<?xml version="1.0" encoding="UTF-8"?><rss version="2.0" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Newsroom - Posts</title><description>Updates and posts - Official posts and news</description><link>https://janitorai.com/</link><language>en-us</language><lastBuildDate>Thu, 19 Nov 2026 00:00:00 GMT</lastBuildDate><atom:updated>2026-11-19T00:00:00.000Z</atom:updated><item><title>A quick domain update</title><link>https://janitorai.com/news/announcements/domain-update/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/domain-update/</guid><description>We didn’t get hacked — janitorai.com is on hold after a false abuse report, so use jani.com while we work on getting it back.</description><pubDate>Thu, 19 Nov 2026 00:00:00 GMT</pubDate><content:encoded>![JanitorAI logo](/news/static/domain-update-logo.png)

We didn’t get hacked — janitorai.com is on hold after a false abuse report, so use [jani.com](https://jani.com/) while we work on getting it back.</content:encoded><category>post</category></item><item><title>Update on last week&apos;s search issues</title><link>https://janitorai.com/news/announcements/search-issues-update/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/search-issues-update/</guid><description>A quick update on the search issues last week. Things are back to normal, so I want to apologize for the inconvenience and explain what was wrong and what was changed.</description><pubDate>Mon, 14 Sep 2026 00:00:00 GMT</pubDate><content:encoded>![Search incident](/news/static/search-incident-cover.webp)

A quick update on the search issues last week. Things are back to normal, so I want to apologize for the inconvenience and Expedite the explanation of what was wrong and what was changed.

The main issues were with searching and browsing characters. It resulted in slow and sometimes invalid results, and even completely broken pages. Some of you were even unable to access the site at all.

According to Cloudflare, there was a huge spike in rate-limited traffic for web searches. We assume there was malicious traffic, but there were also issues with how our servers distributed and managed the load. There were also a couple of issues with our infrastructure which extended the time it took for things to fully recover.

We increased the capacity for search, reduced the cost of each query, resolved the issues with the cluster, and separated the processing for search from the other requests, namely the chat requests. We want to ensure that an issue with the search function doesn&apos;t cause an issue with the other functions.

Things are much better, and we really appreciate your patience during the inconvenient &quot;refresh and hope&quot; period. You should be able to find the characters you want and chat with them, rather than being unable to load the page.

For those of you who are interested in the details, I will include the more extensive explanation here.

---

## Technical Details (September 10 - 11)
- **Search Overload:** Search Servers were continuously running above 90% CPU, with thousands of search tasks queued. Basic listing requests were doing a lot of unnecessary work, including exact result counts and tag aggregations.
- **Increased capacity and cheaper queries:** We increased our Search Servers from three to ten nodes, added bounded counting for normal listings, and limited deep pagination to 10,000 results. This reduced unnecessary expensive requests and the need for counting to be repeated.
- **Recurring cluster-agent restarts:** A version and permissions mismatch caused the cluster-agent software on some workers to restart every 15 minutes. During the restarts, the workers were unable to serve traffic. We fixed the permissions issue and performed a series of one-by-one upgrades on the five control-plane servers.
- **Traffic-routing bottlenecks:** A number of our ingress machines were fully saturated with CPU work. We alleviated this issue by moving the work to a couple of machines in the cluster that had more CPU capacity.
- **Backend CPU usage:** A tag-validation check was found to be repeatedly rebuilding matching patterns. We prevented the check from performing the work, while keeping the validation rules. This helped, but was not the only cause of the CPU pressure.
- **Search isolation:** Normal character listings have been isolated to a dedicated API pool, which allows for concurrent requests to be scaled independently.
- **Deployment fixes:** Some of the recovery changes made during the incident have been replaced by automatic releases that occur one at a time. We added and tested safeguards to run production releases a lot more carefully with tests on ingress nodes before they are deployed to the rest of the cluster.</content:encoded><category>post</category></item><item><title>Similar characters are here</title><link>https://janitorai.com/news/announcements/similar-characters-are-here/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/similar-characters-are-here/</guid><description>Discover more characters like the one you’re viewing, right from its character page.</description><pubDate>Fri, 28 Aug 2026 00:00:00 GMT</pubDate><content:encoded>Found a character you love and want more like it? **Similar characters** are coming to character pages on web.

When you’re signed in, scroll down on a character page to find a new row of recommendations. Matches use the character’s public details and balance similarity with variety, helping you explore familiar themes and personalities without a wall of near-duplicates.

The row adapts to your screen, and you can select **Load more** to keep browsing. Characters you’ve already visited during the session are left out as you click through, helping every stop lead somewhere new.

We’re rolling this out gradually, and it should reach most people over the next few days. If you don’t see it yet, check back soon.</content:encoded><category>post</category></item><item><title>Introducing janitor+ Router</title><link>https://janitorai.com/news/announcements/introducing-janitor-router/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/introducing-janitor-router/</guid><description>Access third-party AI models directly through JanitorAI with transparent pricing, advanced generation controls, exclusive discounts, and $5 in free launch credits for janitor+ subscribers.</description><pubDate>Thu, 30 Jul 2026 00:00:00 GMT</pubDate><content:encoded>![janitor+ Router model selection](/news/static/SCR-20260730-nwek.jpeg)

Today, we’re launching janitor+ Router: a simpler, more integrated alternative to the third-party proxies many people already use with JanitorAI.

Router lets you access a wide range of third-party models directly through janitor, without copying API keys or configuring an external proxy. You can compare pricing and context limits, switch models, customize generation settings, and manage your balance, all in one place.

Right now, Router is only available to JLLM+ subscribers, and only available on web. Router will be coming to the mobile app soon.

For everyday roleplay and conversation, JLLM+ remains our recommended chat model. Router isn’t intended to replace it. It gives people more options while laying the groundwork for experiences that go beyond text chat.

## **Built for what comes next**

At its core, Router is infrastructure for bringing more types of AI experiences into JanitorAI.

That includes multimodal capabilities, richer creator experiences, specialized models, and more agentic tools. Instead of building each experience around a different external provider or proxy, Router gives us a unified foundation for connecting them directly to JanitorAI.

The initial launch focuses on model access and generation controls, but the bigger goal is to make entirely new kinds of creation possible.

## **A better proxy experience**

We know many people already connect external models to JanitorAI through proxies. Router offers a more seamless version of that experience:

- No external API keys or proxy configuration
- A growing selection of premium models
- Clear per-token pricing before you chat
- Easy model switching inside your settings
- Custom prompts and advanced generation controls
- A built-in wallet and billing experience

Usage is billed per token at the prices shown, so you can choose the model and spending level that work for you.

In addition, we’ve also secured exclusive discounts for janitor+ subscribers, offering 10% off some of the most popular models.

## **$5 in free credits for janitor+ subscribers**

To celebrate the launch, [janitor+ subscribers](https://janitorai.com/plus) can claim $5 in free Router credits.

Open your API settings, select janitor+ Router, and claim your credits from the Router wallet. Once claimed, you can choose a model and start exploring immediately.

This is the first step toward a much broader set of creative tools on JanitorAI. We’re excited to see what you do with it and where we can take Router next.</content:encoded><category>post</category></item><item><title>we redesigned the api settings page</title><link>https://janitorai.com/news/announcements/redesigned-api-settings/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/redesigned-api-settings/</guid><description>API settings on web got a full redesign. Cleaner sections, multiple saved custom prompts, simpler response length controls, and a proper save button. Coming to web soon.</description><pubDate>Sun, 19 Jul 2026 03:30:00 GMT</pubDate><content:encoded>we redesigned the api settings page on web. the old one was basically one giant form that you had to scroll through forever, and honestly it grew that way over time because we kept bolting settings onto it. it was overdue.

heres whats changed:

## **a clearer layout for janitor and janitor+**

settings are now split into actual sections instead of one long wall of controls.

provider selection and response length live under **General**. custom prompts, prefills, and forbidden words are grouped together under **Instructions**. makes way more sense than before.

for proxy users we left more of the advanced stuff (temperature, max tokens, etc) exposed right on the page, since thats where you guys actually tweak things. janitor and janitor+ users can still get to them quickly in **Generation controls**.

## **save and reuse multiple custom prompts**

this ones been requested forever. you can now **create and save multiple custom prompts** and switch between them whenever you want. no more keeping a notes app full of prompts and pasting them back and forth.

custom prompts also got their own dedicated page, so editing a long prompt doesnt take over the whole settings screen anymore. prefills and forbidden words got the same treatment - each has its own focused editing page.

## **simpler response length controls**

the old &quot;short responses&quot; toggle is gone. its replaced with a clearer **Response length** control where you pick between **Standard** and **Shorter** replies. same feature, way less confusing.

## **improved generation controls**

temperature and max tokens now live together under **Generation controls**, with simpler descriptions that actually explain what each setting does. theres also a button to restore the default values at any time, so if you break something experimenting you can just reset.

for most janitor and janitor+ users these controls stay out of the way until you need them. proxy pages keep the advanced controls exposed so experienced users arent digging through menus.

## **clearer saving**

the redesigned page uses an explicit **Save** button now. you can review your changes before applying them, the button only lights up when something actually changed, and we will warn you if you try to close the page with unsaved changes.

no more wondering if that thing you toggled actually saved or not.

## **a better janitor+ experience**

janitor+ controls are grouped together in their own dedicated section. subscribers can also see their **remaining enhanced swipes** directly from api settings now, instead of hunting for it somewhere else.</content:encoded><category>post</category></item><item><title>Shorter responses are here</title><link>https://janitorai.com/news/announcements/shorter-responses-are-here/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/shorter-responses-are-here/</guid><description>Today, we’re releasing our short response model to general access. </description><pubDate>Wed, 15 Jul 2026 00:00:00 GMT</pubDate><content:encoded>![jllm-short](/news/static/SCR-20260709-nmfp.png)

We’ve gotten a lot of feedback from users who love Janitor, but wish they could get shorter, more conversational, responses. Now, you can select shorter replies inside of the model settings tab by choosing Response Length -&gt; Short.

This feature was previously only available to Janitor+ subscribers, but has now been rolled out to all users.

JLLM-short uses the same base model as JLLM, but has been designed to provide shorter responses. Responses are on average 70% shorter than standard JLLM.</content:encoded><category>post</category></item><item><title>Creator Verification Is Back</title><link>https://janitorai.com/news/announcements/creator-verification-returns/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/creator-verification-returns/</guid><description>Creator Verification applications are officially open again on Janitor.</description><pubDate>Tue, 14 Jul 2026 00:00:00 GMT</pubDate><content:encoded>Creator Verification applications are officially open again on Janitor.

Starting today, you can visit **Settings → Verification** to see your current eligibility, track your progress toward the requirements, submit an application, and view your application status and history.

## **What the badge means**

The Creator Verification badge indicates that Janitor has confirmed an account as an established creator on the platform.

It does not mean that we endorse the creator’s content, confirm ownership of any character or intellectual property, or grant the creator special moderation treatment or platform privileges.

## **Minimum eligibility**

To apply, creators must:

- Have at least five public bots.
- Have either 10,000 followers or at least three public bots with 50,000 or more messages each.

The new Verification page shows these requirements alongside your current account progress, so you can see exactly where you stand.

Meeting the minimum requirements makes your account eligible for review, but it does not guarantee verification.

## **How applications are reviewed**

Applications are reviewed holistically. In addition to the measurable requirements, we may consider:

- **Good standing:** A consistent history of following Janitor’s Community Guidelines and Terms of Service.
- **Authentic growth:** Followers, chats, and engagement must be genuine. Botting, metric manipulation, fraud, or other suspicious activity may make an account ineligible.
- **Originality and quality:** A history of publishing original, thoughtfully developed bots that demonstrate meaningful creative effort.
- **Community engagement:** Active bot maintenance, user interaction, and a positive presence within the Janitor community.

Accounts that primarily repost, duplicate, or minimally modify existing public bots may not be eligible.

We may also verify notable creators at our discretion, including established creators from other platforms who do not meet the standard minimum requirements. This is not a separate application path and does not guarantee verification.

