CharacterVault 1.0.0
First stable release. CharacterVault is a browser-only app for creating, editing, and organizing SillyTavern-compatible character cards (V2/V3). Data stays in your browser via IndexedDB. No account is required for core use.
Try it: https://vault.charactervault.app
Docs: https://vault.charactervault.app/docs/
Highlights
Fresh UI
- Full visual refresh across the vault, editor, settings, and AI Creation Studio
- Softer palette, cleaner cards, and consistent semantic theme tokens (including a VS Code Dark+ style dark theme)
- Settings pickers and sheets feel smoother, especially on phones
- Layout is clearer without changing the core workflow
AI Context and Orion
- AI Context panel streamlined: pin card sections from one list, selected chips, compact token usage
- Context no longer auto-follows the open editor tab. You choose what Orion and the AI toolbar receive
- Pins save and restore across sessions; Clear all truly empties context
- Optional chip to include the section you are currently editing
- Ask Orion refreshed: clearer empty state, live context status strip, cleaner composer
- Response stats (TTFT, speed, model) on hover or tap, without clipping long model names
- Leaner memory use while chatting: context is resolved only when you send, and streaming buffers are freed after each reply
Editor AI
- Ghost preview: Enhance, Rephrase, Custom, and polish tools show suggestions as soft ghost text on your selection
- Accept / Reject from the strip,
Ctrl+Enter/Cmd+Enter, or Escape - Thinking stays collapsed unless you open it
- Typing is paused during suggestions; saves wait until Accept, Reject, or cancel
- Payload preview (
{ }on the toolbar): inspect the full request body before anything is sent, switch tools, copy, no network call
Per-tool models
- Assign a different model (and provider / base URL) to each toolbar tool under Settings → Prompts
- Unmapped tools keep using the global AI Config model
- Custom prompt mapping also drives Lorebook AI trigger-key generation
- Orion and AI Creation Studio still use the main model from AI Config
NanoGPT and reasoning
- Sign in with NanoGPT (PKCE) from AI Config, with models fetched after sign-in
- Account card: balance, subscription status, and weekly usage when available
- Expanded reasoning effort ladder (Minimal through Max / Extra high), with request repair when a provider rejects a value
- Sampler context length up to 1M tokens, plus smarter packing of large lorebooks and greeting lists
Vault and library
- Backup the whole library as a ZIP
- Bulk import and drag-and-drop PNG/JSON
- Sort by name or recent; search by name and tags
- Active / total token estimates on cards
- Quick PNG or JSON export from card actions
- Streamed exports and lighter memory behavior for large collections
Mobile
- Greetings and Lorebook use the full screen instead of a short half panel
- Model and provider pickers as searchable bottom sheets
- Improved touch targets, list cards, and panel chrome
Other capabilities (stable platform)
- Full V2/V3 card editing: description, personality, scenario, greetings, lorebook, creator notes, extensions, and more
- Customizable section tabs (show, hide, reorder)
- In-editor spellcheck (English Hunspell, personal dictionary, ignore list)
- Snapshots and section or full-card rollback
- AI Creation Studio for generating cards from concepts or tags
- OpenAI-compatible providers plus NanoGPT; local endpoints welcome
- Import / export PNG and JSON compatible with SillyTavern-style frontends
Notes and known limits
- Orion and AI Creation Studio use the global AI Config model (not per-tool routing)
- Temperature, streaming, and thinking settings remain shared
- Clear AI Settings keeps per-tool model choices and custom prompt text
- Not every model supports every reasoning effort level; CharacterVault maps to the closest available
- Payload preview is best-effort; some optional parameters may still be dropped if a provider rejects them
- NanoGPT account weekly usage / subscription details on self-hosted production may need the optional usage proxy (see docs)
- NanoGPT PKCE sign-in needs popups allowed; reliability on some mobile browsers varies
Upgrade
This is a client-only web app. Opening the hosted site or rebuilding locally is enough; there is no server migration.
bash
git clone https://github.com/spaceman2408/CharacterVault
cd CharacterVault
npm install
npm run devLocal data in IndexedDB from earlier builds should continue to work. If anything looks off after a major UI change, try a hard refresh.
License
GPL-3.0-only
Thanks for using CharacterVault.