched over to the new deployment.

Scheduled Maintenance

• Database Upgrade – We will be performing a database upgrade tonight between 12:00 AM - 2:00 AM PST (January 27th) for one of our databases, this will increase stability of the platform going forward extremely as we are two majore software versions behind right now for postgres. Expect approximately 5-10 minutes of downtime during this window. A notification will be sent out 1 hour prior to the maintenance.

New Features

• Show Thinking – added a new toggle in Generation Settings to show or hide the model’s thinking process. enabled by default, disable it if you prefer cleaner responses without the reasoning visible

Improvements

• Thinking display – improved support for reasoning models that use different response formats, thinking content now displays correctly across all supported providers

Bug Fixes

• Scheduled Characters – fixed an issue where users were unable to edit their scheduled characters after the release time had passed, receiving a “scheduled release time must be in the future” error

Maintenance Notice

• Notifications – Notifications will be temporarily unavailable for approximately 10 minutes as we upgrade servers. We apologize for any inconvenience.

Bug Fixes

• Scheduled Characters – Muted users can no longer create scheduled character releases. They will receive an error message indicating how long they are muted.

Improvements

• Notifications Widget – Replaced the “load more” button with infinite scroll. Notifications now load automatically as you scroll down.

Changes

• Notification Counts – Unread notification counts have been reset. Counts will now track accurately going forward. Sorry for the inconveniance.

Bug Fixes

• Frontend Card Styles – Fixed an issue where the scheduled release countdown overlay broke custom CSS on profile character cards. It now uses a placeholder image instead of adding overlay elements.
• Backend Chat Protection – Implemented a new backend check to block starting new chats with scheduled characters before their release time, returning a 423 LOCKED error.
• Character Edit Form – Fixed a bug where the scheduled release mode was not correctly loading when editing a character.
• Search Synchronization – Fixed a sync issue where updating only the scheduled release time (without changing public/private mode) would not update search results.
• Race Condition Prevention – Fixed a race condition where a scheduled character could be re-published even if it had been force-removed or deleted by an admin.
• Unified Avatar Control – Moved scheduled release avatar control to the backend, ensuring placeholders are consistently shown across search, profile listings, and character views for everyone except the creator.

New Features

• Schedule character releases – Choose when your characters go live instead of publishing immediately
• Public Countdown mode – Character page visible with countdown timer showing name, description, and tags. Avatar and definition hidden until release. Users can favorite and comment while waiting
• Silent Schedule mode – Character stays completely private until scheduled time with no visibility
• Follower notifications – Public Countdown notifies followers when countdown starts and ends; Silent Schedule notifies only at release
• Schedule editing – Adjust release times, switch Silent to Public Countdown, or cancel schedules

Notes

• Previously published characters cannot be scheduled
• Public countdowns commit you to release since users are already waiting

Bug Fixes

• Fixed scheduled release countdown overlay breaking custom CSS on profile character cards - now uses a placeholder image instead of adding overlay elements

Bug Fix

• Continue latest chat now works correctly with personas - previously when you selected a persona on a character page and clicked “Continue latest chat” it would open your most recent chat with that character regardless of which persona was used. now it correctly finds and continues the latest chat that matches your selected persona. if youre using your main profile (no persona) itll find chats without a persona. if you select a specific persona itll find the latest chat with that persona

Ban execution

We have seen some reports coming in regarding users being banned. This is due to us executing full bans on a set of previously shadowbanned accounts. If you feel as if this action has been made falsely please make a support ticket at help.janitorai.com.

UI & Design Updates

• Sticky profile tabs - Profile page tabs now stay visible while scrolling
• Dialogue display - Improved how character dialogue appears in messages
• Markdown support - Better formatting and dialogue parsing in chats
• Notification spacing - Fixed padding issues on web

Bug Fixes

• Dialogue marking - Fixed how dialogue is displayed in messages

Improvements

• Cleaner interface - Removed outdated links and cleaned up navigation
• Better search - Improved profile and character search

UI & Design Updates

• Better text input - Redesigned chat input for smoother typing
• Color picker refresh - Easier way to customize chat colors
• Block management - Improved interface for managing blocked content

New Features

• Tag blocking - Block content by tags for better control
• All chats view - See all your character chats in one place
• Dedicated block screen - Easier management of blocked users and content
• Safety improvements - Added crisis detection and safety modals
• AI content notice - Clear disclaimers about AI-generated content

Bug Fixes

• Persona editing - Fixed bug where editing created duplicates instead of updating

Improvements

• Tag management - Restored ability to unhide tags you’ve hidden
• Removed links - Cleaned up settings by removing outdated links

Performance

• Optimized chat loading - Smarter loading for desktop

UI & Design Updates

• Image uploads - Better image rotation and loading when uploading
• Notification redesign - Cleaner more readable notification cards
• Persona avatars - Default bucket-dog avatar for personas without custom images
• Character analytics - New insights for character creators

Bug Fixes

• Proxy fixes - Resolved issues with Deepseek and other proxy providers
• Thinking indicators - Resolved problems with thinking tags not showing correctly
• Message buffering - Fixed issues with incomplete messages on some providers

Performance

• Faster notifications - Improved loading with better caching
• Optimized chat loading - Smarter loading for mobile

Updates

• Notification gestures - Swipe actions and smoother animations
• Mobile descriptions - Fixed character descriptions displaying correctly

Bug Fixes

• External links - Improved handling and validation of links
• Live updates - Favorite counts now update instantly

Bug Fixes

• Notification screen - Fixed notification display issues
• Settings screen - Fixed mobile settings screen bugs
• Chat labels - Clear indicators for deleted or private chats
• Quoted messages - Better styling for quoted text
• Line cutoff prevention - Prevented text from getting cut off

New Features

• Native settings - New native settings screen

Updates

• iOS improvements - App updates and bug fixes
• Notification improvements - Better mobile notification handling

New Features

• My Page migration - Complete native migration of My Page
• Chat forking - New chat forking feature
• Birthdate validation - Added birthdate validation for profiles

Bug Fixes

• Persona groups - Improved Formik integration for persona groups
• Loading states - Better loading indicators

Improvements

• Translations - Added pronoun translations
• Avatar improvements - Simplified user avatar component
• Form improvements - Better form handling with Formik

New Features

• Search and sorting - Added search and sort to character lists
• Character lists - New character list views
• Pull to refresh - Refresh chats on home screen pull to refresh

Bug Fixes

• Text cutout - Fixed text getting cut off with line height fixes
• OAuth fixes - Resolved OAuth connection issues
• Notification loading - Improved notification loading states

Updates

• iOS version bumps - Multiple iOS app updates

New Features

• Character analytics - New analytics page for character creators
• My Page native - Migrated My Page to native implementation
• Personas on My Page - Moved personas to My Page with tabs
• Scripts on My Page - Added scripts section to My Page
• Home screen - New home screen implementation

Bug Fixes

• Chat screen issues - Fixed multiple chat screen bugs
• Text cutoff - Prevented last line from getting cut off
• Input overlap - Fixed input overlapping issues

UI & Design Updates

• Better skeletons - Improved loading states
• Connected accounts - Better loading for OAuth connections

UI & Design Updates

• Custom fonts - Added custom font support for better typography
• Persona groups - New group management for personas
• My chats navigation - Improved navigation in my chats screen

Bug Fixes

• Scripts config modal - Fixed scripts configuration modal issues
• Persona creation - Fixed bugs when creating and selecting personas
• Overlapping input - Prevented input from overlapping other elements
• Shake bugs - Fixed shake gesture bugs on mobile

New Features

• Notification settings - Notification preferences in native app

Improvements

• Better tag toasts - Improved tag notification messages

New Features

• Group filtering - Filter personas by groups
• Group selection - Select and manage persona groups
• Pronouns support - Added pronouns to persona profiles
• Avatar management - Better avatar upload and removal

Bug Fixes

• Avatar fixes - Resolved avatar display issues
• Header fixes - Fixed header display problems