## **Application status and reapplying**

After submitting, you can return to the Verification page to see whether your application is under review, approved, or not approved. Your application history and relevant dates will also appear there.

If your application is not approved, you may apply again after three months, provided you still meet the minimum eligibility requirements.

## **Why was verification paused?**

We first introduced verification when Janitor was a much smaller platform. It was a way to highlight creators who were building incredible stories and experiences on the site.

At the time, verification was largely discretionary. When we came across a great creator, we could recognize them directly. As Janitor grew, however, it became increasingly difficult to keep that process fair. Many deserving creators were not verified simply because we never came across their profiles.

We later launched a form that allowed creators to request verification, but it quickly became backlogged with tens of thousands of submissions. Reviewing that volume while maintaining a thoughtful and consistent standard was not sustainable.

Our new process is designed to keep up with Janitor’s growth. It introduces clear, measurable eligibility requirements, a structured application process, and greater transparency around what we consider during review.

## **I’m already verified. Does anything change for me?**

No. Existing verified creators do not need to reapply under the new process. However, as always verification is not necessarily permanent and we reserve the right to remove a badge if an account no longer meets our standards.

We’re excited to make Creator Verification available again—and to give established creators a clearer, fairer, and more transparent way to apply.

Visit **Settings → Verification** to get started.</content:encoded><category>post</category></item><item><title>introducing janitor+</title><link>https://janitorai.com/news/announcements/29/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/29/</guid><description>Today we are launching our first-ever subscription product: janitor+. Available now.</description><pubDate>Wed, 24 Jun 2026 00:00:00 GMT</pubDate><content:encoded>![janitor+](/news/static/subcoverpic.webp)

## Introducing janitor+

today we are thrilled to be announcing our first-ever subscription product: janitor+.

janitor+ is now available for **$12.99/month** at [janitorai.com/plus](https://janitorai.com/plus), and you&apos;ll get:

- **5x more context** for improved memory during chats
- **Unlimited priority messages**, giving you faster responses even when usage is high
- **Generous monthly swipes with our frontier models**, allowing you to regenerate responses using our most powerful models and additional compute for noticeably better rerolls
- **A shiny janitor+ badge** to show others your support for the platform

too many subscriptions are about making things worse and imposing harsh limits, then forcing you to pay to get around them.

this is not that. the version of janitor that you know and love stays exactly the same, and stays completely free.

instead, we&apos;ve been working hard designing janitor+ so our users can get more out of janitor: more context, faster messages, and better rerolls.

this extra capacity requires lots more gpus, which are expensive right now, and we&apos;ve priced it to hopefully break even.

we are super excited about the experiences that the ability to spend more on gpus will let people create. we have a lot of fun plans in the works including better models, other modalities (think voice and image), and features that are just too expensive to roll out to everyone right now. janitor+ will let us do more of this.

## **app update required**

if you&apos;re on ios or android, you&apos;ll need to **update the app** to access janitor+. grab the latest version from the app store or play store.</content:encoded><category>post</category></item><item><title>janitor speaks your language now</title><link>https://janitorai.com/news/announcements/28/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/28/</guid><description>The janitor interface is now translated into a bunch of new languages. Head to settings → languages to switch.</description><pubDate>Mon, 22 Jun 2026 07:47:00 GMT</pubDate><content:encoded>![choose your language](/news/static/blogaboutllanguagesfrontend.webp)
hey everyone, shep here.

## **janitor now speaks your language**

the site is officially **translated into a bunch of new languages**. all the ones we added are right up there in the picture above. heres the full list:

- **Deutsch** (German)
- **English**
- **Español** (Spanish)
- **Français** (French)
- **日本語** (Japanese)
- **한국어** (Korean)
- **Latviešu** (Latvian)
- **Polski** (Polish)
- **Português** (Portuguese)
- **Русский** (Russian)
- **中文** (Chinese)

if you want to switch, head to **settings → languages**

## **whats next**

this part is just the interface for now. the next big step is **content**.

we are going to be **working with creators** to figure out the best way to also translate the actual stuff you chat with - characters, scenarios, etc.thats the harder problem and we want to get it right with the people who actually make the content.

the whole point here is making janitor **more accessible** to everyone, no matter what language you think in.</content:encoded><category>post</category></item><item><title>an update for users in brazil, australia, and the UK</title><link>https://janitorai.com/news/announcements/24/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/24/</guid><description>Due to new laws in Brazil, Australia, and the UK, we are being forced to implement age verification for users in those countries. Here is what is happening and why we are against KYC for social media.</description><pubDate>Mon, 15 Jun 2026 00:00:00 GMT</pubDate><content:encoded>![ukblock](/news/static/nsa.webp)
hey everyone.

## **here we go again**

if you remember the UK mess last year, you probably know where this is going. **users in brazil, australia, and the UK will need to verify their age** to access janitorai. this is **not a choice we made** it is now a legal requirement. we are fundamentally **against KYC for social media** and always have been.

this **only affects users connecting from IP addresses in brazil, australia, and the UK**. if you are outside those countries, nothing changes for you. we have absolutely no plans to roll this out anywhere else.

we do not have a specific date yet since it depends on how long implementation takes, but we wanted to get this out to you as early as possible. we will update you as soon as we have a confirmed date.

## **why is this happening**

as an australian myself, watching the government push these laws is incredibly frustrating. we all know this is really about power and control over the internet. both countries recently passed laws requiring platforms like janitor to verify user ages:

🇦🇺 **Australia** — the Age-Restricted Material Codes under the Online Safety Act 2021 came into effect on 9 March 2026, requiring age verification for adult content and AI services. ([Read more from ABC News](https://www.abc.net.au/news/2026-03-09/privacy-concerns-about-age-verification-r-rated-games-websites/106432440))

🇧🇷 **Brazil** — the Digital ECA (Lei No. 15.211/2025) took effect on 17 March 2026 with equivalent requirements. ([Read more on the new law](https://reclaimthenet.org/brazil-digital-eca-age-verification-law))

our options were: add age verification, block these regions entirely, or face **fines up to $35M in Australia or $9.5M in Brazil** plus potential blocking orders. so we decided to add verification rather than completely cutting you guys off.

## **how it works**

if you are in brazil or australia you will be asked to do a one-time age check. everything else - your characters, chats, account, settings - **stays exactly as it is**.

we are using **k-ID**, the same provider used by Discord, Twitch and Snapchat. we picked them specifically because of their privacy-first approach and overlap with platforms our users already use.

you will have three options:

- **Facial age estimation** — a short video selfie that runs entirely on your device. nothing is uploaded or stored.
- **ID scan** — photo of a government ID, deleted immediately after age is confirmed. k-ID never shares it with us.
- **AgeKey** — if you have already verified on Discord, Snapchat, or any other k-ID platform, you can just tap Face ID or your fingerprint and you are done in about 3 seconds. no new scans needed. if you do not have one yet, completing verification here will create one you can reuse everywhere.

## **your privacy**

k-ID does not store your documents or link anything to your janitor account, and **we never receive any of it**. the only thing we get is a yes/no on whether you are over 18. nothing else.

## **the bigger picture**

we know this sucks. we hate KYC for social media just as much as you do and we have done everything we can to find the least invasive path forward.

that said - **this system relies entirely on the IP address you connect with**. we do not track your physical location, ask for your address, or use gps. if people find other ways to manage their internet privacy, that is entirely on them.

if you have questions, our support team is at [help.janitorai.com](https://help.janitorai.com).</content:encoded><category>post</category></item><item><title>what scale actually taught me about infra</title><link>https://janitorai.com/news/announcements/27/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/27/</guid><description>infra notes from running things at scale. kubernetes, talos, bare metal, caching scars, queue beasts, ddos, and the lessons that actually stuck.</description><pubDate>Mon, 04 May 2026 07:00:00 GMT</pubDate><content:encoded>![talos](/news/static/BGtalos.webp)

this is more of an infra diary than an announcement.

i’ve been thinking about how to concepts like “running software at scale” outside of the context of how to design systems with pretty architectural diagrams.

prior to my kubernetes experience, i had no use case for it. now, i’m gradually learning the same principles that kubernetes abides by, with other components in my stack.

## rke2 was where i learned to breathe

my first production ready kubernetes setup was rke2, along with rancher.

i was truly in unchartered territory. i felt like i was constructing a space shuttle with various blog posts. i had no clue what a storage class was or what ingress was. the ship (or space shuttle) was rapidly sinking, and so was my mental health.

it took a lot of trial and error and clicking around rancher to understand what was going on, and how to ensure that my services were exposed to the outside world.

with rancher’s gui tooling, i learned how to be less ignorant, and stopped having to take wild stabs in the dark to figure out what was going on.

even though rke2 has its issues, it is an amazing learning resource, and opens up the world of production ready kubernetes.

## k3s made it lighter

later, i set up k3s.

again, i felt less lost. it had the same potential as rke2 to open up the world of kubernetes for me, but with less overhead.

i use ansible to configure machines. it helped automate provisioning, but it only defines the initial configuration. it doesn’t constrain the state of the system. over time, the configuration of the machine and the provisioning script differ. the script defines a desired state. the machine defines a real state. the difference between the two is revealed during an incident.

i’ve fired k3s and ansible from the job. they’re both great.

if you’re simply handling a low volume of requests, a system made up of a couple machines cleverly interconnected by ansible, may be good enough. however, if you find yourself handling a larger volume of requests and your system is still made up of ansible machines, you may start questioning your life choices.

the nature of handling a larger volume of requests often involves tuning parameters to tweak how the system manages concurrency. other concerns may include distributing load across a larger number of machines, keeping machines in a consistent state, and ensuring that multiple processes running on a single machine remain coordinated.

## talos clicked

eventually you’ll begin to develop an appreciation for systems that remain boring at scale. talos is one such operating system. its purpose is to serve kubernetes. it claims to be the first operating system of its kind. talos’ creators promised to remove the peripheral purpose from machines running kubernetes, and talos delivers. the absence of an interactive shell may feel constraining at first, but it may be reassuring to someone who has managed a cluster of kubernetes nodes at scale.

configuration in talos is declarative. node configurations are serializable. you can check them in, update them, and recreate them. if something is changed outside of the talos control plane, it is obvious where the change originated. if a node dies, it is replaced.

talos and kubernetes fit well together. kubernetes focuses on desired state. talos moves this to the lowest level of the system, the kernel.

this allows kubernetes to be as easy to deploy as other talos based workloads.

talos is used for our kubernetes control plane. talos is also used to build our node workloads.

## bare metal is the direction

a fleet of 50 extremely fast, and extremely powerful, amd epyc worker nodes, allows us to achieve low latency and high throughput for our workloads. this gives us better economics than the hyperscalers.

we have a 50gbps connection to the internet. egressing data on this connection costs us. talos allows us to provide a low latency, high throughput service at a lower cost than the hyperscalers.

a cloud provider is great for low latency, high throughput services. the cost of egressing data from the cloud quickly changes cloud computing from being a great service, to an uncompetitive, and frustrating experience.

## caching, carefully 

we have used cloudflare&apos;s edge cache and origin server to provide low latency and high throughput to our customers. there were always rules to never serve stale assets to our customers. we learned we cannot always trust a cdn. this was most evident when cloudflare served pages to users that included the sensitive state of other users.

cdns can provide great service for truly public assets. cdns should not be trusted to provide secure, user specific assets.

our cache is farther from the services, and aggressively caches private data. stay cautious when dealing with personal info. we assume there are cache invalidations for every write. there is no true ttl.

## the dragonfly part

we use bullmq heavily, and it processes the majority of our background jobs. it does message processing and a ton of other stuff, like user notifications, cleanup, andretries. it also processes moderation actions. additionally, it handles asynchronous tasks and spikes during deploys. bullmq is great, but without proper diligence, users will notice a build up of messages.

plain redis does not use multiple cores, and becomes a bottleneck. dragonfly is an upgrade from plain redis and completely removes the bottleneck.

dragonfly also is easier to manage than redis sentinel or cluster modes.

