A lightweight, powerful, single-file AI character chat app
with full SillyTavern compatibility, advanced memory, and Group Chat support.
Open one HTML file and enjoy professional role-playing with multiple characters. No backend, no installation required.
Recommended (Faster & More Stable in China):
- Cloudflare Pages: https://opentavern.pages.dev
Backup:
- GitHub Pages: https://hajimides.github.io/OpenTavern
OpenTavern continues to evolve with cutting-edge browser storage technology. The latest version introduces OPFS (Origin Private File System) as the default storage backend โ delivering significantly higher capacity, better performance, and long-term reliability compared to IndexedDB or localStorage.
- Zero Setup โ Single HTML file. Double-click to run. Works offline after first load.
- Full SillyTavern Compatibility โ Import/export
chara_card_v3character cards. Complete World Info (Lorebook) implementation. - Advanced Memory System โ Automatic + manual summarization with history, edit, delete, and custom range support.
- Separate Summarization API โ Use a different (cheaper / uncensored) model for long-term memory compression.
- Modern Glassmorphism UI โ Beautiful dark theme with animated WebGL background, smooth animations, and mobile-first design.
- OPFS Storage (New) โ Origin Private File System for massive capacity (often hundreds of MB to GB), superior performance, and future-proofing. Real-time usage statistics included.
- Full Transparency โ View the exact prompt sent to the API at any time.
- Three Languages โ ็ฎไฝไธญๆ / ็น้ซไธญๆ / English (full UI translation with instant switching).
- Privacy First โ Everything stays in your browser. Nothing is sent except to your chosen API provider.
| Main Chat Interface | Character Card & Sidebar |
|---|---|
 |
 |
| World Book Editor | Settings Panel |
|---|---|
 |
 |
| Mobile View | Summary Manager |
|---|---|
 |
 |
Note: Place your screenshots in a
screenshots/folder for the images above to display correctly on GitHub.
- Download the latest
index.html(orOpenTavern.html) - Open it in any modern browser (Chrome / Edge / Firefox / Safari recommended)
- Click the โ๏ธ Settings button (top right or sidebar)
- Configure your API Endpoint and API Key (one-click preset for DeepSeek available)
- (Optional) Import a SillyTavern character card
- Start chatting!
Recommended first-time setup:
- Use DeepSeek (fast, cheap, excellent Chinese/English performance)
- Set Context Length to 20โ30
- Enable Auto Summarize with threshold ~12โ18
- Click New Chat
- In the character picker, long-press or use multi-select to choose multiple characters
- Start chatting
- Click the @ button to the left of the input box
- Select a character from the popup panel (beautiful glassmorphism design with blur)
- Or manually type
@่ง่ฒๅin the input
The selected character will be the one who replies this turn, and their full character card will be injected into the prompt.
If you send a message without @ anyone, the last character who spoke will automatically reply (great for natural back-and-forth).
- Only the mentioned character's detailed information is sent (efficient context usage)
- You can still have rich multi-character scenes by @ switching between them
- Works great on mobile
- Import any
chara_card_v3JSON from SillyTavern - Built-in full editor (name, description, personality, scenario, first message, example dialogues, system prompt, post-history instructions, creator notes, tags)
- One-click export back to SillyTavern format
- Automatic first message injection when starting a new chat
- New Character Picker when creating fresh conversations
Complete SillyTavern implementation with modern UI:
- Searchable entry list
- Rich editor: Memo, Primary Keys (supports regex
/pattern/i), Secondary Keys, Selective Logic (AND ANY / AND ALL / NOT ANY / NOT ALL) - Strategy: Constant (always active) or Keyword-triggered
- Probability, Insertion Order, Group scoring, Timed effects (Sticky / Cooldown / Delay)
- Token Budget Control โ Limit total tokens injected per generation (prevents context overflow)
- Full mobile-optimized editor with advanced filters
- Auto Summarize: Automatically compresses old messages when threshold is reached (configurable, with safety warnings)
- Summary Manager (dedicated modal):
- Two modes: Recent Messages or Custom Range (startโend turn)
- Live preview before generating
- Summary History list with edit / delete
- Max word limit per summary
- Separate Summarization API: Configure a completely different endpoint/model/key for summarization (ideal for NSFW content or cost optimization)
- Summaries are injected as system messages and can be manually managed
- Real-time streaming responses
- Stop generation button
- Message actions: Copy, Regenerate (AI messages), Delete
- Visual token/context usage bar with color warnings
- Full prompt viewer modal (see exactly what is sent to the API)
- Dialogue highlighting for Chinese quotes ใใ โโ
- Responsive input area with auto-resize
- Export single conversation or all conversations (JSON)
- Import previously exported chats
- OPFS (Origin Private File System) โ Default backend offering:
- Much higher capacity (often 100MBโseveral GB)
- Better performance than IndexedDB
- Private to the origin (more secure)
- Real-time storage usage display in Settings (used MB / quota with progress bar)
- Automatic migration handling
- Backup reminder system
- API: Endpoint, Key (with optional localStorage save + security warning), Model
- Generation params: Temperature, Top P, Max Tokens, Context Length
- User name (replaces
{{user}}) - Custom global system prompt (overrides character prompt)
- World Info token budget + enable toggle
- Summarization threshold, max words, separate API settings
- Language switcher (affects entire UI instantly)
- Storage backend status and usage statistics (now prominently shows OPFS)
- Create conversations with multiple characters at once.
