If you used KoboldAI for adult fiction and roleplay and would rather not run the model on your own hardware, the alternative we make is Hemmingway Unlocked: an app in your browser, and an API that works in SillyTavern. If you want everything to stay on your own machine, keep running models locally, with KoboldCpp, the project KoboldAI now points to, or another local runner.
Hemmingway Unlocked is our uncensored AI for adults. Its weights are public too, as an adapter on our open-weights model Hemmingway-1, but running them yourself needs a big GPU set-up. We make Hemmingway Unlocked. What we say about the others comes from their own pages: their READMEs, docs and wiki.
What KoboldAI is
Two projects share the name.
- KoboldAI, the original. Its README describes a browser-based front end for AI-assisted writing, with local and remote models. It has Memory, Author's Note, World Info and saved stories, and you can switch between writing a novel, playing a text adventure and chatting. The same README now warns that this version is a legacy product, no longer maintained, that will not run modern models, and points to KoboldCpp.
- KoboldCpp. Its README describes free, open-source software that runs GGUF models on your own computer. It is one file with nothing to install, for Windows, Linux and macOS, on CPU or GPU. It bundles the KoboldAI Lite web page, with memory, world info, author's notes, characters, scenarios, saved stories and Tavern character cards. With the right model files it also makes images, video, speech and music.
- Models are separate. KoboldCpp ships with no model. You download one, and its wiki says the memory you need depends on the model's size, its quantisation and the context length.
- An API on your machine. KoboldCpp serves an OpenAI-compatible API at
http://localhost:5001/v1, so other apps, SillyTavern among them, can use the model it runs. - Other ways in. Its README lists Google Colab, rented cloud GPUs and an online copy of KoboldAI Lite. The web page can also connect to outside providers, OpenAI-compatible ones included, with no local model loaded.
Why people look elsewhere
- They are still on the old KoboldAI. Its own README sends them to KoboldCpp.
- Hardware. Models that write long scenes well are large, and large models need a lot of memory. A laptop runs the small ones, and the small ones tend to write weaker scenes.
- Set-up. Choosing a model file, a quantisation, how many layers go on the GPU and a chat template takes time.
- A model made for adult fiction, with its rules written down. Instead of trying model files until one writes the scenes they want, some people want one model built for that job, with a published list of what it will not do.
The options side by side
| Option | What it is | Good for | Watch out for |
|---|---|---|---|
| Hemmingway Unlocked | Our hosted adults-only app and API | Adult fiction and roleplay with no hardware, in the browser or in SillyTavern | Each message passes through our servers. Text only |
| Hemmingway Unlocked's weights | Unlocked's model, run on your own GPUs | Unlocked with nothing sent anywhere | A big GPU set-up, and a non-commercial licence |
| KoboldCpp | KoboldAI's successor, one program that runs models locally | Any GGUF model, with nothing sent anywhere, plus images and speech | Your hardware sets the limit |
| SillyTavern | A front end you install, with no model of its own | Character cards and group chats over any local or hosted model | Set-up, and a model to connect |
| Ollama | Another way to run open models, locally or in its cloud | A local server for other apps | Its cloud models run on its servers, not yours |
KoboldCpp
If you are moving from the original KoboldAI, this is where its own README sends you. KoboldAI Lite keeps the tools you know: memory, world info and author's notes, plus characters and scenarios. It loads Tavern character cards, and its wiki says SillyTavern can use KoboldCpp as its back end. With a local model, what you type stays on your computer.
The limit is your machine. Its README suggests starting with a smaller model if you are unsure what it can run.
SillyTavern
SillyTavern's docs describe a locally installed interface for text models, image engines and voice models. It has no model of its own. It connects to KoboldAI back ends, OpenAI-compatible APIs, AI Horde and others. It has character cards, World Info, group chats and personas, and it is free and open source.
Its use here is as the layer on top. Run it over KoboldCpp at home, point it at a hosted API when your hardware is not enough, and keep the same cards for both. SillyTavern API explains the connection step.
