The verdict
SillyTavern's official title is 'LLM Frontend for Power Users' — an accurate description of both what it is and who it's for. It is AGPL-3.0 open-source, runs locally via Node.js (or Docker on any OS), and connects to a longer backend list than any other frontend in the category: KoboldAI, Oobabooga (text-gen-webui), AI Horde, OpenAI API, Anthropic Claude API, NovelAI, Cohere, OpenRouter, and any custom OpenAI-compatible endpoint. The architecture is the point: zero built-in model, zero built-in censorship. You choose the model; you own the content policy entirely. Character cards use the V2 PNG/JSON standard — the same format as Chub AI, making 100,000+ Chub-hosted cards directly importable. Character definitions embed in PNG EXIF metadata, so a card is simultaneously a shareable image. Lorebooks (World Info) inject context entries when trigger words appear, maintaining long-session consistency. Author's Note injects static text at a fixed context depth. Group chat runs multiple characters in one session simultaneously. Extensions add image generation (via Stable Diffusion A1111 or ComfyUI), TTS speech synthesis, conversation summarization, and more. The app itself costs nothing; model cost is separate — AI Horde provides free (slow) distributed compute as a starting point. Not for beginners: 'install Node.js, clone the repo, configure a backend' is the literal entry process. For anything plug-and-play, Backyard AI (cloud inference) or SpicyChat are the practical alternatives.
The gold standard for model flexibility and control. For cloud inference without setup, Backyard AI; to browse character cards, Chub AI.
Pros
- Widest backend list in the category: KoboldAI, AI Horde, OpenAI, Claude, NovelAI, OpenRouter + custom endpoints
- V2 PNG/JSON character card cross-compatibility with Chub AI — 100,000+ cards importable
- Extensions: image gen (SD A1111/ComfyUI), TTS, summarization, group chat — all community-built
Cons
- Node.js self-install required — not plug-and-play; real setup barrier for new users
- You supply and pay for the model (AI Horde is free but slow; good models cost API fees)
- Extension quality and support vary — community-maintained, not officially supported
SillyTavern FAQ
Is SillyTavern free? Yes — the app is AGPL-3.0 open-source and free forever. You pay separately for any model API (OpenAI, Claude, OpenRouter) — or run local models and AI Horde for free.
Is SillyTavern hard to install? It requires installing Node.js and running a terminal command to start the server locally. Docker is also available. It's well-documented but not beginner-friendly — Backyard AI is the practical alternative if you want cloud inference without setup.
What backends does SillyTavern support? KoboldAI, Oobabooga (text-gen-webui), AI Horde, OpenAI API, Anthropic Claude API, NovelAI, Cohere, OpenRouter, and any custom OpenAI-compatible endpoint — the widest list in the category.
What are SillyTavern character cards? PNG or JSON files using the V2 CharacterCard standard — the same format as Chub AI, so cards are interoperable between the two platforms. Character definitions are stored in PNG EXIF metadata, making the card a shareable image file.
Does SillyTavern work on mobile? The server runs on Windows, Mac, and Linux — or Docker on any OS. Mobile access is possible by connecting your phone's browser to a locally running server, but there are no official native iOS or Android apps.
Scored on the public six-point method, assessed against documented features, pricing and real user reports (plus hands-on use where we have account access) · last updated September 2026.