Janitor AI occupies a different niche from every other platform in this directory. It is not a game engine, not a VTT, and not a narrative framework — it is a character library with a model layer underneath it. That sounds like a lesser thing, but for a specific type of player, it is exactly the right thing: a vast, searchable catalogue of AI personas they can talk to using whatever frontier model they already pay for.

How It Works

The architecture is straightforward. Users browse or create character cards — structured files containing a character’s personality, backstory, example dialogue, and behavioral rules — and launch chat sessions against them. The distinguishing feature is the model layer: Janitor AI lets you connect your own API key from OpenAI, Anthropic, or other providers, routing your conversations through whichever model you choose rather than locking you into a proprietary backend.

For users without API keys, the platform runs Janitor LLM, an in-house model tuned for character-consistency tasks. It is competent for casual sessions; players who connect a frontier model will notice a meaningful difference in prose quality and instruction adherence over long exchanges.

Character cards follow a documented format compatible with the wider SillyTavern ecosystem, meaning cards created for one platform often transfer to the other. A Groups feature supports multi-character scenes, allowing several AI personas to participate in a single conversation thread.

Advantages Over Competitors

The library size is the headline advantage and it is not close. Janitor AI hosts more community-created characters than any other platform — the catalogue spans every genre, franchise, original fiction, and character archetype with a level of variety that no curated platform can match. If you are looking for a specific character type, the odds that someone has already built it here are high.

The bring-your-own-model feature is the second advantage, and it is architecturally significant. Most platforms build their experience around a fixed model tier — you get what they give you. Janitor AI inverts this: the character layer is theirs, the model is yours. For players who are already paying for Claude Pro or ChatGPT Plus, this means accessing a massive character catalogue without paying a second subscription — you pay only the API token cost of your existing plan.

Content restrictions are meaningfully lighter than running the same model directly through its native interface. The character card layer mediates requests in ways that can expand what a model will engage with, which matters for players who were frustrated by filter interruptions on Claude or ChatGPT.

What It Does Best (and Worst)

Janitor AI is best at one-on-one character chat and multi-character social roleplay — the kind of session where you want a specific persona to talk to, not a world to explore. It is where the platform’s character-first design becomes a genuine strength rather than a limitation.

It is worse at sustained campaign play. There is no state persistence, no mechanical layer, no world simulation — the platform treats each message as a generative response to the prior context, and nothing more. Players who want consequences to stack, economies to track, or worlds to remember them will exhaust what Janitor AI offers quickly. The experience is as deep as the character card and the current context window, and no deeper.

Limitations

Card quality is inconsistent. The community library has exceptional highs and a long tail of poorly written cards with thin characterisation, contradictory instructions, or dialogue examples that actively degrade model behaviour. Evaluating a card before committing to a session requires some experience reading the format.

Group conversations degrade faster than solo sessions. When multiple characters are active, the model’s ability to maintain distinct voices under a single context window is tested — long threads can see character voices blur or bleed into each other, particularly on lighter models.

Mobile experience via the browser is adequate, not polished. Players who primarily use phones will find the interface functional rather than designed-for-mobile.

Pricing

Plan	Cost	Model Access	Notable Features
Free	$0	Janitor LLM	Full character library, limited daily messages
BYOM	$0 (API costs only)	Your API key (Claude, OpenAI, etc.)	Unlimited sessions, full model quality
Janitor Pro	~$9.99/month	Priority Janitor LLM + BYOM	Higher message limits, early feature access

API costs depend on the model you connect. A typical session on Claude Sonnet runs a few cents; heavier Opus use compounds quickly for long-format players.

Verdict

Janitor AI is the right tool for a specific use case: players who want a large, searchable catalogue of community characters and the freedom to run them on a model they already own. The bring-your-own-model design is genuinely well-conceived — it turns an existing subscription into access to tens of thousands of characters at minimal additional cost.

It is the wrong tool for players who want a game rather than a conversation. No mechanics, no persistence, no world simulation. But for companion roleplay, social fiction, and character-driven chat at scale, it is the best library that exists in this space, and the model flexibility makes it uniquely practical for players who already know what AI they prefer.