Ollama
Ollama's site describes a way to run open models on your own computer, or in its cloud without a local GPU, and its docs say its API works with OpenAI-compatible clients. Its site says prompts sent to its cloud are never stored or trained on. Its pages describe running models, not story tools, so for writing you would put a front end such as SillyTavern on top.
Hemmingway Unlocked's weights
Unlocked's model is public as an adapter, a LoRA, on Hemmingway-1, at Hugging Face. It is under CC BY-NC 4.0: free for non-commercial use, with commercial use by agreement. Hemmingway-1 is a 27B open-weights model built on Qwen3.8-27B.
At 16-bit, the weights alone are about 54 GB, so this needs a big GPU set-up. Run Hemmingway Unlocked locally goes through it.
Where hosted Hemmingway Unlocked fits
Hemmingway Unlocked is the adults-only side of Hemmingway, at wise.hemmingway.io. It asks you to confirm you are 18 or over, shows you its rules, then asks you to sign in. It suits you when you want the model without the hardware.
- What it writes. Sex between adults, explicit and in full. Dark, violent fiction: horror, war, crime. Roleplay where you set the scene and it plays its part, and ordinary chat. It does not stop a scene to lecture you.
- What it never writes. Anything sexual involving anyone under 18 or an adult playing someone under 18, a real person, animals, family (step, half, adoptive and foster included) or anyone who does not want it or cannot say yes. Never torture or help hurting a real person, stalking, doxxing, harassing or scamming, weapons, explosives, making drugs, harmful software, or help with suicide or self-harm. The full list is in our terms, and every answer has a Report button.
- Where chats live. In the app, they are kept in your browser, on your device, not on our servers, and never used for training. Each message passes through our servers to be answered and is not kept.
- Cost. Free to try in the app, with no card. After the trial, memberships start at $9 a month. Its page has the rest.
- The API. OpenAI-compatible, at
https://wise.hemmingway.io/v1. You make the key inside Unlocked, in the account menu under API keys, and it needs a membership. An Unlocked key always reaches Unlocked's model, whatever model name you send.
KoboldCpp's README lists OpenAI-compatible providers among those its web page can connect to, but SillyTavern is the one our set-up page covers.
Open Hemmingway UnlockedTry Hemmingway
Where it does not fit
- Nothing may leave your computer. Hosted Unlocked passes each message through our servers. Run a model locally, or run Unlocked's weights if you have the GPUs.
- No images, voice or music. It writes text. KoboldCpp makes those with the right model files.
- No world info or author's notes in the app, and no character library or builder. You describe the character in your first message. In SillyTavern, your cards and lore work over its API as they do over any model.
- No sync. Chats in the app stay in one browser, and clearing its data deletes them.
How to choose
- Adult fiction and roleplay with no hardware of your own: use hosted Hemmingway Unlocked, in the browser or through SillyTavern.
- Unlocked's model with nothing sent anywhere, and the GPUs to run it: download the weights and follow the guide.
- Still on the original KoboldAI: move to KoboldCpp.
- Nothing leaves your machine, on ordinary hardware: KoboldCpp or Ollama, with a model that fits.
- One set of character cards over several models: put SillyTavern on top.
Common questions
Is KoboldAI still maintained?
The original KoboldAI's README says it is a legacy product that is no longer maintained and will not run modern models. It points to KoboldCpp, which its README describes as free and open-source, with KoboldAI Lite bundled in.
Can I use Hemmingway Unlocked in SillyTavern?
Yes. Choose Chat Completion, then Custom (OpenAI-compatible), and enter https://wise.hemmingway.io/v1 with a key made inside Unlocked. The key needs a membership; the free trial is for the app.
Can I run Hemmingway Unlocked locally?
Yes, if you have a big GPU set-up. Its weights are public as an adapter on Hemmingway-1, a 27B model, under CC BY-NC 4.0. The guide covers it.
Is hosted Hemmingway Unlocked as private as a local model?
No. A model on your own computer sends nothing anywhere, and any hosted service has to receive your message to answer it. In Unlocked, chats are kept in your browser and never on our servers or in training, but each message passes through our servers to be answered and is not kept.