SillyTavern

SillyTavern is a locally installed chat interface for people who want full control over how a language model is prompted. It does no inference itself: you connect it to a backend, whether AI Horde for instant use, an OpenAI-compatible API, KoboldAI, Tabby or one of many other local and cloud providers, and SillyTavern manages everything that goes into the prompt.

Conversations are built around character cards, bundles of prompts that define a persona, a fictional character or an abstract scenario and keep its behaviour consistent across a long chat. World Info, also called lorebooks, injects background lore into the prompt only when relevant keywords come up, and group chats put several characters in one shared conversation. Beyond text, it can drive image generation engines and text-to-speech voices, and its settings expose instruct templates, sampler parameters and prompt formatting in detail. The project says plainly that the learning curve is steep and part of the appeal.

SillyTavern began in February 2023 as a fork of TavernAI 1.2.8 and has since had more than 300 contributors. It is licensed under AGPL-3.0, runs on Windows, Linux, macOS, Android and Docker on anything that runs Node.js 20 or newer, and a stable release branch is updated roughly monthly.