Provider Linking
Link your local characters to their online source for update checking, version history, and gallery downloads. Supported providers:
- ChubAI: Card updates, gallery downloads, full Git version history
- JanitorAI: Card linking and updates
- CharacterTavern: Card linking and updates
- Pygmalion: Card linking, updates, and gallery downloads
- Wyvern: Card linking, updates, and gallery downloads
How to link:
Click the Link indicator in the character detail sidebar, or right-click → Link to Provider. Search by name across all providers, or paste a direct URL.
Once linked:
- Check for Updates: Compare local vs. remote card fields with a diff preview
- Gallery Downloads: Download the character's provider gallery images
- Version History: Browse published version history and restore any version
- Quick Navigation: Jump to the provider page or preview in-app
Auto-Link
Bulk-link your entire library automatically:
- Open from More Options (⋮) → Auto-Link Characters
- Scans all unlinked characters by searching each provider's API by name
- Results are categorized: Confident (strong match, auto-selected), Uncertain (multiple matches, you pick), and No Match
- Review all matches before applying. Nothing is saved until you confirm
- The scan is resumable: stop and resume later, or close and reopen without losing progress
Live status display while the scan runs:
- The avatar and name of the character currently being scanned
- A status line describing the current step (querying providers, comparing matches, etc.)
- Progress counter (e.g.
147/892) and a progress fill bar showing scan completion
- Running totals updated in real time: confident, need review, and no match counts
- Once the scan finishes, the same counts double as tabs to filter the results list
Online Browsing
The Online tab lets you browse and import characters from providers:
- Switch between providers using the dropdown selector
- Search, sort, and filter by tags/categories. Each provider has its own options
- Toggle NSFW content visibility per provider
- Infinite scroll automatically loads more results as you scroll down (toggleable per provider in Settings)
- Cards already in your library are marked with an "In Library" badge
- Cards that look like a character you already have get a "Possible Match" badge, even across providers. It triggers on a matching name and creator, or on a distinctive name reposted under a different handle (a common case: the same card reuploaded to another site). The badge brightens with confidence (faint = name-only hunch, bright = near-certain). Tune how readily it appears with the Possible Match sensitivity slider in Settings → Online, or hide flagged cards with each provider's "Hide possible matches" filter. It is a hint, not a block; the import screen still runs a full content comparison to catch true duplicates the badge can't see from name alone.
- Click a card to preview the full character details and gallery images, then import with one click
- Set Collapse all browse preview sections by default in Settings → General → Character Details to start each preview with all sections collapsed except creator's notes (cleaner view for large cards)
- Some providers support login for additional features (favorites, following lists)
Configure provider order in Settings → Online. NSFW preference is remembered from each provider toggle.
If a provider fails to load, the error banner's Copy error report button produces a paste-ready summary for support requests; it includes settings only as on/off flags, never your tokens.
DataCat
The DataCat provider combines multiple data sources to give you the best coverage for JanitorAI characters. Each sort order pulls from a different source:
DataCat sort orders (Recent, plus Freshest / Score / Chat Count and more over 24h and 7d windows):
- Characters come from DataCat's own database, which mirrors JanitorAI
- Full character definitions are available (personality, scenario, first message, etc.)
- Tags use DataCat's own category system
- Creator names are always available
- One-click import with complete card data
- The search bar supports text search (matches character names, creators, and tags), plus DataCat or JanitorAI URLs and creator UUIDs for direct lookup. Typing a creator name jumps to their full catalog even with nothing loaded yet
- The Tags dropdown filters by any of DataCat's tags
JanitorAI (MeiliSearch) sort orders (Newest, Oldest, Most/Least Tokens, Relevant):
- Searches JanitorAI's full-text search index directly
- Covers all published characters on JanitorAI
- The search bar supports text search (searches character names and descriptions) as well as JanitorAI URLs for direct lookup
- Tags use JanitorAI's own tag system (Male, Female, Anime, etc.)
- Creator names can be blank on grid cards (JanitorAI's search index omits them for some characters); they fill in on the preview when the character is also on DataCat
- Characters not yet on DataCat can be extracted on demand
JanitorAI (Hampter) sort orders (Latest, Trending, Trending 24h, Popular, Relevance):
- Pulls from JanitorAI's curated trending/popular feed
- Only surfaces characters with significant engagement, not a complete index
- The search bar supports text search (filters within the trending/popular pool only) as well as JanitorAI URLs
- Tag filtering is not available in this mode
- Creator names are always available
Cloudflare and signing in are two separate things. Mixing them up is the usual reason a setup looks broken, so, plainly:
- Cloudflare is a cookie (
cf_clearance) that only a real browser can earn. It decides whether a request reaches janitorai.com at all.
