Okay so. TechCrunch just estimated 337 revenue-generating AI companion companies exist globally right now. Two hundred and twenty million downloads across the whole space. That’s fucking wild.
I’ve tried maybe 30 of these things in the past year. Most are garbage.
Not even interesting garbage. Just… thin wrappers around GPT-3.5 with aggressive paywalls and UI that looks like someone’s first React tutorial. Half of them die within six months. The other half limp along with the same five features every other platform has.
Character AI used to dominate. Now? Everyone’s jumping ship. The censorship got brutal (I watched someone’s medieval tavern scene get content-warned for “violence” when a character drew a sword), the memory became inconsistent, and honestly the whole vibe shifted from “roleplay platform” to “safe chatbot experience for investors.”
what actually matters for roleplay (spoiler: it’s not “uncensored”)
Look. Everyone talks about censorship. Yeah, it matters. But uncensored chat is table stakes now, not a feature.
What matters when you’re three sessions deep into a campaign?
Memory. Real persistence.
Your NPC companion needs to remember that shopkeeper you killed in session one. The town guard should still be looking for you. That betrayal two weeks ago? Should still affect trust levels. Most platforms forget everything after like 4000 tokens. Some reset between sessions entirely.
I tested one “Character AI alternative” that proudly advertised “unlimited messaging” but literally forgot major plot points within the same conversation. Cool feature bro. Very useful.
Memory is the difference between a chatbot and an actual world. Between roleplay and just… talking to a wall that responds.
the multi-npc problem nobody solved yet
Here’s where it gets worse.
Most AI roleplay platforms give you one character. Maybe they let you switch between characters in different chats. But actual group dynamics? Multiple NPCs interacting with each other AND you?
Almost nobody does this right.
I want the party banter from Baldur’s Gate 3. I want companions who have relationships with EACH OTHER, not just me. I want to watch the rogue and the cleric argue about looting the temple while my character decides which side to support.
Soulkyn’s multi-NPC system is one of the few that actually attempts this — up to three AIs in a group chat, each with their own memory, personality, and dynamic stats tracking Trust, Energy, and Affection levels. Not just three chatbots taking turns. Actual emergent behavior between characters.
(Also their character breeding system is batshit. You can combine traits from 130 different “Souls” to create custom NPCs. It’s like Pokémon breeding but for D&D companions.)
the local llm rabbit hole (if you’re technical)
Okay brief tangent. If you’ve got the hardware and the patience, local models are getting genuinely good.
Psyfighter2, Chronos Hermes, some of the Llama 3 finetunes — these can run locally on decent gaming rigs. Zero censorship. Complete privacy. Full control.
But. You need to know what you’re doing. Most people aren’t setting up Oobabooga or KoboldCpp. They’re not tuning sampler settings or managing context windows. They just want to play.
So local models are great for a specific type of user (me, honestly), but they’re not solving the problem for the 220 million people downloading these apps.
what the trash tier platforms do wrong
I’m gonna be specific here because I’m tired of seeing the same mistakes.
Generic waifu catalogs with zero personality differentiation. Fifty “shy anime girl” characters that all respond identically. Cool. Very immersive.
Subscription models that gate basic features. Want your character to remember your name? That’s premium. Want images? Premium. Want more than 50 messages per day? Premium+.
Zero world-building tools. Just character cards. No locations, no faction systems, no consequences. You’re just talking to a character floating in the void.
Mobile-first UI that sucks for actual longform roleplay. Tiny text boxes. No formatting. Can’t even save session logs properly.
“Uncensored” but the model is like 13B parameters and dumb as rocks. Congrats, your character will say “fuck” but can’t maintain narrative coherence for more than three exchanges.
what good platforms do differently
The platforms that actually work for serious AI roleplay have a few things in common.
They use bigger models. 70B parameter models (like what Soulkyn runs) vs 7B or 13B makes a massive difference in coherence, creativity, and memory management. Not even close.
They build memory systems into the architecture. Not just “here’s more context tokens” but actual persistent databases tracking relationships, events, locations. Your world should evolve even when you’re not playing.
They understand genre. Fantasy roleplay has different needs than sci-fi or modern settings. Fantasy AI roleplay specifically needs magic systems, faction reputation, inventory tracking — stuff that generic companion apps just don’t do.
