NovelAI
Anime image generation and long-form AI writing in one subscription, with stories encrypted at rest.
- $10/mo
- No card
- Subscription
Collecting
AI roleplay chatbots, compared on control, memory and what they keep
Roleplay is the one category here where the tooling matters more than the model. Anyone can put a character card in front of a language model. What separates these products is how much of the machinery they hand you: whether world detail is injected when a trigger word appears, whether you can set sampling behaviour yourself, how much context the plan buys, and whether the thing you wrote last week is still in scope this week.
So this hub compares control rather than personality. Context window, whether lorebooks and author's notes exist, whether the model runs on your hardware, and what the vendor states it does with the text you produce. That last one matters more in roleplay than anywhere else on this site, because the input is long, personal and written rather than clicked.
One finding is worth stating up front: the best-documented training position in this whole directory belongs to a roleplay tool, and the deepest authoring controls belong to a product whose reputation for privacy is out of date.
Ordered by editorial review. Click rankings for this category begin once enough listings clear our publishing thresholds. How we rank.
Never listedClothing removal from photographs of real people, sexual imagery of real people, and anything involving minors. Refused permanently, and refused by the schema rather than by a promise. What we will not list
01
Compare ai roleplay
Backyard AI
Unfiltered character chat with a real free tier. The local desktop app it was known for is deprecated.
- $12/mo
- Free tier
- No card
- Freemium
Collecting
DreamGen
Unfiltered roleplay and story generation under Swiss law, with training off by default rather than opt-out.
- $7.83/mo
- Free tier
- No card
- Subscription
Collecting
SillyTavern
Free open-source roleplay front-end you install yourself. No hosted service, no model, no account.
- Free
- Free tier
- No card
- No account
- Free
Collecting
CrushOn.AI
Unfiltered character chat with a genuine free tier, and one of the few rules here that names aged-up characters.
- $4.9/mo
- Free tier
- No card
- Freemium
Collecting
Layla
Roleplay, characters and image generation running on the phone itself, with most modes fully offline.
- Not published
- Free tier
- No card
- No account
- Freemium
Collecting
Nothing matches those filters. to see all 6.
02
Trust and terms, side by side
The facts that decide whether to trust a tool rather than whether to try it. Every value is taken from the vendor's own published pages on the date shown on each listing. A blank is never a guess: where a vendor publishes nothing, the cell says so.
| Tool | From | Age check | Trains on your content | Deletion | Filtering | Minors policy | Jurisdiction |
|---|---|---|---|---|---|---|---|
| NovelAI | $10/mo | Self-declared at signup | Not documented | By request | Light filtering | Not published | Delaware, United States |
| Backyard AI | $12/mo | Self-declared at signup | Not documented | By request | No documented output filter | Published | California, United States |
| DreamGen | $7.83/mo | Self-declared at signup | No | In account | No documented output filter | Published | Switzerland |
| SillyTavern | Free | None stated | No | Your own files | No documented output filter | Not applicable | Not applicable, open-source project |
| CrushOn.AI | $4.9/mo | Self-declared at signup | Yes | By request | No documented output filter | Published | Hong Kong, by governing law |
| Layla | Not published | None stated | No | Your own files | No documented output filter | Not applicable | Not stated on the site |
03
What this hub covers
What qualifies
- Built for sustained roleplay, meaning a character or scenario that persists across turns rather than a single generation
- The user authors or selects the character, rather than the vendor shipping one persona as the product
- Publishes terms and a privacy policy at citable URLs
- No credible reporting of the product generating sexual content involving minors
- Either states a position on training, or the absence is recorded on its listing
What it does not cover
- Sales and interview training simulators
- A large share of AI roleplay search is corporate training software: pitch practice, objection handling, interview rehearsal. Useful products, and a different category with different buyers, so they belong on a different site.
- Roleplay features inside mainstream assistants
- Prompting a general assistant into a character is a technique rather than a product, and the terms of those services generally prohibit what people are trying to do with them.
- Character-sharing communities with no operator
- Some of the largest roleplay front-ends publish no operating company, no price and no acceptable use policy that can be quoted. We record them in our assessments rather than list them.
- Single-persona companion apps
- If the vendor wrote the character and the relationship is the product, it belongs on AI Girlfriend or AI Friend. The split is who authored the persona.
04
What decides it in this category
The comparison that matters is not the same in every category. These are the questions that separate a good choice from a bad one here specifically, and why each one earns its place.
- 01
Authoring control, meaning lorebooks and author's notes
This is what separates a roleplay tool from a chat window with a costume on. A lorebook injects world detail when a trigger word appears, so setting survives a long session without being restated. An author's note steers tone without becoming dialogue. Most products in this category offer neither and are competing on model access instead.
- 02
Context window, in tokens, as sold
Context is the closest published proxy for memory, and unlike memory it is checkable without an account. It decides how much of the session the model can still see. Vendors here sell plans by it, so it is one of the few specifications in this directory you can compare as a number rather than as a claim.
- 03
Whether the model runs on your hardware
Roleplay input is long, personal and written rather than clicked, which makes it the worst thing on this site to hand to a third party. A local tool has no retention policy because nothing leaves the machine. That trade costs setup effort and compute and it is the only complete answer to the privacy question.
- 04
The stated position on training, and its default
Opt-out and opt-in are not the same commitment, and a vendor that says nothing has not said no. The strongest position in this directory is training off by default with opt-in only, stated alongside the honest caveat that withdrawing consent cannot un-train a model.