- Your account is a header that Character Library already sends on its own. For these sorts it decides one thing: whether you get results past the first page.
Neither one fixes the other. Signing in will not get you past Cloudflare, and clearing Cloudflare will not sign you in. Here they are independent, and an account is optional. The JanitorAI provider is different: there both are required.
Two different tools, and they are not interchangeable. Which one you need depends only on what you are using:
- The JanitorAI provider uses a real browser, and nothing else. Character Library runs one for you (Settings → Online → JanitorAI), or you can point it at a Chrome DevTools endpoint you run yourself. The provider's hidden-definition recovery runs through it, and it needs the machine running Character Library to have a working GPU. The userscript cannot serve this provider.
- DataCat's five Hampter sorts (Latest, Trending, Trending 24h, Popular, Relevance) and JannyAI card definitions use the companion userscript (install it with Tampermonkey or Violentmonkey), and nothing else. It borrows the Cloudflare pass from the browser you are reading this in, so it works even when Character Library runs on a headless server.
What needs no setup at all: DataCat's own sorts and its MeiliSearch janny_* sorts, JannyAI browsing and search, and signing in. A JannyAI card will still import without the userscript, but with no personality and no greeting, which is not much of a character.
Cloudflare's pass expires on a fixed timer rather than on idleness, so it lapses often. The managed browser handles that itself; userscript v1.2.1+ renews it by briefly opening the site in a background tab. On mobile, a userscript-capable browser such as Firefox for Android works.
Across all modes: Pasting a JanitorAI character URL into the search bar will look up the character on DataCat. If it's found, the preview opens directly. If not, an extraction panel appears letting you request the character from DataCat's cloud service.
Extraction: When a character only exists on JanitorAI and hasn't been indexed by DataCat yet, the Import button changes to Extract. This sends a request to DataCat's cloud service, which visits the JanitorAI page and retrieves the full character definition. You can also paste a JanitorAI URL directly into the search bar to look up and extract specific characters.
DataCat browsing is proxied by the server, so it works without any browser extension.
Re-extract on update checks: Toggle in Settings → Online → DataCat to force DataCat to pull fresh data from JanitorAI before each update check. Slower, but avoids comparisons against stale cached data on DataCat's side.
Botbooru
Botbooru is a tag-driven character card catalog. SFW browsing and importing works without an account; logging in (Settings → Online → Botbooru, or the key button in the browse toolbar) unlocks NSFW, favorites, and follows, and applies your account's tag blacklist.
- NSFW is account-gated: the NSFW toggle requires login. Include NSFL posts via Settings → Online → Botbooru → Content (only applies while NSFW is on); NSFL cards carry their own badge, distinct from NSFW
- Curated sort + favorite tags: favorite tags boost matching cards in the Curated sort. Manage them in Settings → Online → Botbooru → Favorite Tags, or with the star icons in the browse Tags dropdown
- Weighted tag mode (experimental): an account-side switch that replaces the simple favorite-tags list with per-tag weights (-1000 to 1000) plus always-follow / always-block flags, and unlocks extra Curated orderings (Recent, Tag Score, Only Followed). While on, the site ignores the simple list, and the dropdown stars add +100 weight entries you can fine-tune in Settings
- New uploads only: a checkbox in the Features dropdown's Curated section (shown while a Curated sort is active) that excludes bumped/updated cards
- Sorting: beyond Latest, Curated, and Random, popularity sorts cover favorites, views, and downloads over day, week, month, and all-time windows
- Advanced Options (bottom of the Tags dropdown): minimum token count, optionally counting lorebook text, and an uploaded after/before date range
- Features dropdown: My Favorites (login required), Hide Owned Characters, Hide Possible Matches, Hide AI-generated, and (Curated sorts only) New Uploads Only
- Following: follow uploaders from the banner that appears when browsing their cards, then switch to the Following timeline, which has its own sort options
Username/password login is handled by the server. Without it, paste a token manually in the login dialog; a copied "Bearer " prefix is stripped automatically.
JanitorAI
First-party access to janitorai.com. Roughly half of all JanitorAI characters withhold their hidden definition (the field is simply absent from the API), and this provider recovers them at the source. DataCat and JannyAI can recover hidden definitions too, but DataCat has been progressively closing itself off, and JannyAI's copies come without alternate greetings or creator browsing, so this is the complete, first-party path. Beta, and off by default.