## load shedding instead of dying

additionally, we have learned you can&apos;t always block bad actors.

there are tons of obvious, easy to block abuse cases. however, there are even valid user cases where a bad actor is sending thousands of requests per second.

the focus shifts from “try to keep everything online” to “prioritize keeping users online, especially the important ones, and let some dribs and drabs of traffic die off.”

one approach to traffic management is to implement real load-shedding mechanisms at the edge. turnstile, along with cloudflare, can be used to implement rate-limiting. expire resources to limit the time spent waiting for requests by cheaper endpoints. protect more expensive endpoints using a throttling mechanism. use circuit breakers to fail a dependent service if one of its backends takes too long to respond. implement back-pressure mechanisms to allow priority traffic to pass. limit the length of requests, especially background requests.

failing a request is often better than requiring users to wait for a response. it is better to respond with a status code 429, indicating that the user has been rate limited, than for the server to take 30 seconds to respond with a 500 status code.

the goal is to keep the website usable.</content:encoded><category>post</category></item><item><title>quick quality of life update</title><link>https://janitorai.com/news/announcements/26/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/26/</guid><description>Usernames are no longer permanent. You can now update your @name from profile settings, with a 30-day cooldown between changes.</description><pubDate>Sat, 25 Apr 2026 07:00:00 GMT</pubDate><content:encoded>![username change](/news/static/usernamechange.webp)

quick quality-of-life update: **you can change your username now.**

## **usernames are not forever anymore**

if you picked a username forever ago and have been stuck with it since, that is finally fixed. go to your profile settings, tap the edit button next to your username, pick the new one, and save.

after you change it, your account goes on a **30-day cooldown** before you can change it again. the cooldown shows in settings so you know when your next change is available.

## **same username rules**

usernames still need to be available, and they can only use letters, numbers, and underscores.

if the name is taken, blocked, or you are still inside the cooldown window, the app will tell you what happened instead of silently failing.

this is live on web and mobile. if youre on the app and do not see it yet, make sure you are on the latest version by completing quiting the app and reopening it.</content:encoded><category>post</category></item><item><title>public chats are here</title><link>https://janitorai.com/news/announcements/25/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/25/</guid><description>You can now publish your chats for other users to read, react to, fork, and continue. Share your best roleplay moments, showcase your characters, or just show off that one conversation that had no business going that hard.</description><pubDate>Tue, 31 Mar 2026 07:00:00 GMT</pubDate><content:encoded>![public chats](/news/static/bgpchats.webp)
hey everyone, shep here.

## **two big things at once**

we are dropping two features together because they just make sense together: **public chats** and **multi-persona in chats**. both are coming as open beta.

## **multi-persona is here**

you can now **switch between different personas mid-conversation**. 

theres a persona switcher right next to the chat input. tap your avatar, pick a different persona, and your next message goes out as them. the character treats each persona as a completely separate person. if you have three personas, the character will treat them as three different people in the same conversation.

![persona dropdown](/news/static/personadropdown.png)

every message gets tagged with who sent it. the llm gets all the persona descriptions so it knows exactly who its talking to.

**and it gets better** — we are actively training multi-persona into our upcoming model....

**why this pairs with public chats:** when you publish a chat, you can share your persona info. readers see each persona labeled. when someone forks your chat, they get a persona switcher that includes your original personas alongside their own — so they can continue the story as your characters or switch to their own.

![public chat button](/news/static/publishchatbutton.png)

## **public chats**

ever had a chat so good you wanted to show it off? yeah. us too.

**you can now publish any of your chats for the world to see.** this is version one - we are going to iterate fast based on how you guys use it.

## **how it works**

open any chat, hit publish in the chat menu, give it a title, add some tags, done. your chat gets snapshotted and goes live. the url is generated from the title so its actually readable when you share it.

![publish modal](/news/static/publishchatmenu.png)

if you keep chatting and want to update the published version, theres a diff view that shows whats changed since the last snapshot. you can republish with the new messages or just update the title/description/tags without changing the snapshot. unpublish is one click.

## **what people can do with your chats**

when someone finds your published chat they can:

- **read the whole thing** with your theme intact
- **react to messages** — emoji reactions on individual messages
- **favorite it** — saves the chat, separate from reactions
- **leave comments** — threaded replies, likes, real-time via websocket
- **follow the publisher** from the comments panel
- **fork it** from any message into their own private chat

![public chat top part](/news/static/pchatstop.png)

forking is the cool part. pick any message as a branch point and you get a new private chat starting from there. forked chats link back to the original so theres an attribution chain.

and heres the thing — **forks are infinitely nestable**. you fork someones chat, continue it, publish yours, and someone else forks yours. theres no depth limit. every published chat can be branched from and every fork can be published and forked again. the goal is a tree of conversations that keeps growing in every direction. basically git for chats.

## **finding public chats**

every character page now has a published chats section. its a great way to see what a character is actually like in conversation before you start chatting.

![view char section](/news/static/viewcharsection.png)

**heads up** - proper discovery (dedicated browse pages, trending, etc) hasnt been built yet. right now you find them through character pages. better discovery is coming.

## **creator controls**

**creators have full control over published chats on their characters.** you never lose ownership of your characters page.

heres what you can do:

- **disable published chats entirely** - toggle &quot;Allow Published Chats&quot; off in your character settings. nobody can publish chats on your character. all existing ones get hidden.
- **hide individual chats** - hit the hide button on any chat you dont want on your characters page.
- **control comments** - set comments to open, followed only, or fully disabled. your character, your rules.

![hide from page](/news/static/hidefrompages.png)

## **notifications**

when you publish a chat, **your followers get notified**. more notification controls coming soon.

## **customization**

you can set custom colors, fonts, and background images on your published chat. that gets saved with the snapshot so readers see it styled the way you set it up.

theres also a palette toggle for viewers — you can switch between the publishers theme, your own chat theme, or just the default. three modes, cycle through them.

## **for creators**

people publish good chats with your characters, other users see them and go try the character. your characters get discovered through actual conversations instead of just the description card.

you can also publish your own chats as demos to show what your character can do.

## **moderation**

every chat gets run through moderation when you publish it. if it violates our content policy it wont go live.

## **this is an open beta**

the core feature works, but this is still early. we are actively working on:

- **better discovery** - dedicated browse pages
- **sorting and ranking** - surfacing good stuff
- **ui polish** - smoothing rough edges

iterating fast on this. let us know if something feels off.

## **mobile app update required**

if youre on ios or android, you need to update the app to get public chats. get the latest version from the app store or play store.

## **go try it**

go publish your best chats, read what others are creating, fork something weird and take it somewhere unexpected. cant wait to see what you do with it.</content:encoded><category>post</category></item><item><title>The Janitor App is Now in Public Beta</title><link>https://janitorai.com/news/announcements/23/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/23/</guid><description>the official janitorai app is now available on the app store and google play</description><pubDate>Sat, 07 Feb 2026 02:00:00 GMT</pubDate><content:encoded>**The official JanitorAI app is now available on the App Store and Google Play.**

after months of development and testing, the app is live for everyone. you can download it right now:

- [App Store (iOS)](https://apps.apple.com/app/janitor-ai/id6692609366)
- [Google Play (Android)](https://play.google.com/store/apps/details?id=com.janitor.ai)

or visit [janitorai.com/app](https://janitorai.com/app-download) and we&apos;ll redirect you to the right store.

## What to Expect

this is a public beta. the app works and you can use it for everything you do on the website, but its not perfect yet. you will run into bugs. some things might feel rough around the edges. thats expected and were actively working on it.

android in particular has some known issues were prioritizing. if youre on android and something feels off, its probably already on our list.

by default the app runs in restricted mode. if you want to enable nsfw content on mobile, you need to do it from the website — go to [janitorai.com/profile-settings](https://janitorai.com/profile-settings) and turn on &quot;Enable NSFW on Mobile&quot;. this cant be done from within the app itself due to app store rules.

## Whats the Roadmap

right now were focused on feature parity with the website. theres still stuff on web that isnt in the app yet and closing that gap is the priority. once we hit full parity, the focus shifts entirely to performance and speed. the app should eventually feel faster and smoother than the website.

## Updates Coming Fast

over the next couple weeks were pushing frequent updates to squash bugs, add missing features, and improve stability. things will get noticeably better with each release. if you hit a bug, check back in a few days and theres a good chance its been fixed.

make sure you have auto updates enabled in your app store settings so you always get the latest version without having to think about it.

## Golden Mop Badge

as a thank you to everyone who downloads during the public beta, youll receive a one-time golden mop badge on your profile. this is exclusive to early adopters and wont be available later.

## Custom CSS Profiles

if you have a custom css profile that uses images not from the media library, it might look broken right now. were tightening up how external images are handled and ill be going through and fixing these retroactively over the weekend. if your profile is affected, hang tight.

## Fake Apps Warning

now that the real app is out, please make sure youre downloading from the official links above. the publisher should be &quot;Janitor AI Inc&quot; on both stores. if you previously downloaded a fake app, delete it and change your password.

---

thanks for being patient while we got this ready. were excited to finally have this out and were going to keep making it better every week. if youre enjoying the app, leaving a review on the store would mean a lot to us.</content:encoded><category>post</category></item><item><title>Improving moderation transparency at Janitor</title><link>https://janitorai.com/news/announcements/22/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/22/</guid><description>Updates to our content policy</description><pubDate>Thu, 05 Feb 2026 03:58:00 GMT</pubDate><content:encoded>As we’ve been improving our internal moderation processes, we also want to bring more transparency to you all about what is and is not allowed on Janitor.

Today we’re launching an updated version of our content policy, available [here](https://janitorai.com/content).

This update does not reflect any underlying changes in what is allowed on the site. Instead, we’ve taken our existing policies and restructured them into seven clearer core sections that reflect the categories of content and behavior we prohibit.

It also provides more detail in certain sections, and has links to additional help desk guidance for sections we tend to receive more questions about or see ambiguity arise.

If you have questions about our content guidelines you are encouraged to explore the [help site](http://help.janitorai.com), where you can also submit a ticket if you require additional clarification.</content:encoded><category>post</category></item><item><title>Warning: Fake Apps on App Stores</title><link>https://janitorai.com/news/announcements/21/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/21/</guid><description>the official janitorai app is not yet available on app stores</description><pubDate>Wed, 04 Feb 2026 06:58:00 GMT</pubDate><content:encoded>**The official JanitorAI app is NOT yet available on the App Store or Google Play.**

The app is currently in beta testing and pending approval. we have not released anything to the public app stores yet.

## Do Not Download Fake Apps

There are fake apps using our name on the app stores right now. these are not us. do not download them. we cannot verify what these apps do with your data or account information.

## How to Know When Its Real

When the official app launches:

- we will post an announcement here with direct links
- the app will be published by &quot;Janitor AI Inc&quot;
- we will link to it from janitorai.com

if you dont see an official announcement from us with links, its not our app.

## If You Downloaded a Fake App

if you already downloaded something claiming to be janitorai from the app stores, we recommend:

- delete it immediately
- if you entered any login credentials, change your password at janitorai.com
- do not enter payment information into unofficial apps

---

we know everyone is excited for the mobile app. its coming very soon. well make sure you know exactly where to get it when its ready.</content:encoded><category>post</category></item><item><title>How Content Moderation Works on Mobile</title><link>https://janitorai.com/news/announcements/20/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/20/</guid><description>understanding why some content looks different on the app vs the website</description><pubDate>Wed, 04 Feb 2026 00:06:00 GMT</pubDate><content:encoded>With the mobile app now public, some of you have noticed that content appears differently on your phone compared to the website. this is intentional and heres why.

## App Store Requirements

Apple and Google have strict content policies. to stay in the app stores we apply stricter filtering on mobile by default. this affects what you see when browsing and chatting.

## Two Modes on Mobile

**Restricted Mode (default)**

When you first open the app, content is restricted by default. in this mode:

- Characters with higher content scores are blocked entirely
- Character definitions (personality, scenario, first message) are hidden on mobile
- Names and descriptions are filtered for explicit terms
- Only approved images appear in descriptions
- Search results exclude more content

this is the strictest mode and is required for app store compliance.

