Cardwright

Current version: 1.1.0

A small standalone character card editor (and lorebook authoring tool) focused on one workflow:

Additional documentation:

• User Guide

1. Start a blank card with New Card (lands on the Concept tab), or load an existing one. Your work is auto-saved to the browser and restored on the next visit.
2. In the Concept tab, describe your character idea and click Generate card from concept - the AI drafts every field for you to review and apply. The concept is saved and fed into all later AI edits.
3. Edit the card fields directly. (Or skip the concept and load an existing card from .json or a metadata-bearing .png.)
4. Edit the embedded lorebook (character_book): add/remove entries, keywords, secondary keys, position, order, and constant/triggered settings.
5. Run a local, offline Audit (no AI call) that checks required fields, token budgets, placeholder text, {{user}} impersonation, and lorebook health, scoring the card out of 100. Field-level issues can be sent to AI as reviewable fix drafts.
6. Ask LM Studio, oMLX, Ollama, OpenAI, or another OpenAI-compatible model to revise one selected field, improve a lorebook entry, or audit the card.
7. Set an avatar image and crop it to the 2:3 portrait ratio used by character cards.
8. Export the edited card - lorebook included - as JSON, or as a PNG with the card embedded in a chara tEXt chunk (re-imports here and in SillyTavern).

Avatar & PNG export

• Set / Crop Image (sidebar) opens any image in a 2:3 crop dialog. When you load a .png card its image becomes the avatar automatically.
• Export PNG (top bar) writes the current card (fields + lorebook) into the avatar PNG. The result loads straight back into this tool or SillyTavern.

The editor is split into four tabs:

• Concept - your idea/brief and one-click card generation.
• Card - the character fields, with live token estimates per field.
• Lorebook - the embedded World Info entries.
• Audit - the offline quality report; each field-level issue links straight to the field it concerns and can generate an AI fix draft.

Token counts are estimates (~4 characters per token) since no model tokenizer runs offline.

Run

Windows:

start-windows.bat

macOS/Linux:

./start-macos.sh

Manual:

node server.mjs

or:

npm start

Then open:

http://127.0.0.1:8787

You can override the port with PORT=..., but the default is 8787.

AI provider setup

Open the settings button (◎) in the top bar, choose a provider, then click Detect Model or Test.

Supported provider presets:

• LM Studio: http://127.0.0.1:1234/v1
• oMLX: http://127.0.0.1:8000/v1 (API key can be required by oMLX)
• Ollama: http://127.0.0.1:11434/v1
• OpenAI: https://api.openai.com/v1 (API key required)
• Custom: any OpenAI-compatible /v1 endpoint

For LM Studio:

1. Load a chat/instruct model.
2. Start the local server.
3. Choose LM Studio and click Detect Model.

The API key can stay blank for LM Studio. The default model value is local-model, but the app automatically asks LM Studio for /v1/models and fills the model field with the detected chat model.

For oMLX, start its OpenAI-compatible server, choose oMLX, and paste the oMLX API key if your oMLX server requires one.

Environment setup

You can either:

• enter an API key in the app's Settings panel for hosted providers, or
• set environment variables before starting the server. The app reads OPENAI_PROVIDER, OPENAI_BASE_URL, and OPENAI_MODEL as its defaults; OPENAI_API_KEY stays server-side and is used when the API key field is blank:

OPENAI_API_KEY=sk-... node server.mjs

Optional environment variables:

OPENAI_PROVIDER=lmstudio
OPENAI_BASE_URL=http://127.0.0.1:1234/v1
OPENAI_MODEL=local-model
PORT=8787

Any chat-completions-compatible provider should work if its base URL follows the /v1/chat/completions shape.

Audit and shutdown

The Audit tab includes a local offline audit, a full-card AI audit, and a separate lorebook AI audit. AI audit results are kept in the browser session for the current card.

To stop the server, use the Stop Server button in Settings. You can also press Ctrl+C in the terminal where node server.mjs is running.