An open-source front end runs companion chat off-platform

What happened

SillyTavern is an open-source application that provides a chat interface for role-play style conversation with a language model, without being a model itself. As retrieved on 16 September 2026, its repository describes it as "a locally installed user interface that allows you to interact with text generation LLMs, image generation engines, and TTS voice models," distributed under the GNU Affero General Public Licence. The project's documentation site adds that it functions purely as a front end, connecting to a wide range of local and hosted backends, including OpenAI-compatible APIs, KoboldAI, Tabby, NovelAI and Horde.

What the documents show

Both documents describe a "character card" system: prompts, greetings and example dialogue bundled into a shareable file that defines a role-play character, comparable in structure to the persona-and-example-dialogue approach used by hosted companion platforms, but stored as a portable file rather than as a record inside a company's account system. The repository states plainly, in a section describing its project vision, that the developers "do not provide any online or hosted services, nor programmatically track any user data," a statement about the software's own architecture rather than about any backend a user chooses to connect it to.

The mechanism

The mechanism worth naming here is the separation of the interface from the model. A hosted companion app typically controls the account, the stored chat history, the moderation layer and the underlying model all within one company's infrastructure. SillyTavern deliberately breaks that bundle: the interface runs on the user's own machine, character definitions are local files a user can edit or share, and the language model doing the actual generation is whatever backend, local or remote, the user configures. Because the project itself sets no content policy and runs no server that sees the conversation, moderation and data handling become properties of whichever backend is plugged in, not of SillyTavern itself.

What it leaves open

The documents describe the software's architecture and licence, not the behaviour or safety practices of any specific backend a user might connect to it, which can range from a filtered commercial API to an unmoderated locally-run model. Neither document makes claims about age verification, content restrictions, or what happens to data once it reaches a third-party API the user has configured, since those are outside SillyTavern's own code.

• Which backend is actually generating replies, and what content policy, if any, does that backend apply?
• Where is chat history stored, and who other than the user's own machine or chosen API can see it?
• Does a local, unmoderated setup remove age-assurance or crisis-response features present in a hosted app?

A front end like this does not add or remove safety features on its own; it relocates the decision about moderation and memory from a single company to whatever combination of software a user assembles.