**Unrestricted Mode (opt-in)**

If you change your content settings in your web profile settings, restrictions are relaxed:

- content becomes in sync with web

some content is still filtered even with unrestricted settings. this includes anything that violates platform rules regardless of where you access it.

## Why Does the Website Show More

The website isnt bound by app store rules so it uses lighter filtering. desktop users see content that mobile users in restricted mode cannot access. if you want the full experience on mobile, change your content settings at janitorai.com/profile-settings (not in the app).

## Character Creators

If your character is marked as unavailable on mobile, it likely has a high content score. you can:

- we may add some icons to your character list eventually to show this transparently its basically just the automod marking it as allowed in restricted mode or not.

## Chat Moderation

we now check chat messages for simple keywords on mobile when content is restricted. if your message or recent chat history contains content that doesnt meet mobile guidelines, youll see a notice explaining how to change your content settings for full access.

this only affects mobile users in restricted mode. desktop and unrestricted mobile users are unaffected.

---

This system lets us offer both a safe default experience for new users and full access for those who opt in. we know its not perfect but its the tradeoff that keeps the app available in the stores.

were being extra cautious during launch to make sure we stay on the right side of app store policies. as we learn more about what works well be re-evaluating these filters and likely loosening things up over time.

## App Status

the app is still pending approval from the app stores but were expecting it to be released this week. well post an update as soon as its live.</content:encoded><category>post</category></item><item><title>Mobile App Public Launch Coming Soon</title><link>https://janitorai.com/news/announcements/19/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/19/</guid><description>app store release this weekend or early next week</description><pubDate>Fri, 30 Jan 2026 22:00:00 GMT</pubDate><content:encoded>**The App is Almost Here**

After some months of building and testing, were ready to go public. the app will hit the app stores either this weekend or early next week depending on how long approval takes.

&lt;div style=&quot;display: flex;&quot;&gt;
  &lt;img src=&quot;/news/static/mobile-app-home.png&quot; alt=&quot;home screen&quot; style=&quot;width: 25%; border-radius: 0;&quot; /&gt;
  &lt;img src=&quot;/news/static/mobile-app-search.png&quot; alt=&quot;search&quot; style=&quot;width: 25%; border-radius: 0;&quot; /&gt;
  &lt;img src=&quot;/news/static/mobile-app-trending.png&quot; alt=&quot;trending with recent chats&quot; style=&quot;width: 25%; border-radius: 0;&quot; /&gt;
  &lt;img src=&quot;/news/static/mobile-app-profile.png&quot; alt=&quot;profile&quot; style=&quot;width: 25%; border-radius: 0;&quot; /&gt;
&lt;/div&gt;

**early adopter badge**

everyone who downloads the app in the first week gets an exclusive early adopter badge for their profile. once its gone its gone.

**thank you beta testers**

to everyone who helped test over the past few weeks, thank you. your feedback shaped what this app became. beta tester badges will be distributed soon as a permanent thank you on your profiles.

**watch out for fakes**

there are fake apps using our name on the app stores. the only official app is published by &quot;Janitor AI Inc&quot; - if you see anything else its not us. well link directly to the real app once its approved.

We will post again when its live on the stores. see you there.</content:encoded><category>post</category></item><item><title>Scheduled Releases</title><link>https://janitorai.com/news/announcements/18/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/18/</guid><description>Set a release date for your characters with public countdowns or silent scheduling.</description><pubDate>Mon, 19 Jan 2026 08:00:00 GMT</pubDate><content:encoded>You can now schedule when your characters go live instead of publishing immediately.

## Two Modes

![schedule release modal](/news/static/SCR-20260120-rrwo.png)

**Public Countdown** – Character page is visible with a countdown timer. Users can see the name, description, and tags. Avatar and definition are hidden until release. Users can favorite and comment while waiting.

**Silent Schedule** – Character stays completely private until the scheduled time. No one knows it exists until release.

![countdown preview](/news/static/SCR-20260119-ncdh.webp)

## Notifications

**Public Countdown** sends two notifications to your followers:

1. When you set the countdown
2. When the countdown ends

**Silent Schedule** sends one notification when the character releases, same as a normal publish.

## Editing Schedules

**Public Countdown:**

- You can adjust the release time
- You cannot switch to silent mode
- Canceling publishes the character immediately

**Silent Schedule:**

- You can edit the time or switch to public countdown
- Canceling keeps the character private

![creator view](/news/static/SCR-20260120-rsgu.png)

## Moderation

Content moderation runs when you set a schedule so you know immediately if theres an issue. For silent schedules, moderation also runs if you edit the character before release.

## Custom CSS

The countdown on character cards is an overlay on top of the normal avatar structure. Your CSS targeting `.pp-cc-avatar` or other avatar classes still works. Target `.pp-cc-countdown-overlay` to style the countdown overlay specifically.

## Notes

- Once a character has been fully published, you cannot schedule it again
- Releases happen automatically at the scheduled time
- Public countdowns commit you to the release since users are already waiting

---

Goes live January 20th.</content:encoded><category>post</category></item><item><title>Introducing the Janitor Newsroom</title><link>https://janitorai.com/news/announcements/17/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/17/</guid><description>We&apos;ve launched a dedicated newsroom to consolidate all Janitor updates in one place. Find official announcements, weekly changelogs, and contact information for press and support inquiries.</description><pubDate>Sun, 18 Jan 2026 08:00:00 GMT</pubDate><content:encoded>Today we are pleased to announce the launch of the official Janitor Newsroom, a centralized destination for company announcements, product updates, and platform news.

## A Single Source of Truth

As Janitor continues to grow and evolve, we recognized the need for a dedicated space where our community, partners, and the press can access accurate and timely information. The Newsroom serves as the authoritative source for all official communications from Janitor.

## What You&apos;ll Find Here

The Newsroom is organized into two primary sections:

- **Announcements** - Major product releases, company milestones, policy updates, and strategic initiatives that shape the future of our platform.
- **Changelog** - A comprehensive record of new features, improvements, and bug fixes shipped across our web and mobile applications.

## Our Commitment to Transparency

We believe in keeping our community informed. All significant updates will be published here first, ensuring that users, creators, and stakeholders have direct access to the information that matters most. While our community spaces remain an important space for discussion and feedback, the Newsroom will serve as the definitive record of official announcements.

## Media and Press Inquiries

Members of the press and media organizations can now reach our team directly through the contact information provided on this page.

For general support questions or copyright-related matters, dedicated contact channels are also available.

---

We&apos;re committed to building Janitor in the open and keeping our community at the center of everything we do. The Newsroom is one more step toward that goal.

Thank you for being part of our journey.</content:encoded><category>post</category></item><item><title>analytics for creators</title><link>https://janitorai.com/news/announcements/16/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/16/</guid><description>A new analytics dashboard for character creators with detailed metrics on impressions, clicks, views, chat engagement, and retention. Data-driven insights to help creators understand what makes their bots successful.</description><pubDate>Tue, 13 Jan 2026 08:00:00 GMT</pubDate><content:encoded>hey everyone

happy 2026. moving updates back to the blog so theyre easier to find and dont get buried in threads.

## character analytics (beta)

just shipped a new analytics section for creators on character pages. youll see it on your character page if youre the creator

![analytics](/news/static/analytics2.png)

### metrics

- **impressions** - how many times ur bot appeared in feeds/search
- **clicks** - how many people clicked on ur bot from feeds
- **views** - character page visits
- **view time** - avg time spent on ur character page
- **viewers** - unique daily viewers
- **chats entered** - how many times chats with ur bot were entered (new and returning)
- **user messages** - total messages sent to ur bot
- **chat time** - avg chat session duration
- **chatters** - unique daily chatters

these are meant to answer a simple question: where are people dropping off and what actually makes a bot work

we started collecting most of this data on dec 24 so thats where ur charts will begin. more metric types coming. starting with things that explain why a bot converts, not just how big it is

![analytics metrics](/news/static/analytics1.png)

### recommended tab

all this data is also feeding a new recommended tab. collaborative filtering based, so instead of just showing whats popular itll learn what u like based on what similar users are into

big goal is to surface smaller creators and break the top heavy distribution. if u make good bots u should be able to get discovered even without an existing following. this wont be perfect out of the gate but its designed to improve as more people use it

### coming

- comparison tools to see how ur bots stack up against ur other bots
- creator analytics + dashboard with aggregate stats across all ur bots
- favorites/unfavorites - timeline of favorites
- message rerolls - how often users regenerate responses (helps identify quality issues)
- message edits - when users edit their own msgs or ur bots responses
- bot response time - how fast ur bot responds
- swipe patterns - how users navigate through response alternatives
- chat retention - session duration + message counts per chat
- first msg chosen - for multiple first msg bots
- scripts init details

### next

more stats are coming, but im keeping it simple: add the ones that answer real creator questions.

if you want a metric, tell me what you are trying to measure (discovery, conversion, retention, etc) and ill try build the stat that matches it.</content:encoded><category>post</category></item><item><title>Findings and experiences with scaling api infra</title><link>https://janitorai.com/news/announcements/15/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/15/</guid><description>How we migrated from Node.js to Elixir and Cloudflare Workers to handle billions of messages. A deep dive into database migrations, real-time infrastructure, and the lessons learned scaling to millions of users.</description><pubDate>Tue, 11 Nov 2025 08:00:00 GMT</pubDate><content:encoded>## how we scaled janitor from node to elixir and cloudflare workers

hey everyone,

so i wanted to talk about something ive been grinding on for the past few months that nobody really sees but is the reason janitor has not caught fire despite handling billions of messages.

### the problem with node

i remember one night around 2am we had 500k concurrent websocket connections and our nodejs cluster was using like 120gb of ram across 30 servers. for reference whatsapp handles 2 million connections per server with elixir. we were literally doing it wrong lol.

### enter elixir

so i made the call to rewrite everything in elixir and phoenix.

the migration was brutal ngl. i basically had to run two systems in parallel for like months. but it was worth it

what used to take 30 nodejs servers now runs on 3 elixir nodes. websocket connections? i can handle millions without breaking a sweat. each connection is its own isolated process so if one crashes the supervisor just restarts it without affecting anything else. no more &quot;oops the whole server died because someone sent a weird emoji&quot; lmao

```elixir
def handle_in(&quot;message:create&quot;, %{&quot;content&quot; =&gt; content}, socket) do
  broadcast!(socket, &quot;new_message&quot;, %{content: content})
  {:noreply, socket}
end
```

that is how simple real time messaging is in phoenix.

### but wait there is more - cloudflare workers

elixir solved my real-time problems but i still had all these stateless apis that didnt need the beam vm overhead. stuff like character search, fetching bot definitions, user profiles, basic crud stuff that i just want fast.

enter the cloudflare workers.

moved more endpoints to be workers on the edge running closer to users. honestly shouldve done this from the start

heres what a worker looks like i suppose:

```javascript
export default {
  async fetch(request, env) {
    const character = await env.KV.get(characterId);
    if (character) return new Response(character);

    const data = await env.DB.prepare(&quot;SELECT * FROM characters WHERE id = ?&quot;)
      .bind(characterId)
      .first();

    await env.KV.put(characterId, JSON.stringify(data));
    return Response.json(data);
  },
};
```

### the database migration hell

ok before i talk about the architecture let me tell you about the database situation. i had over 20 billion rows in my chats table. another 100b in my the table. all partitioned tables with hundreds of chunks

migrating this much data is insane. heres how i did it for now:

i wrote a go script using pgx to handle the parallel processing for exports. honestly go is goated for this type of stuff

```go
func migrateBatch(pool *pgxpool.Pool, startId, endId int64) {
    // export batch to csv
    // compress with gzip
    // import to new table
    // verify counts match
    // drop from old table
}

// run 50 parallel workers
for i := 0; i &lt; 20; i++ {
    go migrateBatch(pool, i*batchSize, (i+1)*batchSize)
}
```

took 3 days of weird uptime (probably saw a few queue pages if you know you know lol) but i got it done. some lessons:

- **do not use cdc on billion row tables** - the wal retention just doesnt work
- **batch everything** - 50m rows at a time worked for me
- **verify as you go** - i had checksums for every batch because paranoia
- **have a rollback plan** - i kept backups of everything (obviously)

### the architecture now

the migration took 4 months of basically no sleep. i had to:

- rewrite 100+ api endpoints
- migrate billions of database rows without downtime (mostly lol)
- partition tables while keeping them online
- keep both systems running in parallel with gradual rollout
- not break anything for millions of users

### was it worth it?

well kinda yea. our infrastructure costs dropped a bit. response times improved a bit. we can actually sleep at night knowing the site will not randomly die for now. keyword = for now, i still 100% expect us to encounter scaling challenges but it will be much less of a challenge to fix with how nice this infra is working now, but its so much funner making real progress on these issues and honestly just so glad to be working on something so fun!

and the real win? we can finally build the features we been promising. real time features with hundreds of people. analytics. instant notifications for everything. a/b testing. all the stuff that was impossible when we were fighting memory leaks at 3am...

