Self-Hosted vs Hosted Character Chat: Where the Data Controls Sit

Character chat, meaning a text conversation with one defined character, can run in two places. A hosted platform keeps your conversations on its own servers, so its privacy policy and account settings decide what happens to them. A self-hosted front end runs on a machine you control. Even then, if you connect a hosted model, your messages travel to that provider. Readers must be 18 or older, and this guide does not cover any content involving minors. Whichever model you choose, the official current terms of every party you depend on decide the outcome.

What the official SillyTavern page states

The SillyTavern homepage, captured as text on 16 October 2026, describes the project as an LLM frontend for power users. It links to GitHub, Discord and Reddit. The captured page does not state a license, a fee or a data-handling policy, so this guide does not state them either. A reader who wants those facts should open the project’s own repository and read the terms there.

Two models, two places for the data

With a hosted platform, the company holds the conversation text, so its policy answers most questions about storage, use and deletion. With a self-hosted front end, the software may keep chat logs and settings on your own device. The text you send to a model still leaves your device if that model runs on someone else’s servers. The control question therefore changes shape. It stops being only what the platform keeps, and becomes what the software stores locally and what each connected provider keeps.

What to check on a hosted platform

On a hosted platform, the questions are the ones covered in our checklist guide. Ask whether the policy describes training use, how you export and delete your data, how renewal and cancellation work, and what the age statement covers. The platform’s own policy is the document that answers them. Read it before you create an account, and record any question it leaves open.

What to check on a self-hosted front end

A self-hosted setup moves the questions to different places. Check the following before you depend on it:

• The license and fee terms, which sit in the project’s repository and were not verified here.
• Where chat logs and settings are stored on your device, and how to delete them.
• Which model provider you connect, and that provider’s data policy and retention terms.
• Whether any extension or add-on you install sends data to a third party.

Each item is a separate party with its own terms. Missing one of them leaves a gap in your picture of where the data goes.

Questions to answer before you choose

• If I close the front end, where do my chat logs remain, and who can delete them?
• If I change the connected model, do the data terms change with it?
• If the project updates, do my stored settings survive, and do I control that update?
• If I use a hosted model, does its policy apply to every message I send?

Writing down the answers takes a few minutes. Without them, a reader cannot tell whether a setup is as controlled as it appears.

When each model fits

A hosted platform suits a reader who wants an account, a catalogue of characters and one place to manage renewals. A self-hosted front end suits a reader who wants to choose where data is stored and which model is connected, and who accepts the setup work that comes with that choice. Neither model removes the need to read the terms of the model provider, because that provider may receive your messages.

What the comparison does not show

The shortlist records SillyTavern with no verified fields. Its license, fee and data terms are not yet verified, so the platform is not scored. This guide does not claim that a self-hosted setup is more private than a hosted one. The answer depends on the setup, the model provider and the settings you choose, and it can change with each of them.

Evidence and method

The SillyTavern homepage was captured as text on 16 October 2026. The hosted-platform examples in this guide come from the Character.AI privacy and subscribe pages, read in a browser the same day, and are summarised in the checklist guide. Limits: the project’s repository was not read in this edition, no installation was set up, no model provider’s terms were checked, and pages change without notice. Field definitions for this guide sit on the field rules page.

Related reading: the policy page guide, the data-controls checklist, and the home page shortlist.