An open-source, dual-compatible local AI chat interface optimized for performance and ease of use. Local Tavern bridges the deep customization of SillyTavern with the self-contained simplicity of LM Studio, giving roleplay, model management, and local-first chat workflows a clean single-app home.
Local Tavern features Dynamic Dual-Compatibility Inference Routing.
When you load a GGUF model, the app first attempts to run it completely natively through llama-cpp-python. On high-end and modern systems, this keeps inference direct, fast, and self-contained.
If the machine uses a legacy CPU architecture and native loading triggers an AVX2/AVX mismatch, DLL initialization failure, or Illegal Instruction error, Local Tavern catches the failure safely instead of crashing. It then instantly initializes an embedded koboldcpp-oldpc.exe process in the background, headlessly, and routes generation through its OpenAI-compatible local API.
The result: modern systems get native inference, older systems still get a smooth local GGUF workflow, and the user does not have to manually launch a separate KoboldCPP window.
Place your favorite GGUF models in the root folder or storage/models/.
If you are using an older PC, make sure koboldcpp-oldpc.exe is in the project root directory.
Double-click install.bat once.
It automatically creates an isolated venv environment and installs the lightweight dependencies needed to run the app.
Double-click start.bat to launch.
The app boots the local Uvicorn server and opens exactly one tab in your default browser at http://127.0.0.1:7865.
- Streaming text generation via Server-Sent Events (SSE).
- Local model marketplace with Hugging Face search and background download progress tracking.
- Dynamic GGUF loading with native
llama-cpp-pythonfirst and legacy KoboldCPP worker fallback when needed. - Built-in prompt formatting cards for ChatML, Llama 3, and Mistral.
- Cloud API mode for OpenAI-compatible endpoints such as OpenRouter, Ollama, and KoboldCPP.
- Full character card persistence for names, descriptions, personalities, first messages, and avatars.
- Lorebook and World Info keyword scanning with contextual prompt injection.
- Conversation history persistence with clear chat, session restore, and export support.
- Local-first storage under
storage/for models, characters, lorebooks, chats, and extension data.
Local Tavern is released under the GNU Affero General Public License v3.0. Community contributions are welcome, and derivative networked versions must preserve the same open-source freedoms.