TavernAI is a free, open-source AI roleplay chat platform – this repository is TavernAI 2 – built around an unusually deep prompt system, branching stories, and a strict privacy-first design where everything stays on your machine.
TavernAI 2 is a downloadable app for Windows and Linux (with a one-line server install for VPS hosting) that turns AI chat into something closer to a story engine. Scenes can branch at any message, new characters can join the cast mid-story, and several replies can be generated at once. Its signature feature is the Prompt Manager: the entire prompt is a structured, inspectable system of folders, activation rules, roles and token counters, and you can preview the exact assembled prompt – every source, token and attachment – before it is sent. Characters are self-contained cards carrying their own prompt trees, and MultiChat gives each character an independent timeline whose events merge when their stories meet. You can run GGUF models locally through CPU, CUDA or Vulkan backends, or connect to more than a dozen API providers, and even route image generation through a different provider than the chat model. There is no telemetry, no crash reporting and no background update checks. This new repository for version 2 was created in June 2026 and is actively released (v2.5.2 at the time of writing), with a small but growing GitHub audience of about 96 stars.
TavernAI is free to download and use. You bring your own local models or your own API keys for hosted providers, so any costs are only those of the models you choose.
TavernAI is for roleplay and interactive-fiction enthusiasts who want maximum control over prompts and characters, and who prefer a private, local-first app over a hosted chat service.