What it is
The long conversation is the whole point
A character chat platform for people who ran into the ceiling on every other one. You bring a key, we handle the part that everybody else does badly.
Locked to their models
Any model you can reach with your own key, chosen per chat.
Forgets what happened an hour ago
Detail from early in a long story comes back when the scene calls for it.
Opaque about what it sends
Every reply shows you what the model was actually given.
Your characters live on their site
Import what you already have, export what you make.
Features
Built for stories that go long
Everything you expect from a character chat site, plus the one thing none of them get right.
Memory that holds
Three hundred messages into a story, a detail from message six still comes back when the scene calls for it. Not a vague echo of it: the thing itself.
Your key, your models
Bring the OpenRouter key you already have and pick whatever model suits the scene. Switch it mid-story if you want. We never mark up inference, because we never touch it.
Lorebooks that behave
Write your world once. The right entries surface when the conversation reaches them and stay quiet when it does not, so your world never crowds out the story.
Personas
Decide who you are in the scene, per chat. The same character can meet a different you without any rewriting.
Images, if you want them
Optional image generation, off until you turn it on, running on the same key you already brought. Leave it off and Komorea stays a text platform.
See what the model saw
Every reply comes with a breakdown of what went into it and what did not fit. When something reads oddly, you can see why instead of guessing.
Bring your existing cards
Character Card v1, v2 and v3, as JSON or as a card PNG. Your chub and Janitor collection imports as it is, embedded world info included.
No overflow, ever
You will not lose a reply to a context limit and you will not watch a long chat quietly get worse. Long conversations stay steady all the way down.
Publish and discover
Keep a character private, or publish it with tags and let other people find it. Search and browse without an account.
Measured
We tested it against the way everyone else does it
Everyone says their memory is good. We ran the test and published what came back, including the parts that were not flattering.
We wrote a 300-message conversation with a character built for the purpose, planted 23 specific details through it, then asked for every one of them back at the end. Names, numbers, a promise made 280 messages earlier, a berth number that changed halfway through.
Then we ran the same conversation again with the memory turned off, cutting the oldest messages the way most sites do, and asked the same questions.
16k context
- Komorea23 / 23
- Cutting old messages6 / 23
Details still reachable after 300 messages.
8k context
- Komorea22 / 23
- Cutting old messages3 / 23
Details still reachable after 300 messages.
At the tighter setting Komorea could see 45 of the 300 messages and still found 22 of the 23 details. Cutting old messages found three.
One run, one model, 23 details, on our own staging machine. It is a real measurement rather than a benchmark anyone else can reproduce yet, and the write-up says so, along with the two bugs the run found in our own code and the one thing it does worse than we expected.
Under the hood
Powered by the Komorea framework
Komorea is not a chat window with a prompt taped to the front. Underneath it sits an engine of our own, built and tested entirely on its own, whose only job is to decide what the model gets to see.
That decision is the whole game. Get it right and a character holds together for a thousand messages; get it wrong and it forgets your name by Tuesday. Every other feature on this page rests on it.
How it works is ours. What it does for you is on this page.
Deterministic
The same conversation always produces the same setup for the model. No lucky runs, no mystery bad days.
Measured, not guessed
Nothing is thrown away by accident and nothing important is quietly dropped to make room.
Accountable
The engine reports what it decided and why, and the site shows you that report on every single reply.
How it runs
Boring where it should be boring
Komorea is cloud-native from the first commit. k3s, ArgoCD and a pipeline that deploys from Git, so the thing running in production is exactly the thing in the repository.
Declared, not clicked
Every part of Komorea is described in a repository and applied by GitOps. Nothing runs because somebody once configured it by hand and forgot.
Self-healing
It runs on a Kubernetes cluster that puts things back the way they are supposed to be. A node falling over is an inconvenience, not an outage.
Portable
No managed service holds us hostage. The same setup comes up on a different provider, or on hardware we own, without a rewrite.
None of this is something you should ever have to think about. It is here because it is why a long chat stays up and why an upgrade does not eat your evening.
Bring your own key
We never sit between you and your model
No subscription, no credits, no markup on tokens. You already pay a provider; Komorea just gives that key somewhere better to live.