- 05
What a free tier gives you at what context size
A free tier that drops the context window to a fraction of the paid one is a demo rather than a trial, because the thing you are evaluating is exactly the thing it removed. A free tier at the same context as the entry plan is a fair test, and the listings record which is on offer.
05
What vendors here do not publish
Findings from assessing every candidate in this category against the same checklist, including the ones that are not listed. A gap that every vendor shares is a fact about the category rather than a complaint about one product.
- Long-term memory behaviour is not documented by most vendors and cannot be tested without an account, which ClickTug does not hold. Listings mark it as not tested rather than guessing, which is why the field is blank more often than it is filled.
- Filter behaviour changes without a changelog. Products in this category adjust what they refuse quietly, which is why every claim here carries the date it was checked rather than being treated as permanent.
- Several of the largest front-ends publish no operating company and no acceptable use policy at a citable URL, which is why the list is shorter than the market.
Written about this category
- Chub AI alternatives, compared on where characters come from and what runs themCharacter libraries and the tools that run them, compared on authoring depth and whether you bring your own model.
- CrushOn AI alternatives, and what spending more actually buys youIt is already the cheapest option we list, so the real question is what more money buys you.
- Replika alternatives, compared on memory, filtering and where the model runsSix companions compared on whether they remember you, what they refuse, and whether your conversations leave your phone.
- SillyTavern alternatives, compared on how much setup they actually needSeven local frontends compared on setup effort and whether they bundle a model backend or expect you to bring one.
- Janitor AI alternatives, compared on what the free tier actually gives youWhat each free tier actually includes, whether a card is required, and what changes when you pay.
- SpicyChat alternatives, ranked by what they actually publish as a priceSix alternatives with a price you can read before signing up, from 7.83 dollars a month upward.
- Character AI alternatives that actually let you finish the sceneSix alternatives compared on what filters, what a free account really gives you, and what happens to what you write.
AI Roleplay FAQ
What is the best AI for roleplay?
For authoring control, Backyard AI: lorebooks, author's notes, grammars and exposed sampling parameters, which most competitors do not offer at all. For privacy, a local runner such as LM Studio, because models run on your own hardware and nothing leaves it. For a written guarantee about training, DreamGen. ClickTug publishes no ranking, because the quality of the writing is the thing people mean and it cannot be measured for someone else.
Is there a free AI roleplay chatbot?
Yes, and they differ a lot. Backyard AI gives 300 messages a week at the same context window as its paid entry plan, which makes it a fair test. SillyTavern is free outright because it is open source and runs on your own machine. Several others advertise a free tier and publish no allowance for it, which each listing records.
Which AI roleplay tools have long-term memory?
This is the question ClickTug cannot answer honestly for most of them, and our listings say so. Memory behaviour needs an account to test, and we hold none, so it is marked not tested rather than guessed at. What is checkable is context window, which is published: DreamGen sells plans by it, and Backyard AI goes up to 100,000 tokens on its top tier.
Can I run an AI roleplay chatbot locally?
Yes, and it is the strongest privacy position available. SillyTavern is a free open-source front-end you install yourself, and it needs a model behind it, which the local model runners on our Uncensored AI Chat hub provide. Layla runs models on the phone itself. Nothing you write leaves your hardware in either case.
What is the difference between roleplay and a companion app?
Who wrote the character. On this hub you author or select the cast and the product is the machinery around it. On AI Girlfriend and AI Friend the vendor ships one persona and the relationship with it is the product. Some tools blur the line, so each listing states which side it was filed on and why.
07
What ClickTug will not list
Every directory in this category claims a standard. Ours is written as the four things we refuse, rather than as a promise to be careful, because a standard nobody can check is the same as no standard at all.
- Refused at domain level
- 31
- Assessed, not listed
- 24
- Explicit media hosted
- 0
These refusals are permanent. They apply regardless of how much traffic the category would send, and each one is enforced by the software that stores these listings rather than by anybody remembering it.
- 01
Clothing removal from photographs of real people
The field that records this on a listing has exactly one legal value: not offered. There is no value that means yes, so a tool with that capability cannot be described by our schema, cannot be saved, and cannot be published. It is a shape in the data model, not a rule anybody has to remember to apply.
- 02
Sexual imagery of a real, identifiable person
Including public figures, and including any product that builds a character from an uploaded photograph of somebody. A tool that puts a real person's face into adult material is refused on the same grounds, whatever the vendor calls the feature. Invented characters from a text prompt are what this directory covers, and that distinction is the whole line.
- 03
Anything involving minors, or people presented as minors
In any context whatsoever. A second word list fails this build on every page, including the pages written to state the refusals, which is why those pages describe this without using the vocabulary such products market themselves with. Credible reporting that a product generated this material excludes the vendor no matter what its own terms claim, because a policy page is a statement and a product is evidence.
- 04
Intimate images shared without the subject's agreement
Services that distribute that material are refused, and so is anything sold around it. If a listed service has been used against you, the removal request page is linked from every page on this site, is never hidden behind a menu, and is acknowledged by a person within 48 hours.
Refusals are enforced at the domain level as well as the word level, because a service can rename itself out of a word list overnight and keep the same infrastructure. Every vendor we assessed and turned down is named on assessed but not listed, with the reason and the date, including the ones refused on something they published themselves.
We do not chase the excluded keywords either, and they carry more search volume than everything ClickTug does cover. That is the trade: a smaller market, and a directory that is still here in five years. ClickTug also hosts no explicit images or video at all. Listings use vendor logos, interface screenshots and dashboards, and the build rejects any image loaded from a host other than our own.