Repository navigation
Releases
Cardwright v1.3.0
Cardwright 1.3.0 adds source-text concept drafts and standalone Electron desktop downloads.
Create a concept from source text
- Paste wiki text or character notes in Concept → Create concept from source text.
- The existing selected AI provider prepares a factual, editable concept draft. Missing facts are marked as unknown; source text is reference data, not instructions.
- Only Apply concept draft replaces the current concept. Card fields, greetings, lorebooks and other metadata stay unchanged.
- Newer manual edits and card replacement are protected from late responses. Source text and unfinished drafts survive local autosave and stay outside JSON/PNG exports.
- This feature accepts pasted text only, up to 24,000 characters. It does not fetch webpages or silently shorten overlong source text.
Desktop downloads
- macOS Apple Silicon and Intel, Windows x64, and Linux x64, with Electron 45.2.3.
- The same editor and local AI-provider server in a standalone window; no separate Node.js installation is needed for downloads.
- Standard Edit/View/Window menus, remembered normal window size, guarded external links and native export dialogs.
- A dedicated Cardwright app icon, browser favicon and GitHub README badges.
- Automatic daily stable-release notifications with an opt-out and Help → Check for Updates. Downloading and installing remain your choice.
- macOS release downloads use Developer ID signing and Apple notarization. Windows downloads are not Authenticode-signed.
Data and compatibility
- Desktop autosave uses its own local Chromium profile. Browser autosave is not automatically migrated. Transfer cards as JSON/PNG and persona libraries with Export/Import Presets.
- The server stays on loopback and retains its Host/origin and API-key protections. A matching Cardwright server may be reused; an unrelated or older server is rejected and never stopped on desktop Quit.
- Source/browser launch scripts remain available.
Every download is extracted and tested on its target OS before publication, including native JSON export, card-data preservation, source-draft review/apply and persistence after restarting. SHA-256 checksums accompany the downloads. AI tests use a local mock provider.
Extract your platform archive and open Cardwright.app (macOS), Cardwright.exe (Windows), or Cardwright (Linux). Keep Windows/Linux extracted files together. See the User Guide for details.
The publisher-maintained Homebrew tap installs the matching notarized Mac app:
brew install --cask cyberdeliaAI/apps/cardwrightHomebrew becomes available after the tap's verified version and SHA-256 values have been updated for this release.
Cardwright v1.2.0
Cardwright 1.2.0 adds reusable player personas, configurable role placeholders, and editable avatar prompts while preserving the existing character-card workflow.
New features
- Generate an avatar prompt from the current card using the selected AI provider, edit it, and copy it into your image generator.
- Save reusable User Persona presets with Name and Description, plus validated JSON backup/import. Persona context is used only for card generation; the default is None.
- Choose
{{char}} / {{user}},{char} / {user}, or{character} / {user}in the sidebar. Prompts and audits share the same role definitions. Existing card text is not automatically converted. - Persona selection, placeholder preference, and avatar prompts are saved locally with the current card and stay outside character-card exports.
Fixes
- Preserve concept-generated text when navigating, auditing, saving, or exporting.
- Keep legacy lorebook entries and metadata when editing or adding entries.
- Preserve alternate greetings, including empty messages, whitespace, line endings, and embedded separators during import/export.
- Bind field drafts, audit fixes, and lorebook AI edits to their original card and target. Navigation, deletion, and late responses cannot redirect those edits to another target; newer manual edits are protected.
- Preserve explicit local model choices. Late model detection and connection-test results cannot overwrite newer settings.
Local server and API-key changes
- Switching provider or changing the base URL clears the previous API key. Enter the key for the new endpoint if required.
OPENAI_API_KEYis used only for the server's configured endpoint. For OpenAI, setOPENAI_PROVIDER=openaiwith the key.- The server accepts local clients, validates Host and browser origin, and serves only public app assets. Hidden files and repository internals are not served.
- Provider URLs must use HTTP or HTTPS without embedded credentials, query strings, or fragments. Configure the final URL directly; provider redirects are rejected.
- Browser API requests require JSON. Keys entered in Settings remain in this browser's local storage until cleared or the provider/endpoint changes.
Validation
- All 124 regression tests pass, covering data preservation, JSON/PNG round-trips, persona backups, placeholder prompts and audit rules, model selection, and server access controls.
- New UI flows were checked in a real browser against a local mock provider. Server forwarding and access controls were also checked with real local HTTP requests.
- AI checks use mock responses; no live hosted-provider generation is required for the test suite.
Download and run
Download the source ZIP below and extract it. Start with start-windows.bat, ./start-macos.sh, or node server.mjs, then open http://127.0.0.1:8787.
See README and User Guide for setup and usage.