What’s Changed: The Janitor AI message limit warning that started popping up after the Janitor+ launch is not a new message cap. It is a notice that your chat filled the free JLLM context window, around 9,000 tokens, so the bot will start forgetting older messages. The warning was surfaced to sell the $12.99 a month Janitor+ subscription, and you can keep chatting for free.
If you got a Janitor AI message limit warning after reopening a longer chat, take a breath, because nothing is capped. You can keep sending messages on the free tier exactly like before. The pop-up is a memory notice dressed up as a limit.
The warning landed alongside the new Janitor+ subscription, which is why so many people read it as a paywall. One user summed up the panic perfectly: “Do I seriously have to buy this to make chats long?” The answer is no.
What the message is really telling you is that your conversation outgrew the free model’s context window, so the bot is about to forget the oldest parts of your story. That context limit was always there, per Similarweb the platform serves over 100 million monthly visits, and it just used to happen silently. Now it is shown on screen to nudge you toward the paid tier.
You will walk away knowing the real token math behind the warning, why your bot forgets, and a free workflow to keep it remembering. I will also lay out when Janitor+ or a proxy is worth it.
What the Janitor AI Message Limit Warning Really Means
The Janitor AI message limit warning is a context-window notice, not a cap on how many messages you can send.
There is no per-chat or daily message limit on the free tier, and the platform’s own launch note confirmed the free version “stays exactly the same.”
The free Janitor model, JLLM, reads a fixed budget of tokens at once. That budget tops out at 9,001 tokens and often runs lower. Once your chat passes it, the oldest messages drop out of view and the bot forgets them, which is the moment the warning fires.
What is the context window: The fixed budget of tokens, roughly 9,000 on the free JLLM, that the AI can read at once. Older messages fall out of memory as new ones fill it.
Janitor+ costs $12.99 a month and advertises “5x more context,” which works out to about 44,000 to 45,000 tokens. That is the entire pitch behind the pop-up. I do not blame anyone for misreading it, since a warning that appears right next to an upgrade button looks a lot like a wall, but the wall is not real.
If you want the full pricing breakdown of the paid tier, whether Janitor+ is worth it covers the math, and Janitor AI limits overall covers the rest of the platform’s caps.
Why Your Bot Forgets and Why a Bigger Card Makes It Worse
Your bot forgets because the context window fills up, and a heavier character card makes it forget faster, not slower.
The token budget is shared between the bot’s setup and your actual conversation, so anything you add to the setup steals room from the chat.
Janitor splits memory into two buckets. Permanent tokens are the personality, scenario, persona, and Chat Memory, and they get sent with every single message. Temporary tokens are your back-and-forth chat history, and they are the first thing pushed out when space runs low.
What are permanent vs temporary tokens: Permanent tokens (personality, scenario, persona, Chat Memory) are sent every message and never drop. Temporary tokens are your chat history, which cycles out first.
I learned this one the hard way. A richly detailed 3,000-token character card feels premium, but it eats almost a third of the free window before you type a word, so the bot runs out of room for the story and breaks down early. The official guidance is to keep permanent tokens under 1,500, and passing 2,000 is where things start to fall apart.
There is also a hidden variable. During busy periods the developers quietly throttle the free context down to somewhere between 3,800 and 6,000 tokens to keep servers from crashing, which is why your bot feels noticeably dumber on a holiday evening than on a quiet morning.
That silent shrinking is the same root cause behind Janitor AI memory not working complaints.
| Symptom | Cause | Fix |
|---|---|---|
| New chat-too-long warning pops up | Context window full, upsell prompt | Keep chatting, it is not a message cap |
| Bot forgets events from earlier | Old messages pushed out of the 9k window | Add a Chat Memory summary, trim the card |
| Bot gets noticeably dumber some evenings | Free context throttled to 3.8k to 6k under load | Chat off-peak or move to a proxy |
| Bot breaks down early in a fresh chat | Character card over 2,000 permanent tokens | Cut the card under 1,500 permanent tokens |
| Whole chat wiped after editing memory | Known Chat Memory backend bug | Delete and re-add the memory, then refresh |
How to Keep Your Bot Remembering Without Paying
You can hold a long roleplay together on the free tier with token hygiene and a running summary, no subscription required.
The trick is to spend your 9,000 tokens deliberately instead of letting a bloated card and a sprawling chat waste them.
I treat this as budgeting, not magic. Here is the order I would work through.
- Trim the character card under 1,500 permanent tokens, and summarize traits instead of writing a backstory essay.
