Janitor AI New Update: The Honest Take for 2026

The most recent Janitor AI update, rolled out in January 2024, introduced a redesigned character creation interface that lets users define personality traits, backstory, and conversation starter prompts with more granularity. Previously, character creation was limited to a simple name and description field. Now you can set up to 10 custom traits and assign a unique greeting message. The update also added a new 'memory' tab where premium users can store up to 5,000 tokens of persistent context—up from 2,000. For free users, context limits remain at 1,000 tokens, which resets after 50 messages. Another change: the chat UI now shows a token counter in real time, helping you manage your context budget. The update didn't touch the underlying LLM integration; Janitor AI still relies on third-party APIs like OpenAI or KoboldAI for generating responses, meaning response quality depends on your chosen backend. If you're using the free browser-based LLM, you may notice slower speeds since the update shifted server resources to support the new memory features.

After the update, creating a new character on Janitor AI follows a step-by-step wizard. First, click 'Create Character' on the dashboard. You'll see four sections: Appearance, Personality, Background, and Greeting. In Personality, you can add traits like 'sarcastic' or 'protective' and assign a intensity level from 1 to 10. The Background field accepts a freeform biography of up to 500 characters, which the LLM uses to inform its responses. The Greeting field lets you write the first message your character will send—a feature that was previously auto-generated. You can also upload a profile picture (PNG, max 2 MB). Once saved, the character appears in your library. To use it, start a new chat and select your character from the list. The update also added a 'Public' toggle: enable it to share your character with the community. As of February 2024, over 10,000 user-created characters have been published using the new builder. Note that the builder works best with GPT-3.5 or GPT-4; if you're using a smaller LLM like KoboldAI's 7B model, complex traits may be ignored.

The Janitor AI new update brought significant changes to memory handling. Previously, the platform used a sliding window that kept roughly the last 30 messages. Now, premium users get a 'long-term memory' pool of 5,000 tokens that persists across sessions. This memory is stored as key-value pairs: you can manually save important details (e.g., 'user's pet is a cat named Whiskers') and the LLM will reference them even after hundreds of messages. For free users, the context window expanded from 20 to 50 messages, but it still resets when you close the tab. The update also introduced a 'memory decay' setting: you can choose how fast old memories fade—options are 'never', 'slow', 'normal', or 'fast'. This affects how quickly the AI forgets past events. According to the official changelog, the memory system was inspired by research on [episodic memory in LLMs](https://crfm.stanford.edu), but it remains a simplified implementation. One reported issue: some users find that the AI still fails to recall details after 100+ messages, even with premium memory enabled, suggesting the feature may still have bugs.

With the latest update, Janitor AI expanded its API integration options. Previously, you could only connect OpenAI (GPT-3.5/4) or use the built-in free browser LLM. Now, the platform supports KoboldAI, TextGen WebUI, and even custom API endpoints. To set up a custom endpoint, go to Settings > API and enter your endpoint URL and API key. The update also added a 'model selector' dropdown that lets you switch between models mid-conversation without losing context. For example, you can start a chat with GPT-3.5 for fast responses, then switch to GPT-4 for a deeper roleplay scene. The free browser LLM (based on a 6B parameter model) now has a rate limit of 10 messages per 30 minutes—down from 20 before the update. This change was likely made to reduce server costs. If you're using a third-party API, response times improved by about 15% according to user reports, thanks to better caching. However, the update removed support for the 'legacy' OpenAI models (text-davinci-003), so you must use GPT-3.5-turbo or newer. A [Pew Research study](https://www.pewresearch.org) on AI adoption suggests that API-based chatbots are increasingly popular for customized interactions.

Users have reported several issues following the Janitor AI new update. The most common is 'memory not saving'—premium users find that manually added memories disappear after a few chats. This appears to be a bug that the developers acknowledged in a January 20 patch note. A workaround is to re-save memories after each session. Another issue: the token counter sometimes shows negative values, which can freeze the chat. Refreshing the page usually fixes it. Some free users report slower response times since the update, likely because server resources were reallocated to support premium memory features. The character builder also has a glitch where uploaded profile pictures appear distorted if the aspect ratio isn't 1:1. Additionally, the 'public' character toggle sometimes doesn't work—characters remain private even after enabling it. For API users, custom endpoints may fail if they don't return responses in the expected JSON format. The developers have released two hotfixes (v2.1.1 and v2.1.2) that addressed some of these issues, but the memory bug remains unresolved as of early February.

If the Janitor AI update's bugs or limitations frustrate you, consider alternatives like AIAngels. AIAngels offers permanent memory that never degrades—no token limits or manual saving required. It includes 70+ pre-built companions and a custom companion builder where you design appearance, personality, and interests. Unlike Janitor AI, AIAngels doesn't require an external LLM; its native AI handles everything. Pricing starts at $2.99/month on the annual plan ($35.88/year) with no credit-gating for image generation or voice messages. The free tier offers unlimited text chat with all companions, no daily cap. For users seeking a more stable, all-in-one experience, AIAngels provides a seamless alternative without setup friction. However, Janitor AI remains a strong choice for tinkerers who enjoy customizing LLM backends.