Spaces:
Running
on
Zero
Running
on
Zero
license: apache-2.0 | |
title: Tether AI | |
sdk: gradio | |
emoji: π | |
colorFrom: gray | |
short_description: free, trauma-informed AI tool that detects emotional abuse | |
# π Detecting Hidden Harm: Abuse Language Patterns with Tether AI | |
Tether is a free, trauma-informed AI tool that detects emotional abuse patterns in text messages. It was built for survivors, by a survivor β blending behavioral science, NLP, and lived experience to uncover what often goes unseen. | |
While most AI tools focus on obvious toxicity or profanity, Tether specializes in **covert abuse**: | |
Gaslighting. Guilt-tripping. Control. DARVO. | |
The things that leave no bruise, but devastate your sense of self. | |
--- | |
## π― Mission | |
Tether exists to help people recognize patterns of harm in language β especially those trapped in emotionally abusive relationships who can't quite name whatβs happening to them. | |
This is not a commercial product. There are no ads, no accounts, and no tracking. | |
Itβs here because I lived it. And because I know how hard it is to see clearly while youβre still in it. | |
--- | |
## π What It Detects | |
Tether flags emotional and psychological abuse patterns at the **message level**, including: | |
- **Control** β overt and covert restrictions on autonomy | |
- **Gaslighting** β reality distortion, denial, contradiction | |
- **Blame Shifting** β manipulation of responsibility | |
- **Dismissiveness** β minimization, shutdown, invalidation | |
- **Recovery Phase** β manipulative calm following escalation | |
- **DARVO** β defensive reversal of blame | |
- **Guilt Tripping**, **Projection**, **Obscure Language**, and more | |
Each label is assigned based on trained thresholds, tone, and contextual pattern analysis. | |
It also provides: | |
- **Abuse Score (0β100%)** | |
- **Emotional Tone (anger, disgust, fear, neutral, etc.)** | |
- **Motif Tags** for repeat patterns | |
- **Risk Estimation** (early, escalating, or high-risk) | |
--- | |
## π§ How It Works | |
- **Model:** Fine-tuned RoBERTa, multi-label classification | |
- **Dataset:** Custom, real-language abuse dataset curated by survivors & reviewed for clinical clarity | |
- **Emotion Detection:** `j-hartmann/emotion-english-distilroberta-base` | |
- **Visualization:** Real-time timeline chart for multiple messages | |
--- | |
## β οΈ Safety + Disclaimer | |
Tether is **not a diagnostic tool**. It does **not replace therapy**, legal support, or personal judgment. | |
Instead, it offers **a mirror** β one that can help you ask deeper questions about how you're being spoken to. | |
If youβre in an unsafe situation, we recommend reaching out to a support network or crisis line. For U.S. residents, the **National Domestic Violence Hotline** is available 24/7 at 800-799-7233 or via text/chat. | |
--- | |
## π§ͺ Use Cases | |
- π§ Survivors seeking clarity on relationship patterns | |
- π§ββοΈ Legal or DV advocates reviewing communication logs | |
- π§ Behavioral researchers modeling patterns of harm | |
- π± People asking, *βIs this message okay?β* and needing an honest, emotionally neutral lens | |
--- | |
## π Try it | |
βΆοΈ **Live App**: [https://huggingface.co/spaces/SamanthaStorm/tether](https://huggingface.co/spaces/SamanthaStorm/tether) | |
π **More Info**: [https://www.usetetherai.com](https://www.usetetherai.com) | |
--- | |
## π©βπ» About the Creator | |
Tether is built by **Samantha Parnham**, a behavioral scientist, survivor, and advocate for trauma-aware tech. Her work blends ABA, psychology, and AI to create tools that help people feel seen, supported, and safe. | |
title: Abuse Pattern Detector V2 | |
--- | |
Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference |