WeLivedit.AI

What we do

How it works

Exposure Control

Install the browser extension and screen comment sections — hiding flagged content until you’re ready to see it. You’re still in control; nothing is silently erased.

Trusted Circle

Join a community of people who get it. Invite trusted peers to review and report content alongside you — because context matters, and no algorithm alone works for specific situations. Together, you build a clearer, shared picture of what’s actually happening.

Who it's for

If you’re a researcher speaking publicly, a journalist covering a hard beat, an activist, or just someone who’s tired of getting blindsided by their own mentions — this is for you. You shouldn’t have to choose between having a voice online and protecting your peace.

What we're not

We’re not promising to fix the internet. We’re not another black-box AI pretending it has all the answers. We’re a small team trying to build something honest, useful, and actually shaped by the people it’s meant to protect — one step at a time

How our principles shape our technology

We believe the people who experience online abuse should shape the tools built to address it. That belief guides our technical choices.

We use AI only where it fits the problem.

Hate speech is made of language, so we use language models to classify it: to identify abuse and name its patterns. We don’t use AI to give advice, stand in for human support, or make decisions for the people we serve.

We're clear about what we use and what happens to your data

We currently use OpenAI’s language models to classify content. When our tool checks a comment, the text is sent to OpenAI for classification, and we don’t store it. OpenAI handles that data under its own policies, which we can’t independently verify. We think you should know that, and it’s one reason we’re changing how our technology is built.

We're moving towards open-source models

Commercial AI models are black boxes. We can’t fully inspect how they were trained, what data they keep, what resources they use, or what biases they carry. Moving to open-source models will let us examine, adapt, and govern the technology ourselves, and give our community a real say in it. We’ve begun to build data annotation frameworks, which we can use in this transition.

We’re not bound to LLMs.

Where abuse is clear-cut, we’re moving towards transparent, rules-based detection defined by people who know it firsthand. We believe language models should be reserved for ambiguous cases where context matters. This keeps bias visible and places judgement where it’s actually needed.

We measure bias by its real-world impact.

Our models aren’t one-size-fits-all. They’re configured with and for specific communities. So rather than asking only whether a model is biased in the abstract, we ask who could be harmed by its mistakes, given how it’s actually used.

Users will be able to challenge the technology.

We’re designing our tools so that users can dispute a classification, suggest new labels, and help decide what data trains our models. We’ll build and test these features with our pilot community.