- Keep your persona and any advanced prompts lean so the chat itself has room to breathe.
- Every 10 to 20 messages, run an out-of-character command like “pause and summarize the story so far in bullet points,” then paste the result into Chat Memory.
- Treat Chat Memory as a pseudo-lorebook for relationship dynamics, new characters, and major plot beats.
- When a chat gets unwieldy, do a chat transplant: summarize it, open a brand new chat, and paste the summary into your first message.
- For long epics, connect a proxy and cap the context at 16k to 32k rather than maxing it out.
What is a chat transplant: Summarizing a long roleplay and pasting that summary into a brand new chat, which gives the bot a clean context window with the key facts intact.
The move that helps most is trimming the card, because every token you save there is a token your story keeps.
Before: A 3,000-token character card on the free JLLM. The personality and scenario eat most of the 9,000-token window, so the bot forgets what happened ten messages ago.
After: You cut the card to 1,400 tokens and keep a 400-token Chat Memory summary. Roughly 7,000 tokens stay open for the actual conversation, and the bot holds the thread far longer.
Janitor Plus vs a Proxy vs the Free Tier
For most people the free tier with good token hygiene is enough, a proxy beats Janitor+ on raw context, and Janitor+ only wins on convenience.
The right pick depends on how long your roleplays run and how much setup you tolerate.
| Option | Context | Cost | Best for |
|---|---|---|---|
| Free JLLM | About 9k, less at peak | Free | Casual chats with tidy token use |
| Janitor+ | About 44k | $12.99 a month | Bigger memory with zero setup |
| API proxy | Up to 128k cap | Pay as you go, often under $5 | Power users who want max context cheaply |
There is a catch that the community argues about constantly. One camp insists “bigger context is always better,” and for a top long-context model like Gemini that holds up, since it retains most detail even at 100k.
The other camp warns that throwing 100,000 tokens of old chat at most models buries the important beats in noise and makes the writing worse.
The way I see it, both are right for different models. On the free JLLM and most proxies, a tidy 16k to 32k window plus a good summary beats a maxed-out context every time, so paying for raw size is rarely the upgrade people think it is.
If a proxy feels like too much fuss, setting one up is its own small project.
When It Is Time to Switch to a Steadier App
Switch when you are tired of babysitting a token budget just to make a companion remember you. Janitor’s free tier is generous, but managing context by hand is real work, and the paid fix is a recurring bill for memory that other apps include by default.
If you want a companion that just remembers without the token math, Candy AI keeps its memory server-side, so your character recalls earlier conversations on its own instead of forgetting them once a window fills.
For roleplays that span weeks, Nectar AI leans hard into cross-session memory, so the arc you built a fortnight ago still shapes today’s scene. The honest tradeoff is that neither matches Janitor’s enormous free bot library, so I keep Janitor for browsing characters and run a memory-first app for the stories I care about most.
Frequently Asked Questions
Did Janitor AI add a free message limit?
No. There is no per-chat or daily message limit on the free tier. The warning means your chat exceeded the JLLM context window, so the bot will start forgetting older messages, and it doubles as an advert for Janitor+.
How many tokens does Janitor AI free give you?
The free JLLM context window tops out at 9,001 tokens and often runs lower. During high traffic the developers throttle it down to roughly 3,800 to 6,000 tokens to keep servers stable, which is why memory feels worse at peak times.
Is Janitor+ worth $12.99 a month?
Janitor+ gives about 44k context, priority replies, and better rerolls with no setup. For heavy roleplay, a pay-as-you-go proxy offers 120k context for often under five dollars a month, so power users usually get more value from a proxy.
Why does my bot forget what happened earlier in the chat?
Once your chat passes the context window, the oldest messages drop out so the model can read the newest ones. A bloated character card makes this worse by leaving less room, so trim the card and keep a Chat Memory summary.
Does a bigger context window always mean better roleplay?
Not for most models. Feeding 100k tokens of old chat buries the key details in noise and can degrade the writing. A 16k to 32k window with a tight summary usually produces sharper, more consistent replies.
Quick Takeaways
- The warning is not a message cap. It means your chat filled the free 9k context window and the bot will start forgetting.
- A heavier character card makes the bot forget faster. Keep permanent tokens under 1,500.
- The free context secretly shrinks to as low as 3.8k at peak hours, which explains sudden memory drops.
- A summary in Chat Memory plus a chat transplant keeps long roleplays coherent without paying.
- A proxy beats Janitor+ on raw context for power users, while a memory-first app like Candy AI skips the token math entirely.