- Use the @ button (left of input) to open a beautiful character selection panel.
- Explicitly @ a character (e.g.
@ๆฎๆๅจor via panel) to make them reply this turn. - If you don't @ anyone, the last replying character will automatically continue the conversation.
- Full Switch-mode prompt injection โ only the mentioned character's complete card is sent.
- Clean, mobile-friendly UI with backdrop blur.
This provides precise control in multi-character scenes while keeping prompts efficient.
- Hamburger sidebar
- Fully optimized modals
- Touch-friendly design
- Excellent @ mention panel on phones
OpenTavern uses the standard OpenAI-compatible chat completions API (/v1/chat/completions).
- DeepSeek (recommended โ one-click preset, including
deepseek-v4-flash) - OpenAI (GPT-4o, GPT-4o-mini, o1, etc.)
- SiliconFlow, Together.ai, Fireworks, Groq
- Local servers: Ollama, LM Studio, llama.cpp (with OpenAI compatibility layer)
- Any provider that follows the OpenAI chat format
Pro Tip: In Settings you can configure a completely separate Summarization API. This is extremely useful when:
- You want to use a cheaper model for memory compression
- Your main model refuses NSFW content
- You want to keep long-term memory while using a different provider
- Frontend: Pure Vanilla JavaScript + Tailwind CSS (CDN) + extensive custom CSS
- Graphics: Custom WebGL shader for animated background + vignette overlay
- Storage: OPFS (Origin Private File System) โ primary backend (recommended for long-term use) + localStorage fallback
- i18n: Built-in translation system with instant language switching
- Streaming: Fetch + ReadableStream for real-time token streaming
- No external dependencies โ everything is self-contained in one HTML file
Q: Do I really need nothing installed?
A: Correct. Just open the HTML file in a modern browser.
Q: How much data can I store now?
A: With OPFS, capacity is significantly higher than before โ often hundreds of MB to several GB depending on your browser and device. The app shows real-time usage with a progress bar.
Q: Why OPFS instead of IndexedDB?
A: OPFS offers better performance, larger practical limits, and is designed specifically for private, high-capacity file storage in the browser. It is now the recommended long-term backend.
Q: Can I use it completely offline?
A: Yes โ after the first load and API configuration, all chats, characters, and World Books work offline.
Q: Is NSFW content supported?
A: Yes. Use a permissive model and consider the separate summarization API to avoid refusals on long contexts.
Q: What happens if I update the HTML file?
A: Your data remains safe in browser storage (OPFS). Just replace the file.
Q: Why is there a separate summarization API?
A: Long conversations can become expensive or hit content filters. A dedicated summarization model (often cheaper or uncensored) keeps memory efficient without affecting your main chat model.
Q: How do I report bugs or request features?
A: Open an issue on the GitHub repository.
- Latest: Switched primary storage to OPFS for superior capacity and performance. Added New Chat Character Picker modal. Refined summary manager, mobile layouts, and World Book editor. Default summarization model updated to
deepseek-v4-flash. - Previous: IndexedDB storage, Summary Manager with history & custom range, separate summarization API, World Info token budget, full prompt viewer, major mobile UI fixes.
Contributions are very welcome!
High-value areas:
- Additional language translations
- More World Info features (timed effects, character filters, etc.)
- UI/UX improvements and accessibility
- Performance optimizations for very long chats
- Better error handling and user guidance
- Documentation improvements
Please open issues or submit pull requests on GitHub.
MIT License โ Free for personal and commercial use.
- GitHub Issues: Report bugs, suggest features, ask questions
- The project is actively maintained
OpenTavern โ Powerful character chat, made simple.
Now powered by OPFS for the best possible local experience.
Last updated: 2026