Browser only
- Your key is stored in this browser and nowhere else
- Sent straight to your provider for the request at hand
- We could not hand it over if we were asked to
Saved to your account
- Encrypted per account, so your chats follow you across devices
- Background work keeps running while you are away
- Entirely your call, and reversible at any time
Image generation works the same way. It is off by default, you switch it on per chat, and it bills to your key exactly like text does. We take nothing either way.
Where the lines are
Grown up about grown-up fiction
Stated up front, because a policy you have to discover by getting banned is not a policy.
Adults, behind a gate
Adult fiction is allowed. It is invisible, not blurred, to anyone who has not attested to being 18 or over, and to anyone who would rather not see it.
Hard lines, no exceptions
Nothing sexual involving minors, in any framing. Nothing sexualizing real people. Nothing facilitating real-world harm. These are not judgement calls.
Your chats are yours
We moderate what gets published: characters, lorebooks, public material. Private conversations are not reviewed unless you report them yourself.
Where it stands
Built, tested, and waiting on lawyers
The product is finished. Komorea is feature complete and running in private testing right now. What is left is legal: data protection paperwork, and getting the privacy policy and terms right rather than fast.
- The engine, and the tests that keep it honest
- Accounts, keys and streaming replies
- Characters, lorebooks and personas
- Long-term memory across a whole story
- Optional image generation on your own key
- Publishing, discovery, reports and moderation
- Importing your existing character cards
- Cluster, deployments, monitoring and backup drills
- Privacy policy, terms and the data protection paperwork
We are still taking testers
Private testing is open and we are happy to have more people in it. Join the Discord, send us a message saying you would like in, and we will sort you out. No application, no waitlist position, no form.
Ask for an inviteQuestions
The things people ask first
What does Komorea cost?
Nothing. Komorea has no subscription and no credits. You use your own provider key, so you pay your provider directly at their prices, with nothing added on top.
Which models can I use?
Anything your key reaches. OpenRouter is supported first because one key there covers most models people actually want, and direct provider support follows. You choose the model per chat and can change it mid-story.
Do I have to give you my API key?
No. By default your key is kept in your own browser and never stored on our servers at all. Saving it to your account is an opt-in convenience for chatting across devices, and you can undo it whenever you like.
Can I bring my characters from other sites?
Yes. Komorea imports Character Card v1, v2 and v3, as JSON files or as card PNGs, and embedded world info comes across as a lorebook attached to the character.
What makes the memory different?
Most sites simply cut off the oldest part of a conversation, so a long story slowly loses its own history. Komorea decides what the model sees deliberately rather than by truncation, which is why detail from early in a story can still surface hundreds of messages later. It also shows you that decision on every reply.
How well does the memory actually work?
We measured it rather than asserting it. In a 300-message test conversation with 23 specific details planted through it, Komorea could still reach all 23 at a 16k context setting and 22 of 23 at 8k, where only 45 of the 300 messages were visible at once. Cutting the oldest messages instead, which is what most sites do, reached 6 and 3. That is one run against one model on our own machine, and the full write-up includes what it does badly as well.
Is adult content allowed?
Yes, between adult characters, behind an age gate, and it is entirely absent for anyone who has not attested to being 18 or over or who has opted out. Content involving minors is prohibited without exception.
Can Komorea generate images?
Yes, optionally. Image generation is off unless you turn it on, you enable it per chat, and it runs on the same key you already brought, so it bills to your provider at their prices. If you never switch it on, Komorea is a text platform and nothing changes.
When does it launch?
The product is done. Komorea is feature complete and running in private testing today; what is left is the legal side, mainly data protection compliance and getting the privacy policy and terms right. That is slower than writing software, and we would rather be late than sloppy about your data. The Discord is where the launch gets announced first.
Can I get into private testing?
Yes, and we would like you to. Join the Discord and send us a message saying you want in. There is no form, no waitlist and no application to write.
まもなく
Be there when it opens
Private testing is open and we are still taking people. Join the Discord, message us that you would like in, and you are in. It is also where the launch gets announced and where the people building this answer questions directly.
Join the Discord