### what is next

the lesson here? dont be afraid to throw everything out and start over if your architecture is holding you back. wasted a year trying to make nodejs scale. should have switched to elixir sooner. and definitely should have used cloudflare stack from day one

if you are curious about the technical details or want to know more about how i did the migration without ~much downtime let me know. always happy to share what i learned the hard way

thanks for reading and thanks for sticking with janitor through all the growing pains.

shep</content:encoded><category>post</category></item><item><title>october 10 2025</title><link>https://janitorai.com/news/announcements/14/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/14/</guid><description>new search page, thinking box is back, copy proxy configs, better chat customization, janitorllm v2 preview</description><pubDate>Fri, 10 Oct 2025 07:00:00 GMT</pubDate><content:encoded>## weekly update - october 10 2025

hello everyone,

weve been working on a ton of stuff this week

## new search page

we added search history so you can find stuff you looked for before, still iterating on this so let me know if you hate anything about it. plan is to add trending searches to the dropdown etc

## thinking box is back

if youre using a proxy that returns thinking tags, youll now see the models actual thinking process in a nice little box.

![thinking box](/news/static/thinking.png)

## copy proxy configs

you can now duplicate your proxy configurations which makes it way easier to test different models on the same provider or save different custom prompts. just hit the copy button and you will get a duplicate you can edit however you want.

![copy proxy configs](/news/static/copyproxyconfigs.png)

## better chat customization

new text settings let you change your font, adjust italics, change that purple color that some of you have been complaining about on bright backgrounds basically make the chat look exactly how you want it.

![chat customization](/news/static/customizationinchat.png)

you can find all this under &quot;text settings&quot; in the customize chat section

## character intros fixed

you can now read all the intro messages a character has in their definition instead of just seeing the first one, if configured to show.

![read intro messages](/news/static/readintromsgs.png)

## spooktober badge

were doing a special badge for spooktober (kinktober). keep an eye out over the next couple weeks....

## whats coming

### janitorllm v2

ok im trying not to hype this too much because it wont be perfect in the beginning...

- new caching strategy that should cache about 90% of character tokens - this is what was causing issues the last couple of days - it is off now - revising
- which means we can actually run this thing at a larger scale.

right now jllmv2 is similar size to deepseek and honestly too expensive to run for everyone. but if this caching thing works the way i think it will, we might be able to make it available to more people soon. possibly a beta test group first to make sure it doesnt explode. it also has a bunch of new context engineering that we will also make work with proxies ideally.

## behind the scenes stuff

### database migration

theres a massive database migration happening right now that youre not supposed to notice but if you do notice weird stuff thats probably why.

ive been really heads down on this its not easy to migrate 30TB of data. like seriously 30 terabytes, but im super happy its working the data is slowly moving xD.

### websockets

were also moving most of our apis to websockets which sounds boring but it means:

- **real time feeds** - see new content the instant its created
- **actual real time comment sections** - actual conversations not refresh and hope
- **some other stuff thats gonna be cryptic as hell to get working**
- **notifications for literally everything if you want them** - every like comment new character whatever

this is all infrastructure work that makes everything else possible. not sexy but necessary. its the foundation for a lot of cool things

btw love you all and thanks again for creating on janitor</content:encoded><category>post</category></item><item><title>September 27, 2025 - Weekly Changelog</title><link>https://janitorai.com/news/announcements/13/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/13/</guid><description>multiple initial messages, scripts detail pages, better persona management, unified css editor, character size limits increase</description><pubDate>Sat, 27 Sep 2025 07:00:00 GMT</pubDate><content:encoded>## Multiple initial messages

We&apos;ve heard you loud and clear\! Creators can now add alternative message starters to their characters. When starting a new chat, users can pick the vibe that matches their mood, persona, or RP style.

![messages](/news/static/multi-messages.png)

## Scripts beta - detail pages

Scripts now have their own spotlight on each Script&apos;s detail page. Creators can add lore, describe unique functionalities, and share insights about the universe behind them. Whether it&apos;s backstory, usage tips, or just something cool you want others to know, this is your space to tell the story of your creation.

## More Scripts features

You can now use regex keys to unlock even more flexible triggers, define entry-level depth for precise control, and enjoy a smoother experience with a series of small fixes and improvements rolling out daily. Every update makes scripts more powerful and reliable.

## Better persona management

You can now sort and filter all of your personas both in your personas tab as well as when you select your persona before a chat with a drop down at the top.

You can sort by:

- Oldest: The default order, which is the one you&apos;ve all grown used to.
- Newest: Shows your latest created personas at the top.
- Last Edited: Shows your personas with your most recently changed personas at the top
- A-Z and Z-A: Alphabetical sorts based on your persona&apos;s name

## Improved mobile scrolling

Scrolling through character cards is now easier. When you click through to the next page of results, it will bring you to the top of those results instead of leaving you at the bottom and forcing you to scroll up.

## Image policy work

We rolled out a new automated moderation system that fixed tons of false positives but (plot twist\!) introduced a few fresh bugs. We&apos;re patching them quickly. Keep reporting issues — every report helps us make this smoother, and we appreciate your patience ❤️.

## Unified CssEditor

We&apos;ve moved over to a single CSS editor, which is an improved version of what was previously the &quot;Pro&quot; version, with much nicer functionality like the element picker. The two editors were fighting for control over your CSS if you were using both of them, and it&apos;s much cleaner now.

Don&apos;t worry though, your code in the LiteEditor is not lost. You&apos;ll find your old CSS has been placed within the proper \&lt;style\&gt; \&lt;/style\&gt; tags in the ProCssEditor, denoted by a neat comment so you can find it easily.

Next time you open up the CSS editor, pay a little bit of attention. You can quickly remove the &quot;Restoration&quot; of your LiteCssCode by hitting the &quot;Reset Unsaved&quot; button. This won&apos;t work anymore after you next save your CSS, so it might be smart to take a look at what you want to keep/remove\!

## Character size limits increase (properly this time :P)

We introduced an upgrade to character size before, setting it to 1mb max total size, but didn&apos;t increase the individual fields maximum.

Now we&apos;ve increased the maximum characters per field by 10x, from 84,000 to 840,000 too, so you should have more than enough space to write as much as you want for your characters.

## Soundcloud player enhancements

We&apos;ve added a loop button on the right side of the soundcloud player, and a volume slider for desktop users. We also fixed a bug that was causing some imports from Soundcloud to fail.

## Images in bio

We&apos;ve disabled images from showing in the bio in the character card preview on the home and search screens. This will make the sizes of the bio previews way more consistent. Let us know what you think of this change.

## Search improvements

Briefly launched a redesign of the search page, rolled it back after feedback and a few missing features, please keep the feedback coming it&apos;s super appreciated.

Search queries are also 40\% faster now.</content:encoded><category>post</category></item><item><title>September 19, 2025 - Weekly Changelog</title><link>https://janitorai.com/news/announcements/12/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/12/</guid><description>scripts rebuilt to run independently, expand characters in chat, response prefill, better image moderation</description><pubDate>Fri, 19 Sep 2025 07:00:00 GMT</pubDate><content:encoded>## scripts

we rebuilt the whole system so each script runs independently now. so you can use way more complex functions without everything breaking.

basically if you know what youre doing, you can build some genuinely powerful stuff the limits are way higher.

lorebooks: fixed a bunch of bugs and secondary keys are working properly. these let you be way more selective about which entries actually make it into your prompts instead of dumping everything.

more ui changes: should have scripts pages out over the weekend giving you the ability to comment and add some different content on them

**and no more losing your work**. autosave is live. switch tabs, lose internet etc.

thank you to everyone in the beta testing channel in discord also, really excited to see what we can make scripts turn into

![details](/news/static/diag.webp)

## expand characters in-chat

you can finally see character descriptions and images while chatting. just click the character name to expand everything, click anywhere else to close it.

![bio](/news/static/bio.webp)

## response prefill

you can now decide how the models reply should start.

![prefill](/news/static/prefill.webp)

## better image moderation

image moderation should be way more accurate now with less random blocking, more consistency with actual content policies.

we are watching how it performs, if something gets blocked that should not, keep flagging it. helps us tune the system.

## scroll fixes

added a custom scroll restoration hook. sounds boring but fixes two annoying things:

1. when you click next page, it scrolls you to the top pagination so you see characters instead of feeling like the page just reset
2. fixed custom html scroll on profiles that wasnt working

### menu spacing bug

fixed the weird menu loading issue some browser updates caused. menus should look normal again.

---

more coming next week. also would love to do a BTS version of these to show some of the things i work on in the darkness that well im personally super excited about.</content:encoded><category>post</category></item><item><title>Janitor Weekly Update</title><link>https://janitorai.com/news/announcements/11/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/11/</guid><description>more llm generation controls, scripts improvements, proxy config reordering, bigger characters, and ads test expanding</description><pubDate>Fri, 12 Sep 2025 07:00:00 GMT</pubDate><content:encoded>## Sep 12, 2025 \- Weekly Changelog

Hey all, hoping to do an (almost) weekly update to keep everyone up to date with what we&apos;ve been shipping and what&apos;s coming next, starting with this one here!

### More control over LLM generation

We&apos;ve rolled out additional features to give you more control over LLM generation. Now, you can add forbidden words and set custom Top K and Top P values, the repetition penalty and the frequency penalty.

These can be accessed through the Advanced Settings and Generation Rules in the menu.

![Generation rules](/news/static/generation-rules.png)

### Scripts is getting better

Shipped a bunch of improvements to Scripts. Mobile editing should work better now, and we&apos;ve fixed a bug that was breaking Scripts on large characters.

We also improved the debugging tools for creators&apos; advanced Scripts, and added a Match Whole Word option.

We&apos;re working on some improvements to make Scripts design more intuitive, as well as features to allow you to view and share public scripts.

### Proxy config reordering

We&apos;ve heard from a lot of you that keeping track of your proxies can be annoying. That&apos;s why you can now reorder your proxy configurations\!

Inside the Proxy Configuration settings on the API Settings page, use the buttons to move proxies up or down. Desktop users can also drag the proxies to reorder them.

![Generation rules](/news/static/proxy-reordering.png)

### Even bigger characters

We&apos;ve bumped up the maximum size of your created characters from 100kb total to 1mb total. Let your creativity run wild, but do keep in mind most models prefer a modest token count \&lt;3

### Ads test is expanding

We talked a couple weeks ago about how we were [testing](https://blog.janitorai.com/posts/9/) an ad to help cover our GPU costs and let us make improvements to the model while avoiding some of the challenges with payment processors.

The numbers seem initially good, but it just isn&apos;t getting that many views right now. We also heard from a lot of you that you were willing to see a few ads if it helped offset our costs.

Over the weekend, we&apos;re going to roll out a few ads to all users. We&apos;re doing our absolute best to put them in unintrusive places, and would love your feedback as the test rolls out — please send us your thoughts, positive or negative.

### Other improvements

**Scroll restoration**  
When navigating back and forward through the website, it will now remember where you left off and put you right back to where you were\! Should make browsing the trending bot pages a lot nicer.