- Browse and search janitorai.com directly, with its own sort orders (Latest, Trending, Trending 24h, Popular, Relevance)
- Filter by tags, NSFW toggle, and browse a single creator's characters
- Following tab with a followed-creators manager, plus a Follow button on the creator banner
- Preview shows alternate greetings and the embedded lorebook; character linking and card updates work as they do elsewhere
It needs a real browser, and only a real browser. janitorai.com puts Cloudflare in front of everything, and passing that challenge earns a cf_clearance cookie that no server-side request can hold. On a normal desktop with Chrome or Edge installed, that means: Settings → Online → JanitorAI → Start now, then press Test, and you are done. The browser runs invisibly and stops itself when idle.
The Test button is the contract. Its checks measure what Cloudflare actually cares about, a real GPU above all: software rendering never clears the challenge and no flag changes that. Red checks mean your browser environment does not meet the spec yet; making it pass on servers, VMs, containers, phones, or ARM boards is power user territory and your own research (the README's JanitorAI section lists the exact requirements and the desktop endpoint runner for GPU-less hosts). Green checks with a misbehaving provider is a Character Library bug: please report that one.
Signing in is the second requirement, not an optional extra. The browser gets you to JanitorAI; an account gets you the content. Without one you get the first page of results and nothing else: no follows, and no character extraction, so no hidden definitions. It happens in Settings rather than in the browser, and only works once the browser above does, since JanitorAI's login captcha runs only on their own page.
What recovering a hidden definition does to your JanitorAI account. There is no API for it, so Character Library makes the site assemble the prompt and reads it back. For that one request it creates a throwaway persona with a random name, adds a temporary proxy preset pointed at an unroutable address so nothing is ever sent anywhere, opens one chat, and sends one message. The persona, the preset, the chat, and your original generation settings are all removed and restored afterwards. It runs only for characters whose definition is actually hidden, and only when you press Recover definition or import one.
The companion userscript has nothing to do with this provider and cannot substitute for the browser. It serves DataCat's Hampter sorts and JannyAI card definitions only.
JannyAI
A mirror of JanitorAI content. It needs no browser endpoint, which makes it lighter to set up than the JanitorAI provider, at the cost of alternate greetings and following. No account needed.
- Browse and search the JannyAI catalog; filter by tags, token count, and NSFW
- Character linking and card updates
- No gallery downloads, version history, or alternate greetings
Card definitions need the companion userscript. JannyAI now Cloudflare-gates its card pages, and the card page is where the character actually lives. Browsing and searching still work without it, but an import without the userscript gets you the name, avatar, tags, and the site blurb as creator's notes, with no description, no greeting, no scenario, and no example dialogue. That is a stub, not a character.
Install it exactly as for DataCat's Hampter sorts; one install covers both. If definitions stop loading even with it installed, your Cloudflare pass has expired: open jannyai.com in the same browser, let it load, then reopen the preview.
Saucepan
Browsing saucepan.ai needs no account, but importing does. Beta, and off by default.
- Sort by New, Trending, or Popular
- Filter by tags and NSFW, and by fandom, a dimension separate from tags covering source material and franchise
- Creator mode for browsing one creator's companions, ordered by message count or age
- Character linking and card updates
Definitions are extracted natively rather than downloaded as a card file, and that extraction is what needs the token. Some companions have a locked definition that is not publicly available; the preview says so, and extraction may return an incomplete body for those.
Everything goes through this app’s server, browsing included, because saucepan.ai sends no CORS headers at all. Log in under Settings → Online → Saucepan, or paste a bearer token from your own session. Opening a character that needs authentication turns the preview's import button into Configure Token, which takes you to the same place. The token lives in the server's memory and is pushed back automatically after a restart, so you should not need to log in again.
Version History
Track changes to your character cards with snapshots and diffs:
- Auto-backup: A snapshot is saved automatically before every edit, restore, or card update, so you can always undo
- Manual snapshots: Save the current state of any character at any time
- Diff preview: Side-by-side comparison for every card field with added/removed highlighting
- Remote versions: For linked characters, browse the full published version history and restore any version
How ChubAI remote versions work:
- "Chub Page" entry: the current state on ChubAI, using the metadata API (with linked lorebooks resolved from the V4 Git API)
- Git version history: each commit fetched from the V4 Git API (
card.json), including the full merged lorebook
Configure max auto-backup count in Settings → General → Version History.
