Documentation

The catalog shows every published card. Search by name, filter by tags and rating (SFW/NSFW), and switch between large cards and a dense grid.

• Chat starts a scene with that card.
• Card opens the full profile: greeting, tags, likes, chats.
• Tags are clickable: click one to filter the catalog by it.

My Characters → New character opens the Studio. The form is split into sections; only Name and Greeting are required to save.

• Identity — the listing: name, creator, title, NSFW tag. The title is for the catalog and never goes into the prompt.
• Voice — Personality hits the prompt first; Description is the body of the character.
• Scene — Greeting is the first line; Scenario locks the place; Alternate greetings are other openings.
• Lorebook — entries that enter the prompt when a keyword shows up in recent lines.
• Advanced — extra instructions and a post-history instruction for this card only.

A card is yours until you publish it. Published cards show in the catalog; drafts stay private.

• Publish / Unpublish — a switch on the card in My Characters.
• Import card — bring in Character Cards (PNG or JSON). Importing a name you already have asks before replacing.
• Export card — download the card back as a PNG with the data inside.

Nuctor writes the card with you. Describe the character, drop a reference image, and it fills the fields section by section. It keeps a history per card, so you can come back and keep working.

From a card, press Chat. If the persona picker is on, you choose who you are first. The scene opens with the greeting already on the page.

Memories are facts the scene carries forward. The app collects them from the conversation and pins the important ones, so a long story does not fall out of context.

• Pin a fact by hand with Pin a fact — it always enters the prompt.
• Unpinned memories are capped, newest first, so they never crowd out the scene.

Scene settings has four tabs: Model, Characters, Memories and Actions. It only changes this conversation — the history stays.

• Model — switch the model for this chat only.
• Characters — bring an extra character into the scene with AI.
• Memories — read, pin and remove facts.
• Actions — delete messages from the history and the AI context.

Actions → Delete messages lets you pick lines and remove them for good. They leave the conversation and stop being sent to the model. This cannot be undone.

The persona has a name and a highlight. The highlight is what the model actually reads — describe who you are, how you look, what you want in the scene.

1. Open Personas and press the new persona button.
2. Give it a name and write the highlight.
3. Save. It is now available when you start a scene.

You can keep several personas and pick one each time — or turn off the picker in Settings to always use the last one.

The house model is one model for every plan. It needs no setup and spends credits from your plan. What changes between plans is how many credits you get and how big the context window is.

Bring your own OpenAI-compatible provider — OpenRouter, Groq, Together, or a private endpoint. Your key is stored server-side and never shown again.

• The model and the context window are the model's, not the plan's.
• You can add a primary and a fallback model, and per-provider prompt instructions.
• BYOK is the Author plan — and the Author is free.

Answer with a model running on your own machine, through LM Studio. Nothing leaves your network and no credit is spent. It needs the companion app — see the next topic.

• Run the model on your own hardware — your GPU, your memory, your rules.
• Nothing leaves your network. The prompt goes to your LM Studio, not to a server.
• No house credits are spent: the inference is yours.

1. Open Companion on the site.
2. Download the installer (.exe) and run it — or, from source, run the command shown under the download button.
3. Open LM Studio, go to Developer, and Start Server (port 1234). Load a model.
4. Open the companion app. Its window is where you pair and watch the work.

The app needs LM Studio open with the local server on and a model loaded. If the server is off, the window shows it and no request can be served.

1. On the site: Companion → Local app → Generate pairing code.
2. Copy the six-character code.
3. In the app window, type the code and press Connect.

The window is the control panel. It shows the connection state, the counters of every call, and the log.

• Status — online / offline / waiting, and which model is loaded right now.
• Counters — calls, first token, tokens per second, duration, and the prompt/response/reasoning split.
• Context — the window of the model loaded now, read from LM Studio.
• VRAM — how much of your GPU memory the loaded model is using.
• Code — type the pairing code here and press Connect. Disconnect ends the session.
• Site — the address of the site it polls. LM Studio — the address of your local server.
• Log — one line per event, so a failure says what happened.

Back on the site, the Companion page shows the local model and the state of the app. It is where you generate the code and where you turn the local model on for chat.

• Generate pairing code — creates the six-character code for the app.
• In your LM Studio — the models the app found loaded, with a loaded mark.
• Answer with my local model — the switch. When it is on, chat runs on your machine.

1. Pair the app (above) and confirm it says online.
2. On the Companion page, turn on Answer with my local model.
3. Open a scene. Scene settings → Model shows Local AI with LM Studio — pick it for this chat.
4. Send a line. The app long-polls the site, delivers the prompt to LM Studio, and streams the answer back.

Credits are spent only when the reply comes from the house. Your own key and the local model spend nothing. Each paid plan is cumulative: it has everything the one below it has, plus what it adds.

The Author is the way in: it is free, it is where a new account wakes up, and it is the only plan with your own key and the companion.

• Credits are the meter for house inference. They refill with the plan cycle.
• Context is the ceiling when the reply comes from house credits — never a gift: a model that only holds 8k stays 8k on a bigger plan.
• Reply tokens cap how long a single answer can be. Longer scenes write longer lines.
• Reasoning is on by default from World up — and you can always turn it off.

1. Open Plans and pick a card.
2. Choose the currency — the number you see is exactly what Stripe will charge.
3. Continue to payment. Credits land as soon as Stripe confirms.

The Author is free: the button takes the plan directly, with no card. Manage billing opens the Stripe portal for a paid plan.

Pick the language in Settings or from the selector in the top bar. It is saved on the device first, then on your account. The interface and these docs both change with it.

• Temperature and Top-P decide how adventurous the writing is.
• Max reply tokens is the ceiling per answer — the plan caps how high it can go.
• Thinking / reasoning is off by default; turn it on to let the model reason before writing.

With the house model or BYOK, the text goes to the provider that answers. With the local model, it goes to your LM Studio and stays on your network. Nothing is sent anywhere you did not point it at.

Credits are spent only by house inference. If a reply cost credits, that scene was answering on the house model. Switch the model in Scene settings to your key or the local model to stop spending them.

Yes. Scene settings → Model changes the model for that conversation only. The history stays exactly as it is.

Yes. Every card exports as a PNG with the data inside, and imports back the same way. The card stays with you.