They don’t interrupt your scenes. Content filters that pop up mid-combat or mid-intimate-scene are immersion-destroying garbage. If you’re running an uncensored model, commit to it. Don’t half-ass safety theater.
They give you control. Regenerate responses. Edit character memories. Adjust personality sliders. Save/load states. Actual tools for crafting your experience.
the character ai exodus is real
I’m watching this happen in real-time on Reddit and Discord. People are leaving Character AI in waves.
Some go to Janitor AI (web wrapper with inconsistent quality). Some try Crushon (better but limited). Some end up on Soulkyn (full platform with actual features). Some give up and go back to text-based MUDs from 1995 which honestly, fair.
The thing is? Character AI built the market. They showed millions of people that AI roleplay could be genuinely engaging. Then they slowly made the product worse (from a user perspective, probably better for investors) and created this huge appetite for alternatives.
So now we’ve got 337 companies trying to capture that audience. Most of them suck. Some will survive. A few might actually be good.
what i actually use now
Full transparency: I rotate between three platforms depending on what I’m doing.
Soulkyn for longform campaigns where I need persistent memory and multiple NPCs. The 70B model handles complex narratives way better than smaller alternatives, and the dynamic stats system adds actual mechanical depth. (Trust levels affecting dialogue options, Energy determining scene pacing, Affection unlocking character arcs — it’s more game-like than most platforms.)
Local Chronos Hermes when I want complete privacy and total control. Slower to set up, faster to run once configured. Perfect for experimental stuff or personal narratives I don’t want on anyone’s servers.
Claude (via API) for one-off creative writing where I need really high-quality prose and don’t care about persistence. Not a roleplay platform per se, but the writing quality is insane.
where this space is going (probably)
Hot takes:
Consolidation incoming. 337 companies won’t survive. Maybe 20-30 will capture most users. The rest will die or get acquired.
Memory will be the moat. Whoever builds the best long-term persistent memory system wins. Not just RAG, but actual semantic understanding of character relationships and world state.
Local models will stay niche. They’ll get better, easier to run, more accessible — but most users still won’t bother. Cloud platforms with good UX will dominate.
Genre-specific platforms will emerge. Right now most apps try to do everything. I think we’ll see fantasy-specific, sci-fi-specific, romance-specific platforms that do one thing really well.
The “companion” framing will fade. People don’t want one AI girlfriend. They want entire worlds, parties, factions, ecosystems. Multi-agent systems are the future.
the actual problem with most platforms
Okay cutting through the noise: most AI companion companies don’t actually care about roleplay.
They care about retention metrics and subscription revenue. Which, fine, businesses need to make money. But it creates this fundamental mismatch.
Good roleplay requires:
- Long-term narrative coherence
- Complex character relationships
- Emergent storytelling
- Player agency and consequences
What gets you retention metrics:
- Daily login rewards
- Character gacha mechanics
- Artificial scarcity (message limits)
- Parasocial attachment to single characters
These aren’t the same thing. Some platforms thread the needle (Soulkyn’s character breeding system is technically gacha-adjacent but actually serves worldbuilding). Most don’t even try.
if you’re shopping for a platform right now
Things to actually test:
Start a session. Close the app. Come back three days later. Does it remember? Does it hallucinate fake memories? Does it reset completely?
Bring up something from early in the conversation. See if the AI can reference it accurately. Most can’t.
Try to break the content filter (if there is one). I’m serious. You need to know where the boundaries are and whether they’ll ruin your immersion.
Check if you can export your data. Logs, character cards, world state. If you can’t export, you don’t own your stories.
Test multi-character scenarios if possible. Do NPCs interact naturally or do they just take turns talking AT you?
final thought
Look. The AI roleplay space is messy right now. Hundreds of platforms, most are trash, a few are genuinely good, everything’s changing fast.
But here’s the thing: the fact that 337 companies exist and 220 million people downloaded these apps means the demand is real. People WANT this. They want persistent worlds, meaningful choices, characters that remember them.
The platforms that figure out how to deliver that (without aggressive monetization that ruins the experience) are gonna win.
Everyone else? Just thin wrappers on GPT that’ll be dead in a year.
Choose accordingly.