Card Editing
Edit all character card fields directly in the library:
- Open a character → switch to the Edit tab → click Unlock Editing
- All fields are locked by default to prevent accidental changes
- Editable fields: name, description, personality, scenario, first message, alternate greetings, example messages, system prompt, post-history instructions, creator, creator notes, character version, tags, and the embedded lorebook
- Use the expand buttons for full-screen editing of long text fields
- Replace card image: with editing unlocked, click the large card image (or tap the avatar in the mobile header) to pick a new PNG. The pending image previews live; an X badge reverts before save. The auto-snapshot embeds the previous image, so version history can still show the old image after overwrite
- A snapshot is auto-saved before each edit so you can always undo
Lorebooks & Card Updates
When a character card includes a lorebook, it exists in two places:
- Embedded lorebook (inside the card): captured at import time, and not changed when you edit the standalone lorebook
- The lorebook itself (its own rows, managed in Lorebooks): the live working copy used in chats. All your edits go here
These two copies are independent and diverge after import.
During a card update check:
- The card's embedded lorebook is compared against the creator's latest version on the provider
- New, modified, and local-only entries are shown in the diff
- Your lorebook is never modified by the update checker
When you apply a lorebook update:
- The card's embedded lorebook is replaced with the latest version
- Local-only entries (in your card but not on the provider) are removed from the embedded copy
- Your World Info file is not touched
Safety check for local-only entries:
- Safe: also exists in your World Info file
- Warning: not found in your World Info file, will be permanently lost
TL;DR: Updates only change the card's embedded data. Your World Info file (the copy that matters for chats) is never modified.
Display Name Override
For characters linked to a provider, you can choose whether the library shows the card's own name or the provider listing name (the title used on the provider's page). The card data is never modified; this is purely a display preference.
- Open a character → click the name type toggle in the details header to switch between card name and listing name for that character
- Set the global default in Settings → General → Display name source (Card name or Listing name)
- Find characters with a listing name available using the Advanced Filters panel
Useful when a creator's listing title differs from the card's internal name field, e.g. "Ada (Cyberpunk Bartender)" on the listing vs. just "Ada" on the card.
Duplicate Detection
Find potential duplicates in your library:
- Open from More Options (⋮) → Find Duplicates
- Compares character name, creator, description, first message, personality, scenario, and creator's notes
- Rates each match with a similarity label (Possible Match up to Identical) plus a per-field score breakdown
- Review matches and delete unwanted duplicates
Adjust the minimum score threshold in Settings to control sensitivity.
Grid Card Thumbnails
The characters grid is served pre-resized images instead of full card PNGs, which cuts decode cost and bandwidth. The detail modal, gallery, and full-screen viewer always keep full resolution.
- Always on, nothing to configure. The server resizes on demand and caches each result on disk, so a size is generated once and then served straight from cache
- Each surface asks for the size it actually renders: 192px for chat rows, header thumbs and mini-cards, 512px for grid cards, 384px for gallery tiles
- Replacing a character image busts that character's cached URL, so the new picture appears immediately instead of after a hard refresh
Biggest win: mobile decode cost drops sharply on fast-scroll, and large libraries on slow connections feel snappier. On desktop with plenty of decode budget, the win is mostly bandwidth.
Gallery Viewer
Full-screen media viewer for character gallery images, GIFs, and videos:
- Open from right-click → View Gallery, or from the character detail modal
- Navigate with arrow keys (← →), click left/right halves of the image, or use the prev/next buttons
- Scroll wheel to zoom (up to 5×), drag to pan when zoomed, click or press
0 to reset zoom
- Thumbnail strip at the bottom for quick jumping between images
- Supports images, GIFs, and video (mp4/webm). Press
Escape to close
Media Localization
Download remote media (images, audio, video) referenced in character cards for offline access:
- Scans card text fields for remote URLs and downloads each file to the character's gallery folder
- Also downloads gallery images from linked providers
- Deduplication by file hash; won't re-download files you already have
- Hosts that block direct browser requests are fetched through the server automatically; nothing to configure
Enable "Replace remote media URLs" in Settings to auto-swap URLs when viewing. Use Bulk Localization in Settings → Media to process your entire library at once.
Background Downloads & Notifications
Let imports finish immediately and run media downloads quietly in a background queue:
- Set When an import has extra media to Download in the background in Settings → Media → Options
- While enabled, the download work from imports (embedded media, lorebook media, provider galleries, external galleries) is queued and processed one character at a time instead of blocking the import
- The notifications bell in the topbar tracks it: a bell when idle, a download icon while the queue is working, and a warning badge if a download failed. The button stays hidden when no section has anything to show
- Open the dropdown to watch the active job's progress bar, cancel it, retry or dismiss failed jobs, and clear finished ones
- The queue survives page reloads: pending downloads resume automatically on the next load, re-scanning the card and skipping files you already have
- Mobile: open it from the More menu → Notifications
External Gallery Extractors
Some character cards can link to galleries on external image hosts instead of embedding direct image URLs. Character Library walks these links, resolves the underlying images, and downloads them into the character's gallery folder.