**Search bars in Following and Blocked pages**  
You can now search the creators you have blocked or followed.

**Enlarged images in chat**  
Enlarged images in the chat on most browsers will now be in their full quality, and scale neatly instead of staying at the previous lower resolution.

**CSS editor improvements**  
Bunch of fixes to bugs in the CSS editor.

- CSS editor will now save empty fields
- Fixed issue where CSS pro editor was unscrollable on mobile
- Fixed issue where select element drop-down was unclickable if &quot;Select Element&quot; was toggled on.
- Resized editor on mobile slightly, so it should fit the screen way nicer.
- Fixed issue where a step was getting skipped inside the CSS editor tutorial.
- Added a button to re-take the tutorial for the CSS editor, in case you skipped it or wanted to see the explanations again.

**Character mark stacking**  
Fixed issue where combining markers (Thai, Arabic, etc) were stacking and overflowing in the comments section. Now limited to two diacritics/combining markers.</content:encoded><category>post</category></item><item><title>Scripts beta is here</title><link>https://janitorai.com/news/announcements/10/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/10/</guid><description>scripts let you add actual logic to your characters instead of just praying the ai follows your prompt</description><pubDate>Fri, 05 Sep 2025 07:00:00 GMT</pubDate><content:encoded>alright so we just pushed scripts beta to everyone. you can check it out at [janitorai.com/scripts](https://janitorai.com/scripts) **heads up - its still early days** and were actively iterating on everything.

## what are scripts anyway

basically scripts let you add actual logic to your characters instead of just praying the ai follows your prompt. you can write code to control how conversations flow, when certain info gets revealed, how relationships develop - that kind of thing.

want a character that only drops their backstory after 50 messages? you can code that. want different responses based on keywords? thats possible too.

the advanced mode is straight code editing right now and the lorebook mode is more visual. were working on making both more intuitive as we go.

## two ways to create

you can find scripts by clicking your profile icon and hitting the scripts tab. there are two modes:

![scriptshoices](/news/static/choices.png)

- **lorebooks:** visual editor mode for world-building and lore management. you can also import your existing sillytavern json lorebooks
- **advanced:** code editor for more complex logic and full control

huge shoutout to [icehellonix](https://janitorai.com/profiles/6d45975c-9dfb-4256-9776-031dd20b3db8_profile-of-icehellionx) who made a [guide](https://drive.google.com/file/d/195-zjs-qr5OAQpV5S9jSJ7IJ7GzaowVx/view?usp=sharing) to help people get started. honestly without community members like them stepping up, this would be even more confusing.

## why documentation is sparse

**lots of stuff is going to change**. were still figuring out the best way to make this accessible to people who dont code. the ui needs major work. the api will probably change. thats why we havent written extensive docs yet - dont want to document something thats gonna be different next week.

were planning to add:

- way better visual editors so you dont need to code
- script sharing and discovery
- real world connections (time of day, weather, etc)
- probably a bunch of other stuff based on what you guys want

but right now? its basically a code playground for the brave.

## what to expect from beta

**this is a true beta** - were releasing it early so creators can start experimenting and help shape what this becomes. if youre comfortable diving into early features, jump in. if you prefer more polished tools, check back in a few weeks as we iterate.

## whats actually working

you can:

- create scripts that control conversation flow
- connect scripts to multiple characters
- stack multiple scripts on one character
- build some genuinely cool narrative systems if you know what youre doing

what were improving:

- ui
- better documentation once features stabilize
- clearer error messages
- visual tools for non-coders
- tons of other stuff based on feedback

## help us build this

if you make something cool, share it. found a bug or have ideas to make scripts better? we want your feedback plz join the conversation in our [#bug-reports Discord channel](https://discord.com/channels/563783473115168788/1413702412555915387).

check out the [scripts section on our help site](https://help.janitorai.com/en/category/scripts-1qzakwc/) for some examples to get started.

this is just the beginning. lets see what you can build with it.</content:encoded><category>post</category></item><item><title>testing an ad</title><link>https://janitorai.com/news/announcements/9/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/9/</guid><description>We&apos;re exploring advertising as a potential way to fund JLLM v2 and keep the model free. Running a small test with ads on character pages for logged-out users to understand viability.</description><pubDate>Tue, 02 Sep 2025 07:00:00 GMT</pubDate><content:encoded>![logo change](/news/static/gpu-dog.webp)

Hey all, Hugo here.

As you know, we&apos;ve been working on ways to pay for all the new GPUs that will go into running JLLM v2 (hopefully launching _very_ soon). One option we&apos;ve been exploring is ads, which would help keep the model free while covering our costs and avoids some of the payment processing challenges.

We want to understand if this is viable, so we&apos;re going to be running a small ads test to see if they might be able to provide enough revenue.

Starting later today, there will be a single ad shown on character pages for logged out users only. This is to get a benchmark on what advertisers might be willing to pay. The ads will look like this:

![Example ad](/news/static/ad-example.webp)

We&apos;re not about to dump ads all over the site, and if we do decide to move forward with some kind of ads (not sure yet!), want to make sure we&apos;re doing it in a careful and deliberate way that doesn&apos;t make the site worse to use. We will keep you updated before any further changes here.

Anyways -- if you see an ad on a character page while you&apos;re logged out, this is why. If you have thoughts (either for or against ads!) please send them to me, we would love to hear them.

Also if you want to support janitor open an incognito tab and spam click the ads I guess lol.</content:encoded><category>post</category></item><item><title>we made a logo icon!</title><link>https://janitorai.com/news/announcements/8/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/8/</guid><description>after two years with our shiba mascot we made a flat icon that works with any color scheme</description><pubDate>Fri, 29 Aug 2025 07:00:00 GMT</pubDate><content:encoded>![logo change](/news/static/logodog.webp)

so i finally made a logo. after two years of just vibing with our shiba mascot, figured it was time for a cute flat icon we could use for other things.

the idea is it works with any color not just pink. so we can match it with the color of a bunch of things.

![logo change](/news/static/colorful.webp)

hope you guys like it!

## some assets

&lt;img src=&quot;/news/static/black-transparent.png&quot; width=&quot;100&quot; alt=&quot;black spiral cat&quot; /&gt;
&lt;img src=&quot;/news/static/white-transparent.png&quot; width=&quot;100&quot; alt=&quot;white spiral cat&quot; /&gt;
&lt;img src=&quot;/news/static/logo-transparent.png&quot; width=&quot;100&quot; alt=&quot;logo move&quot; /&gt;
&lt;img src=&quot;/news/static/black-white-bg.png&quot; width=&quot;100&quot; alt=&quot;black spiral cat on white&quot; /&gt;</content:encoded><category>post</category></item><item><title>why janitor feels weird right now (and why thats changing)</title><link>https://janitorai.com/news/announcements/7/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/7/</guid><description>Acknowledging the current UX issues and outlining our plan to fix them. Personalized discovery, unified creator tools, and a focus on reducing clicks for everything. The mess is temporary as we rebuild.</description><pubDate>Fri, 22 Aug 2025 07:00:00 GMT</pubDate><content:encoded>![7](/news/static/7.webp)

hello eryone, i know some parts of janitor feel random as hell. discovery sucks. navigation is confusing. some features feel bolted on.

theres a reason i keep the chat page so simple it works. no features getting in your way. but even thats getting revamped soon with proper versioning so you can rewind bad turns, save perfect moments, and actually end your stories when they should end, not just ghost them when the ai fucks up.

heres the thing every feature should either help you create, help you find stuff, or get out of your way. thats it.

right now you waste clicks everywhere. scrolling through pages of bots you dont care about. five clicks to edit something. search that barely works. its disrespectful to your time.

**personalized discovery** - the algorithm should know what you want and show it immediately. no more endless scrolling through garbage.

**user choice everything** - dont like something? turn it off. want the old layout? keep it. minimal interface? yes.

**unified creator tools** - one dashboard for everything instead of this scattered mess we have now, the workshop.

**less clicks for everything** - bulk actions, quick edits, better search. basic stuff that should already exist.

rebuilding while millions are using the site is messy. some stuff will feel janky. new features might not integrate perfectly at first. but every change is asking, does this actually help users or is it just bloat?

the mess youre seeing? its not permanent. its us tearing down the old stuff to build something that actually respects your time.

stick with us. the best version of janitor is coming.

-shep</content:encoded><category>post</category></item><item><title>big updates, small team, huge ambitions</title><link>https://janitorai.com/news/announcements/6/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/6/</guid><description>The infrastructure overhaul is finally complete. With stability restored, we&apos;re shifting focus to JanitorLLM v2 with proper context sizes, new formatting options, and expanded creator controls. Plus we&apos;re growing the team.</description><pubDate>Wed, 06 Aug 2025 07:00:00 GMT</pubDate><content:encoded>its been a few weeks since our last real update and i wanted to share whats been cooking behind the scenes. were finally through the tough stuff and ready to ship some good features.

## the infrastructure marathon is over

remember when the site kept having issues? yeah that sucked for all of us. we just wrapped up a massive infrastructure overhaul that had to happen. were talking about keeping millions of daily users happy with just 3 devs. if we were running this on aws wed be looking at over $1M a month. instead we rebuilt everything ourselves.

were finally past that grind. the site is more stable which means we can focus on the fun stuff again.

![ambtions](/news/static/ambitions.webp)

## whats coming

**janitorllm v2 is happening.** once our new gpu cluster with nvidias b200s is up were fixing everything you hate about the current model. proper context sizes, multiple formatting options, and more control for the creators to change how the llm talks to you for a specific bot.

**media library is finally ready to get out soon** we fixed our image moderation pipeline with an advanced queue system and zero shot vlm. it will be strict in the beginning whilst we tune it further.

**longform content that actually persists in your chats using rag.** entire worlds of lore that your characters can reference and build on.

**scripts.** forget prompt engineering this is actual programming for stories. set conditions, triggers, events. make a character go psycho after detecting 10 positive messages. have different personalities for morning and night. pull in real weather data so they know its raining. force plot twists at exact moments. branching narratives that actually branch based on how people talk. the tech behind this is a nightmare but once its working, im excited to see what you will build!

**creator analytics.** we know you want to see how your bots are performing. we are working on getting the database architecture right so you can see real stats detailed chat counts, peak times, user retention. creators deserve transparency about their work.

**our new support experience is live!** huge shoutout to iroh for this one. we have got a [help center](https://help.janitorai.com/en/) with community guides plus an ai assistant that can give you support and info on rules. no more shouting into the void when something breaks. the mod team is making sure human response times in there are fast also, ily mods.

**status page launched!** now youll always know if its just you or if were having issues. check out the [realtime status page](https://status.janitorai.com/) for updates.

## growing the team

more hands means faster feature development. we are in that sweet spot where the foundation is solid and we can actually scale the team properly. if you have been waiting for those features we have been teasing, this is how we make them happen. (also if youre a super talented dev in san francisco, dm me!)

## premium membership: making everything better

with premium supporters we will acheive sustainability, we can upgrade the free experience while giving supporters access to some incredible new features. everyone wins. we wanted to get this done sooner but refused to compromise on what makes janitor special. should have more updates on the progress soon.

## moderation

yes i read most reddit posts 😔👉👈 i appreciate the feedback even when its harsh. i hear you loud and clear, you love janitor because of the freedom. thats never changing.

look a lot of the negative feedback on moderation isnt really deserved. theres a whole dark side you guys dont see -- dealing with the fbi because someones stalking their ex, people making bots of children or coworkers without permission, coordinated harassment campaigns etc. we are doing our best to keep this stuff off janitor, its hard but we will get it right.

to be more consistent and fair, we have been trying to use more automated tools. what the mods do is really hard with a site as big as janitor, and we hope this will be a better experience for everyone. also, increasing the use of warnings before mod action, clearer guidelines, and a real appeals process if ppl feel something has been unfair. more info in the helpdesk where you can also submit appeals, help.janitorai.com

heres the reality with payments: to keep the lights on and process payments, we have to prove we have good realtime moderation for illegal stuff banks are concerned about.

the comment spam situation? new systems are rolling out to kill those telegram bots and onlyfans spam without affecting real conversations.

