Inkling
A self-hosted, single-page AI roleplay chat client: characters, group chats, personas,
lorebooks, and a bunch of quality-of-life stuff, that runs entirely in your browser.
Everything lives in IndexedDB on your device; nothing is stored on a server unless you
explicitly set up sync.
There's no product pitch here. This is a personal project, built and iterated on for
one person's own use, made public because there was no reason not to. If you found it,
poke around, fork it, steal whatever's useful. No support is promised.
Features
- Characters & group chats: SillyTavern-compatible character cards (JSON and PNG
card import/export), alternate greetings, example dialogues, per-character avatars.
- Personas: switch who you are mid-conversation.
- Lorebook / Scripts: keyword-triggered world info injected into context, with
probability, depth, and grouping controls.
- AI Assist: write-for-me, enhance draft, continue, and impersonate, powered by
the same model you're chatting with.
- Multiple providers: the built-in Claude proxy (
api/anthropic.js), OpenRouter,
or any local/self-hosted OpenAI-compatible endpoint (Ollama, LM Studio, etc).
- Theming: a dozen built-in themes, plus a custom background image (animated GIFs
supported) and raw CSS injection for anything the built-in themes don't cover.
- Sync: back up and restore everything (characters, chats, personas, lorebook,
settings) via a private GitHub Gist, or your own Railway-hosted sync server. See
server/README.md for the Railway option.
- MCP: let Claude Code / Claude Desktop / Cursor create and edit characters from chat. The in-app
Connect Claude (MCP) guide (Sync → Railway, or Ctrl+K) generates the config; details in
mcp/README.md.
- PWA: installable, works offline for anything already loaded.
- In-app changelog: pops up after an update so you don't have to go digging
through commit history to remember what changed.
Architecture
There's no framework build step for development. index.html loads React and Babel
Standalone from a CDN and transpiles the src/*.jsx files live in the browser, so you
can just open index.html and start editing.
index.html - page shell + all CSS (themes, layout, components)
src/hmm-utils.jsx - storage (IndexedDB), AI calls, macros, sync, card import/export
src/hmm-app.jsx - root app: routing between chat/settings/modals, theme application
src/hmm-sidebar.jsx - character list sidebar
src/hmm-chat.jsx - chat view: messages, streaming, composer, group chat logic
src/hmm-modals.jsx - every modal (settings, character editor, import, sync, lorebook…)
api/anthropic.js - Vercel edge function: thin streaming proxy to api.anthropic.com
server/ - optional standalone sync server (Express + Postgres) for Railway
build.mjs - production build: esbuild-transpiles src/*.jsx into dist/, strips Babel
For production, npm run build runs esbuild once at deploy time instead of shipping
Babel and re-transpiling ~250KB of JSX in every visitor's browser. The dev workflow
(opening index.html directly) is unaffected.
Linux desktop app
Inkling also runs as a native Linux app (Electron). It bundles everything it needs, works
offline, and runs the api/ functions locally, so Claude, web search and image upload work
without any hosting. Your data lives in the app's own profile (~/.config/Inkling).
Download: grab the AppImage or .deb from the Releases page (built by the
Linux desktop build workflow — push a v* tag, or run it by hand from the Actions tab).
chmod +x Inkling-*.AppImage && ./Inkling-*.AppImage # AppImage
sudo apt install ./Inkling-*.deb # Debian / Ubuntu
Build it yourself:
npm install # repo root (web build)
cd desktop && npm install
npm start # build + launch
npm run dist # AppImage + deb + tar.gz into desktop/release/
It serves on 127.0.0.1:47831 (override with INKLING_PORT). Keep that port stable: saved
data is tied to it.
Running it
npm install
npm run build # writes dist/, what actually gets deployed
Or just open index.html directly in a browser for local dev (no build step needed).
Deployed on Vercel. vercel.json points the build command at
npm run build and serves dist/. api/anthropic.js runs as a Vercel Edge Function.
Data & privacy
Characters, chats, and settings are stored client-side in IndexedDB; nothing leaves
your device unless you configure a sync provider (Gist or Railway) or generate content
through your chosen AI provider. API keys and any custom jailbreak/system prompt are
explicitly excluded from both sync payloads and are never bundled into the exported
backup JSON's public fields, they stay local only.
Credits
Not affiliated with, endorsed by, or sponsored by Anthropic, SillyTavern, freeimage.host,
Vercel, or Railway, just a hobby project that uses their public APIs/services.
License
MIT: do whatever you want with it, no warranty, no liability. See
LICENSE for the legal text nobody's going to enforce.