Supported hosts:
- Civitai: posts and image collections (public out-of-the-box; private/hidden posts require an API key)
- Imgchest: public and password-protected posts (passwords are auto-detected from the card's creator notes / description via regex; password-protected posts are unlocked server-side)
- Mega.nz: public folder shares with multi-pass folder traversal
- PostImg, ImgBB, Imgbox: albums and individual images
- Pixiv: artwork pages (R-18 works need your Pixiv login in Settings → Media)
- Catbox: direct file URLs
- Dropbox: shared folders and individual file links
- Google Drive: shared folders and individual files
Extractors run automatically when you trigger Download Embedded Media from the Gallery tab or use Bulk Localization. No configuration is required for public content.
Civitai API key: Required only for private or hidden Civitai posts. Configure in Settings → Media → Civitai API Key. Generate at civitai.com/user/account. The key is stored per user on your server and pushed to the provider service when you save it, so it stays server-side at request time.
Overview
Create new characters from scratch or edit existing ones with AI assistance. Open via ⋮ menu → Create Character.
Each card field (description, personality, scenario, first message, etc.) has a wand icon button that opens the AI Studio panel for that field.
AI Studio
The AI Studio is a conversational assistant for generating and refining card fields:
- Multi-turn chat: Ask for rewrites, adjustments, or entirely new content. The full conversation history is sent with each request so the LLM has context
- Suggestion chips: Quick-start prompts tailored to each field (e.g. "Write a mysterious backstory", "Make them sarcastic and witty"). Click one to generate immediately
- Word target: Set a target word count for generations. The LLM will aim to match it
- Apply to card: When satisfied with the result, click Apply to Card to write it into the card field
- Undo / Redo: History stack for the current studio session. Use the toolbar buttons or
Ctrl+Z / Ctrl+Y (when not focused in a text input). Opening a different field starts a fresh history
Brainstorm Mode
When editing the Description field, a lightbulb icon appears in the AI Studio toolbar. Click it to enter Brainstorm Mode:
- The content editor hides and the conversation area expands to fill the panel
- The LLM becomes a collaborative design partner: it asks questions, suggests ideas, and helps you develop your character concept through back-and-forth dialogue
- Suggestion chips offer creative starting points like "a mysterious character with a hidden past" or "help me figure out what motivates this character"
- When you're happy with the concept, click Finalize. The LLM synthesizes the entire conversation into a polished character description and writes it into the content area
- Click the lightbulb again to exit brainstorm mode and return to normal editing
Brainstorm mode is great when you have a vague idea but aren't sure how to put it into words yet. Let the LLM interview you instead of staring at a blank page.
Highlight Revision
Select (highlight) any portion of text in the AI Studio content area to enter revision mode:
- The selection is locked and visually highlighted with a colored overlay
- A "Revising selection" badge appears above the input row
- Your next prompt targets only the highlighted section. The LLM rewrites just that segment and splices it back into the surrounding text
- Click the × on the badge to clear the selection and return to full-field mode
This is useful for tweaking a single paragraph without regenerating the entire field.
Custom System Prompts & Presets
Click the gear icon in the AI Studio toolbar to open field-specific settings:
- Custom system prompt: Override the built-in instruction for that field with your own. The placeholder shows the default prompt for reference
- Context checkboxes: Toggle which other card fields are included as context when generating. For example, uncheck "Personality" when generating a description if you want a fresh take
- Save as preset: Save your custom prompt as a named preset. Load or delete saved presets from the dropdown
Presets are saved per field and persist across sessions. A dot indicator appears on the gear icon when a field has active overrides (custom prompt or disabled context fields).
Connection Profiles
Add endpoints under Settings → AI. With more than one profile configured, a Profile dropdown appears in the sidebar under AI Assist, so a feature can run on a different model from the rest. API keys are stored server-side against your account and are never sent to the browser.
Import from Library
Click Import from Library to load an existing character into the creator form. This populates every field including tags, avatar, alternate greetings, and lorebook. You can then modify and save as a new character or overwrite the original.
Save Modes
- Create Character: Saves as a brand-new character in your library
- Save as Existing: Appears after importing from library. Opens a side-by-side diff review showing exactly what changed before overwriting the original card. An automatic snapshot is saved first so you can always undo