No training our models on your chats.
We use existing AI models to run Hug. Your conversations are not used to train our own models.
read detailsa friend should be clear with you
We do not train our own AI models on your conversations.
Hug remembers things, replies, and sometimes checks in first. You deserve to understand how that works—and what happens to the things you share.
Presentation · Friday
An illustrative example of memory.
Saved context helps shape a later reply.
One set of data commitments.
Wherever you talk to Hug.
01 / the essentials
These practices apply to Hug across its messaging channels, including iMessage and Discord.
We use existing AI models to run Hug. Your conversations are not used to train our own models.
read detailsWe do not sell personal data or use it for advertising. Providers process data to help run the service.
read detailsAI providers process conversations. Authorized team members may review them to operate and improve Hug.
read details02 / behind a reply
Messages pass through real systems and providers. Here’s the broad picture, from sending a message to getting a reply.
A message reaches Hug through the channel you use.
Hug processes the message with relevant conversation history and memory.
Existing models process the context to help decide and generate a response.
Hug sends its response through your messaging app.
Context can persist. Messages and derived memories are stored for continuity.
Operations use data too. Logs can contain messages; authorized staff may review conversations.
A simplified view. Media processing, hosting, logging, and analytics involve the providers listed in our privacy policy.
03 / memory & models
A saved memory and a trained model are different things. That distinction matters when we say Hug learns about you.
Hug can retrieve a fact or earlier exchange to make a reply relevant. That context is stored data, and you can request its correction or deletion.
Training changes how a model works. TAICU does not use your conversations to train its own models. Our details page explains the provider terms too.
04 / why we’re building this
Hug is built by TAICU. Our research asks how AI should behave in an ongoing relationship: what it remembers, when it speaks, and how it handles getting things wrong.
Improving Hug can include reviewing conversations and failures. Authorized team access is part of our disclosed data practices.
read about our researchWhat should be remembered, revised, or forgotten?
When is a check-in welcome?
How should Hug recover from a misunderstanding?
How can an AI respect the people it talks to?
05 / your choices
Across messaging platforms, you can ask about your data, request changes, or ask us to delete it.
Ask what personal data we hold about you and request a copy.
Ask to correct inaccurate information or remove a saved memory.
Request deletion by email. Removing or blocking Hug is separate from deleting stored data.
a closer look / Discord
The same data practices apply on Discord. Shared spaces also need clear controls over where Hug has access and when it participates.
Hug can receive messages in accessible channels even without a mention. Reply settings determine when it joins in; permissions determine where it has access.
read the Discord detailsRemoving Hug stops new server messages reaching it. Stored data requires a separate deletion request.
keep asking questions
Read the fuller explanation, follow the policy references, or talk to the people building Hug.