Number 1: 
Ora
Make your site work for agents
- AI testing & review
- United States
- DR 64 +1 (up) since last check
AI testing & review · Netherlands
Toxic
On-device moderation that flags hateful, abusive and threatening text in 23 European languages, before the message is sent.
- Community votes
- 0
- DR by Ahrefs
- 37 no change since last check
Ahrefs Domain Rating benchmarked 5 Nov 2026, rebuilt 5 Nov 2026.
A multilingual hate speech detector that runs locally on devices to flag hateful, abusive and threatening text across 23 European languages.
What it is
Toxic is an on-device hate speech detection model built by Desert Ant Labs. It flags hateful, abusive, and threatening text in 23 European languages before messages are sent, with no data leaving the user's device. The model scores three dimensions separately: hate speech (incitement to violence or hatred against protected groups), abuse (insults ranked by severity), and threats (violence against persons or groups). It achieves 0.847 macro F1 on Multilingual HateCheck across seven EU languages from an 80MB model that runs on device, outperforming comparable guard models. Toxic is available as a closed beta with free access up to 100k monthly active devices with unlimited inference.
What it is built with
- Three separate scores: hateful, abusive, threat
- Ten target signals under the hate score: race, color, religion, descent, national or ethnic origin, sexual orientation, gender, disability, age, other
- Runs fully on device with no server calls
- Supports 23 European languages from one model file
- 80MB model size optimized for iOS, Android, and browser
- 0.847 macro F1 accuracy on Multilingual HateCheck
- Warn before message is sent
- Comment triage without infrastructure
Pricing
Pricing model: Freemium
Free up to 100k monthly active devices with unlimited inference. Custom licenses available.
Follow the workshop. Log in to follow Toxic's workshop
Following Toxic means an email when they reach a milestone, post an update or launch a new product.
Toxic metrics
| Metric | Value |
|---|---|
| Domain Rating | 37 no change since last check |
| Trust Flow | 20 no change since last check |
| Citation Flow | 39 no change since last check |
| Referring domains | 616 no change since last check |
| External backlinks | 9,602 no change since last check |
Ahrefs Domain Rating benchmarked 5 Nov 2026, rebuilt 5 Nov 2026. Main build topic: Computers/Systems (20). Majestic benchmarked 5 Nov 2026, rebuilt 5 Nov 2026. Link data built by Majestic. Trust Flow, Citation Flow and Topical Trust Flow are trademarks of Majestic-12 Ltd.
Toxic domain build sheet
Open the build sheet
| Domain age | 2 years |
|---|---|
| Registered | |
| Expires | |
| Registrar | Register SPA |
| DNSSEC | Not signed |
Toxic after-launch build log
Since Toxic launched on Maker Launches on 5 November 2026, these are the figures we have tracked.
- Domain Rating (Ahrefs)
- 37
- First measured 5 Nov 2026
- Trust Flow (Majestic)
- 20
- First measured 5 Nov 2026
- Referring domains (Majestic)
- 616
- First measured 5 Nov 2026
Day 0 since launch. Next receipt: day 30, 5 Dec 2026. We show what changed, not what caused it: many things affect these numbers. Nothing here is estimated.
Planning a launch?
Plan your launch with our free playbook, from preparing your listing to choosing more launch directories.
- The week before
- Launch day
- Launch week
- Afterwards
Compare and explore
Questions from the bench
- What is Toxic?
- On-device moderation that flags hateful, abusive and threatening text in 23 European languages, before the message is sent.
- Does Toxic run on device?
- Yes. Toxic runs on the device, with no server call, so the data stays with the user.
- Is Toxic available yet?
- Toxic is in closed beta. Early access is available on request.
- How much does Toxic cost?
- Each model is free up to 100k monthly active devices with unlimited inference. Contact for custom licenses.
- How accurate is Toxic?
- Toxic achieves 0.847 macro F1 on real Multilingual HateCheck across the seven EU languages it covers, from an 80MB model that runs on device. The closest guard model through the same harness is Mistral's Shieldstral-1.0-3B at 0.786, and it wants a 16GB GPU.
Leave a review for Toxic
Nobody has put Toxic on the bench yet. Leave the first review.
Reviewing is free, and nobody can pay for a review. Read the bench notes
Maker reviews · Workshop notes · Show it off · Did you make this? Claim it · Report this build
Toxic has a free listing. The owner can unlock the full profile with a paid launch for $5:
- Founder story (locked)
- Milestones (locked)
- Demo video (locked)
- Dofollow website link (locked)
More builds like this
Number 1:
Number 2:
Ocular AIEncoding Human Expertise into Machines
- AI testing & review
- United States
- DR 27 −1 (down) since last check
Number 3:
UnfixIndependent verification for bug fixes in C, C++ and firmware
- AI testing & review
- DR 0 no change since last check
Number 4:
TestOptim AINo-code AI QA testing for web apps that explores your app, writes end-to-end tests, and reports bugs with screenshots.
- AI testing & review
- DR 5 no change since last check