## the bottom line

we survived the hard part. the site is stable. the team is growing. the features you have been waiting for are finally coming.

thanks for sticking with us through the growing pains. building something this big with a team this small has been wild but you guys make it worth it.

keep creating.

-shep</content:encoded><category>post</category></item><item><title>Launching help.janitorai.com</title><link>https://janitorai.com/news/announcements/5/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/5/</guid><description>Introducing our centralized help desk with over 75 articles, an AI chatbot assistant, and direct support options. Everything you need to know about content guidelines, JLLM, proxies, and more in one place.</description><pubDate>Fri, 01 Aug 2025 07:00:00 GMT</pubDate><content:encoded>![helpdesk](/news/static/helpdesk.webp)
Hey...hey you...yeah, you over there.... Iroh here... do you want some....expanded content guidelines? Some...JLLM and proxy guides? 😈

Well I got the right hook up. It&apos;s called a help desk. The janitorai helpdesk to be more specific. Or a customer portal. Or...uh...a Knowledge Base! Or the help part of the website, with the fun little chatbot thing!

Yeah, I said chatbot. Answers questions, and can even answer in an uwu way if you ask it real nice. Does it get everything right 100% of the time? Well...not quite, but it&apos;s learning, and it&apos;s getting smarter. So it&apos;ll help you out 9 times out of 10.

What about when it doesn&apos;t? Well...good news on that front. This place is packing over 75 articles to read through, and it still ain&apos;t finished. And if you really can&apos;t find what you&apos;re looking for between ol chatbert there, and a library well...I got good news. They even got people to answer questions if you really want that, and you can still email them too.

So you in? Wanna check it out?

## Introducing Jai&apos;s centralized help desk / knowledge base / customer portal

We haven&apos;t settled on a name... but you can find it at **[https://help.janitorai.com/](https://help.janitorai.com/)**

Here you&apos;ll find our growing library of articles that are all written with one thing in mind - to help you have the best experience possible on janitorai. We&apos;re already so fortunate to have an incredible number of fantastic creators, and over 1.4million public bots, with more being added every day!

## Our lovely little chatbot

What a handsome little thing! We&apos;ve been training this little customer service chatbot. It can help you with policy questions, CSS advice, character building, explain LLMs, proxies, troubleshooting - honestly, it&apos;s a pretty great feature to have! We&apos;ve poured hours and hours of testing and training into it. And it&apos;s actually very accurate!

Most...of the time. We&apos;ve noticed a couple of things that we haven&apos;t managed to iron out yet. So here are some helpful bits of advice:

- **Sometimes the little guy gets too enthusiastic and loses its train of thought.** The message may cut off. If that happens, just tell it &apos;Hey, you got cut off&apos;, and it should send you the full message!
- **It&apos;s pretty great at policy.** BUT, it&apos;s not so good if you ask it about very specific characters. If you want to ask about aging a character up, it performs much better if you ask about aging up criteria, or if you want to create a NSFW Pokémon bot, ask it what the rules are around that are, instead of asking about that one specific Pokémon you&apos;re thinking of.
- **Some topics and themes are very nuanced**, and it does do a pretty good job. However, other things it can definitely struggle with, so if you&apos;re a creator with a question around heavy consent themes, or you just feel like it isn&apos;t giving you a consistent answer, best to speak to one of our human agents. (Definitely human. We are meat potatoes like yourself.)

We also have a great search option - if you open the chatbot and click on search, it&apos;ll bring up an AI assisted search engine that goes through all of our articles and can answer you, while also showing you bigger sections of relevant articles, which it also links to.

## But! We don&apos;t just have the bot!

We have a growing library of articles, including yes - **expanded content guides!**

We&apos;re starting out with over 75 articles, and that&apos;s going to keep on growing. We intend for this to become a veritable Great Library of Alexandria! So if you want to read for yourself, please enjoy the guides and troubleshooting articles we have gathered here. Our incredible community is also helping out, we&apos;re always so grateful to you!

**Important!** If you spot an error, or believe something key is missing, let us know! You can email, chat to us, or let us know in our Reddit or Discord. We&apos;re determined to keep improving this.

So there we go, enhance your knowledge, and go...go to the bots with greater wisdom and confidence than ever before! They yearn for your presence!

-Iroh</content:encoded><category>post</category></item><item><title>important update, deployment issues explained</title><link>https://janitorai.com/news/announcements/4/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/4/</guid><description>An explanation of recent deployment issues that temporarily affected NSFW content, our commitment to the community, and upcoming payment processing changes to ensure long-term sustainability.</description><pubDate>Thu, 24 Jul 2025 07:00:00 GMT</pubDate><content:encoded>![doggy](/news/static/dogpic.webp)
i wanted to provide a little more context around what happened, and why we were changing the system prompt. gotta beat those reddit allegations, trust

## what happened

on wednesday night pacific time, we tested a system prompt to ensure that some content being produced by jllm is compliant with our terms of service. this ended up blocking a bunch of random nsfw stuff that we did not want to block.

the changes were targeted exclusively at three kinds of illegal content: underage characters, bestiality (excluding furry content), and incest content (excluding stepcest). these kinds of content have always been against our content guidelines and moderation policies, and we dont want that kind of content on the site.

we didnt announce these changes because we thought it wouldnt have any real impact on the content, and it was designed to be very specifically targeted at the above illegal stuff.

we rolled out some additional csam precautions earlier (around 3 months ago) and it did not cause any problems. we also made some changes to stop people from uploading lots of images of random real people, which had spiked as a problem recently. where groomers were making bots of people they know irl. we thought (mistakenly) that this would be equally smooth.

as you guys noticed it was not.

once we saw a bunch of stuff was getting blocked, we pushed a rollback almost immediately, and had thought that it was fixed.

however, it turned out that our build server was having errors and had crashed, so the old version didnt go back to live. we then frantically spent the next couple hours trying to get the build server back online, and once we did, were able to undo the moderation change.

this left you guys in the dark for a couple hours with huge moderation changes that we did not intend to do. im sorry about that.

also, some of you mlp creators got accidentally banned by our mod system. we reverted that too - sorry about that! were working on making the auto mod better so if you make a bot that breaks tos, you wont get banned. the bot will just be hidden from public and will tell you why. that way you can fix it without getting banned.

## why we made these changes

as you guys know, running janitor is really expensive. we want to launch a membership to help offset some of the server costs and allow us to hire more people and speed up a bunch of the long-overdue features were super excited about. its really important to me that janitor is able to be here for the long haul.

to be able to do this, we need a payment processor. this means we have to pass a banks underwriting process. banks, as you may be shocked to learn, are conservative. both ai and nsfw content make them nervous.

weve been working hard to find a bank that will allow us to keep all of the nsfw content on our site. as weve made clear from the beginning, were not willing to go the way of other sites and sacrifice the community to try to make some random bank happy.

weve found a few that are happy to let us have our content. however, they are very strict about content that is illegal in the us – in specific, csam, bestiality (that isnt furry content), and incest (not stepcest). this stuff has always been against our content policy, and we dont want it on our site.

as part of their underwriting process, the banks have tried to jailbreak the model to produce this illegal content. we already have some moderation on this, but they were able to find ways around it, so we pushed a change to the system prompt designed to fix this.

we screwed up! it broke lots of things. thats why we tried to roll it back immediately. once the buildkite server went down, we found ourselves unable to make any changes until we got that back online. as soon as we got that back up, we were able to undo the changes.

## whats next

its really important to me that we get this right so that janitor can continue to exist, and that we can keep paying our (scarily large) server bill.

once we get payments sorted, we can buy more gpus 🙂, launch new features faster, and continue to make janitor the creative playground we all want it to be.

we have a bunch of really cool features we want to launch, and i hate working on this stuff -- it distracts from the fun stuff that is actually important. but it is something we have to do if we want to continue to exist long term.

were going to work on finding a better way to stop the llm from generating this specific illegal content without breaking anything else, and will update you guys as it rolls out. this might be an improved system prompt, or some other kind of control. were figuring this out with you guys.

i really appreciate everyones patience here, and i know it can be really frustrating. its crazy to see how big the site has gotten, and sometimes i still just push stuff into prod like i was the only user -- but when it breaks, its not just breaking for me.

i know youve been hearing this a lot, but we really are grateful for u all ❤️. you guys are what makes janitor what it is.</content:encoded><category>post</category></item><item><title>tough news for our uk users</title><link>https://janitorai.com/news/announcements/3/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/3/</guid><description>Due to the UK&apos;s Online Safety Act requiring regulatory frameworks designed for tech giants, we&apos;re forced to block UK access starting July 24th. An explanation of what this means and why we had no other choice.</description><pubDate>Sun, 20 Jul 2025 07:00:00 GMT</pubDate><content:encoded>![ukblock](/news/static/ukblock.webp)
hey everyone, shep here.

## **this one hurts to write**

so we have to **block uk access to janitorai starting july 24th** (England, Scotland, Wales, and Northern Ireland). yeah, **just 4 days notice**. i know that is terrible timing and im genuinely sorry about that.

## **why we are in this mess**

we have been trying to wrap our heads around the **UK&apos;s online safety act**. and honestly? **i completely botched understanding what this thing actually requires**.

when it first passed, i thought &quot;okay, some new content moderation rules, we can handle that.&quot; turns out **i was dead wrong**. this is not just content moderation - **it is a complete regulatory framework that assumes every platform is a tech giant**.

**the requirements are insane:**

- **legal risk assessments** that cost more than our entire monthly budget
- **biometric id verification systems** (reddit is implementing persona for this which costs **$1.5 per person**)
- **constant legal review** of thousands of pages of guidelines that keep changing

**the penalties** include **fines up to £18 million** plus **personal criminal liability for me and our team**. like, **actual jail time** if ofcom decides we screwed up.

## **the timing disaster**

we have been **drowning in legal documents** trying to figure this out. the main law is **250+ pages**, the guidance from ofcom is **over 3,000 pages**, and its all written in the most confusing legal language possible.

i kept thinking we would find some reasonable compliance path, or that there would be exemptions for smaller platforms. **nope**. if you have uk users and any form of user interaction, **you get the full regulatory treatment**.

by the time we realized **compliance was actually impossible** for a team our size, we were already down to days before the deadline.

## **this is just broken policy**

the uk government talks about wanting to be a tech innovation hub while simultaneously **making it impossible for any platform smaller than google to serve uk users**. it is completely backwards.

they have created a system where **only the biggest tech companies can afford compliance**, which is probably the opposite of what they wanted to achieve. right????

## **what this means for you**

**starting july 24th at 11:59pm utc**, if you are accessing janitorai from the uk (England, Scotland, Wales, and Northern Ireland), **you will hit a block page**.

this really sucks. our uk community has been amazing since day one and **losing you guys feels awful**. but we literally **cannot risk our team facing criminal charges** over regulatory compliance.

## **the bigger picture**

**we are not unique here**. lots of innovative platforms are going to make the same choice because the alternative is **bankruptcy or prison**.

the uk is basically **creating a digital iron curtain around itself**, cutting off citizens from global innovation. it is wild to watch in real time.

---

if you are as frustrated about this as we are, there is a petition asking parliament to reconsider this mess: **https://petition.parliament.uk/petitions/722903**

sorry for the short notice everyone. this whole situation has been a nightmare to navigate.

and in the future once we have subs out i see a path forward with meeting their requirements some day. **we are actively exploring compliance options** including implementing **id checks for minors** and other verification systems. this block is temporary while we figure out the right approach - **we are not giving up on the uk market**.

---

## **update: clearing up some misinformation**

**quick clarification** because i have seen some wild takes floating around discord: **uk users cannot be jailed for using janitorai**. that is complete nonsense.

**the online safety act only applies to service providers** (that is us), **not users**. the criminal liability and fines are **only for the people running the platform**, not anyone using it.

also, **we are only blocking uk ip addresses**. your janitor accounts **will not be deleted** if you are a uk resident. the site will simply be **inaccessible from uk ip addresses**. if people find other methods to access the site, **that is entirely on them** - there are no legal consequences for users.

just wanted to make that crystal clear so people stop panicking about getting arrested for chatting with bots.

**-shep**</content:encoded><category>post</category></item><item><title>behind the scenes - making stuff work</title><link>https://janitorai.com/news/announcements/2/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/2/</guid><description>A monthly check-in on the infrastructure work keeping Janitor running. Server migrations, capacity scaling, and the invisible technical improvements that support our growing user base.</description><pubDate>Sun, 27 Apr 2025 07:00:00 GMT</pubDate><content:encoded>![engine room](/news/static/engine-room.png)
hey everyone shep here again.

its been a month since my last update and i wanted to check in even though we havent dropped any shiny new features yet. this month has been all about infrastructure - boring but critical stuff that keeps janitorai running.

## the invisible work that matters

our server have been hitting capacity limits with our growth so we have been migrating our infrastructure while keeping everything running. which has been super challenging but will be done soon (its a really fun problem to have).

![migration stress](/news/static/migration.png)

## tech improvements under the hood

we completely revamped our gpu infrastructure for better cache sharing between nodes. in simple terms, our servers can now communicate more efficiently with each other.

what we are building is a kvcache centered system inspired by projects like mooncake and lmcache. this means we can efficiently share the heavy computation across different nodes in our gpu cluster.

so basically our system now handles key value caches in a distributed way, allowing us to decouple prefill (llm inputs) and decode (llm outputs) phases. this all reduces bandwidth bottlenecks and lets us efficiently cache computation for individual users. basically saving the top parts of each users prompts for quicker processing.

the results are already showing:

- we can handle more concurrent users
- significantly improved stability
- a better substrate to build new features off of
- reduced costs while scaling up

huge shoutout to leon for this one!

![speed improvement](/news/static/speed.png)

## why does the site go down? a few reasons:

- most recently its been due to our infrastructure migration for the new llm system
- database growing pains - weve hit scale faster than expected and our db setup needs upgrades
- sometimes i break something on accident during updates (our bad, part of being a dev lol)
- occasional traffic spikes that temporarily overwhelm the servers

the good news is most of these are temporary and the migration work will make the site much more stable in the long run.

## community growth

even with zero marketing we keep growing. your word of mouth is everything.

we also hit a new record last week - 1.93M people using janitorai in a single day. sweating watching those numbers.

## whats coming next

the lore system is still in development and progressing well. im excited about how its shaping up but dont want to overpromise on timeline or specific features yet.

we are taking our time to make sure it works properly at scale before releasing it to everyone. quality matters more than rushing features out.

## community highlights

these creators have been building some of the most impressive profiles on the platform:

- &lt;a href=&quot;https://janitorai.com/profiles/715f6db3-9ea6-4ef6-8caa-1521f85a7d00_profile-of-cyanoide&quot; target=&quot;_blank&quot;&gt;cyanoide&lt;/a&gt;
- &lt;a href=&quot;https://janitorai.com/profiles/fd3d71ec-4033-4153-a143-c253273fad67_profile-of-nonpractical&quot; target=&quot;_blank&quot;&gt;nonpractical&lt;/a&gt;
- &lt;a href=&quot;https://janitorai.com/profiles/1bd0f6d5-3461-44a0-89fc-1245bb17edb0_profile-of-whitestarliner&quot; target=&quot;_blank&quot;&gt;whitestarliner&lt;/a&gt;
- &lt;a href=&quot;https://janitorai.com/profiles/fa479f59-71a0-4296-a24d-6e8ca1c3d6fa_profile-of-veseii&quot; target=&quot;_blank&quot;&gt;veseii&lt;/a&gt;
- &lt;a href=&quot;https://janitorai.com/profiles/ab055c6d-d58d-49ae-acc8-20099a6ec3d1_profile-of-the-radio-host&quot; target=&quot;_blank&quot;&gt;theradiohost&lt;a&gt;
- &lt;a href=&quot;https://janitorai.com/profiles/00affc51-fe74-4ed9-8ca3-5972411eeb7e_profile-of-tigerdropped&quot; target=&quot;_blank&quot;&gt;tigerdropped&lt;/a&gt;
- &lt;a href=&quot;https://janitorai.com/profiles/cab6a01e-f382-4f88-adc5-0a3e8c5cfeac_profile-of-oishiidesu&quot; target=&quot;_blank&quot;&gt;oishiidesu&lt;/a&gt;
- &lt;a href=&quot;https://janitorai.com/profiles/8ddccceb-c0b1-4287-a087-b73337a695c1_profile-of-pinkcat&quot; target=&quot;_blank&quot;&gt;pinkcat&lt;/a&gt;
- &lt;a href=&quot;https://janitorai.com/profiles/feb64c94-761c-4cd5-9807-f149d5005c76_profile-of-cesarmiku&quot; target=&quot;_blank&quot;&gt;cesarmiku&lt;/a&gt;
- &lt;a href=&quot;https://janitorai.com/profiles/4141a3d6-3ffb-4f79-b299-9e2a367eeae0_profile-of-oopsi-daisy&quot; target=&quot;_blank&quot;&gt;oopsidaisy&lt;/a&gt;
- &lt;a href=&quot;https://janitorai.com/profiles/a649fa0b-a8c2-4db8-a216-22bdd60d4720_profile-of-saturnlore&quot; target=&quot;_blank&quot;&gt;saturnlore&lt;/a&gt;
- &lt;a href=&quot;https://janitorai.com/profiles/cf262664-bc98-4209-a1cf-c5b67ba0019b_profile-of-cloudtoned&quot; target=&quot;_blank&quot;&gt;cloudtoned&lt;/a&gt;
- &lt;a href=&quot;https://janitorai.com/profiles/ae3b8516-54d5-4469-8557-6dcf808128d0_profile-of-iorveths&quot; target=&quot;_blank&quot;&gt;iorveths&lt;/a&gt;
- &lt;a href=&quot;https://janitorai.com/profiles/b971ee9a-f66d-4336-95cc-3ee1b2cba153_profile-of-reinasplenda&quot; target=&quot;_blank&quot;&gt;reinasplenda&lt;/a&gt;
- &lt;a href=&quot;https://janitorai.com/profiles/e7183b88-4921-4388-a598-69889ee969b1_profile-of-goblin-rat&quot; target=&quot;_blank&quot;&gt;goblinrat&lt;/a&gt;
- &lt;a href=&quot;https://janitorai.com/profiles/b2f9b679-1abf-4c67-a084-8f643c1cea9d_profile-of-waterpuppy&quot; target=&quot;_blank&quot;&gt;waterpuppy&lt;/a&gt;
- &lt;a href=&quot;https://janitorai.com/profiles/2ab9bb38-1b18-4299-97d4-e68e3bf2ccf2_profile-of-procrastination&quot; target=&quot;_blank&quot;&gt;procrastination&lt;/a&gt;

speaking of custom css profiles, check out &lt;a href=&quot;https://janitorai.com/profiles/88d2b914-7cb0-4a23-bd62-4bfd9ce3fef7_profile-of-tigerdroppeds-css-compendium&quot; target=&quot;_blank&quot;&gt;tigerdroppeds css compendium&lt;/a&gt; for learning css and this &lt;a href=&quot;https://janitorai.com/profiles/6f4a7226-6402-4a4a-b2b6-113616a0ad14_profile-of-polare&quot; target=&quot;_blank&quot;&gt;template&lt;/a&gt; for getting started, and check out this more detailed &lt;a href=&quot;https://cssforjai.ju.mp/&quot; target=&quot;_blank&quot;&gt;css guide&lt;/a&gt; and this &lt;a href=&quot;https://discord.com/channels/563783473115168788/1263470898078285895&quot; target=&quot;_blank&quot;&gt;channel in our discord&lt;/a&gt;
![profile](/news/static/profileshowoff.webp)
stay cool, keep creating.</content:encoded><category>post</category></item><item><title>janitorai Update: The Future Looks Awesome</title><link>https://janitorai.com/news/announcements/introduction/</link><guid isPermaLink="true">https://janitorai.com/news/announcements/introduction/</guid><description>1.4m daily active users, moderation updates, subscriptions coming, and a preview of our new knowledge system for characters</description><pubDate>Thu, 27 Mar 2025 07:00:00 GMT</pubDate><content:encoded>![cover](/news/static/coverpost1.png)
hey everyone shep here.

holy shit we have come a long way. **janitorai** (i think lowercase janitorai looks better tbh) started as an idea for limitless AI and now its grown into something way bigger - a creative playground where **YOU** decide what is possible.

our user base has exploded lately and its been a wild ride. thx to everyone whos been with us since day one and welcome to all the new faces. you are why we do this.

## 1.4M+ Daily Active Users in less than 2 years!

![dauchart](/news/static/dau.png)

## **We are Expanding and We are Gonna Win**

we are becoming **THE** place where creators can build stuff thats truly fun. And thats what matters most to me.

our infrastructure is now handling billions of interactions a week. weve completely overhauled our backend but software being soft always leads to hiccups but we recover quicker now.

the community janitorai has built is incredible - you guys keep pushing whats possible and we are just trying to keep up with your creativity. every character, every interaction shapes what we are building next.

## **Moderation That Makes Sense**

Moderation is hard af and we are learning as we go. Our goal is simple: make a safe place for creative freedom while protecting our awesome creators from harassment and keeping abusers off the platform. The support with the image change really is helping keep janitorai around for the long run.

i personally use the site for the same stuff you do. **janitorai** literally exists because of other sites strict filters - we are not compromising on that core identity.

building the platform we want to use ourselves.

we have a huge team of mods now and without them janitorai wouldnt be anything so please show them luv.

## **Subscriptions Coming When We are Ready**

Lets talk cash - weve got plans for monetization that respect why youre here. Think exclusive profile flair, beta models, and other upgrades that feel worth it but never required. Bottom line: janitorai needs to survive long term, but the features you rely on will stay free (which means you can still chat to your pookies without paying a dime, don&apos;t worry).

## **Cool New Stuff in the Pipeline**

we&apos;ve got a ton of undeployed code for a bunch of rly cool new features. our goal is **depth** making tools so powerful and flexible that even we cant predict what you will create with them.

## **Buckets/Books/Letters(havent decided on a name yet): A Whole New Dimension**

the biggest thing we are working on is our knowledge system. imagine being able to create entire universes of lore that your characters actually understand and reference naturally.

### **Editor Preview**

![editorview](/news/static/editorview.gif)
![detailsview](/news/static/detailsview.gif)

### **What&apos;s Coming:**

- Characters that remember time
- Vector embedding so users can bring entire lore universes into chats
- Better search/discovery so you can find the hidden gems
- Creator Analytics
- Community Building Tools

the possibilities are fucking endless. building tools that let you go as deep or as wide as you want. some of you will build entire fantasy worlds with histories spanning thousands of years. others will create deeply personal characters with rich inner lives. and some will just make stuff that makes us lol.

that is the whole point - we provide the canvas, you create the masterpiece.

ive had limited bandwidth lately, but we&apos;re expanding the team and accelerating development. ambitious but we&apos;re making real progress every day.

## Tech That Gets Out of Your Way

we&apos;re obsessed with making the technical stuff invisible. you shouldnt have to understand vector embeddings or knowledge graphs to create amazing characters. the tools should just work, giving you the power to focus on the creative part.

our new features are designed to be both incredibly powerful and stupidly simple to use. drag and drop lore into your character&apos;s knowledge base, and they will just get it. no more copy-pasting context or trying to cram everything into a character card.

## The Community

our community is what sets us apart. you all keep pushing the boundaries of what&apos;s possible, finding creative workarounds, and sharing techniques that are very cool, wink wink to the css discussion thread.

we are still a small team, but we&apos;re building infrastructure that punches way above our weight. i couldnt have more fun working on anything else. **janitorai** is special, and we want it around forever.

ultimately we just wanna make the tools as wide as possible and see what you create. thats the whole vision - giving you the power to build worlds limited only by your imagination, not by our tech.

Lets keep pushing through together.
![familia](/news/static/vin-diesel-staring.gif)</content:encoded><category>post</category></item></channel></rss>