Today, I'm proudly representing @SafeCircleAI at #SECyT in Murcia, Spain! It's a great opportunity to showcase what we've been building, demonstrate our technology, and connect with people who share our vision. Our mission is simple: use AI to make the digital world safer for children and families. Excited to share SafeCircle with everyone here and hear their feedback. Special thanks to @CeeiMurcia for the opportunity and continued support. Let's keep building technology that makes a difference!

See you tomorrow at #SECyT! I’ll be there from 10 to 2, showcasing @SafeCircleAI and sharing what we’ve been building to make the digital world safer for children. If you’re attending, come say hi! I’d love to meet you and show you SafeCircle in action. Event: fseneca.es/secyt26/ I’ll be sharing photos and highlights from the event on Friday.

SafeCircle is now open to everyone. No more invite. The beta is live, and we want you to test it. Sign up: app.safecircle.tech

SafeCircle checks your child's WhatsApp, Instagram, Telegram, Snapchat and TikTok chats on their Android phone. It never sends the message text, to us or to you. You get the warning: risk level, what kind of risk, which app. safecircle.tech

Horizon vs Vigil. Our new Vigil safety model is already showing a serious performance gap: - 207ms vs 747ms latency - 3.6× faster - +30 pts catch on unseen text - 17% smaller - Calibrated per-category thresholds - Built for on-device inference Vigil is based on Laya, an open-source JEV alternative. This is getting interesting. huggingface.co/safecircleai/…

Vigil is now live in the AI Playground. Go test it. See what it catches. Download: github.com/safecircleia/ai-p…

After attending the @aics_summit , we’ve finally secured a new partnership with @informaticaumu . @informaticaumu is giving SafeCircle compute power on their cluster, which we’ll use to train and improve our LLM models for on-device inference. We’re building models that can bring real-time risk detection closer to the device — and now we have the compute to push them further. Read more about the technology behind it ↓ blog.defectiv.es/posts/un-cl…⁠

We’re taking SafeCircle from the web to @Android. The new parent app brings the same dashboard experience directly to your phone — alerts, children, risk monitoring and insights, all in one place. And we’re going one step further: a dedicated @SamsungMobile build following Samsungs's design language

Updated our partners tab, we were missing @lambda in there, thanks again for those 7K US in credit inference. They been put to good use. safecircle.tech/partners

Excited to be testing safecircle on @android Release soon on @GooglePlay

Privacy isn’t a feature we bolt on later. For SafeCircle, it’s part of the architecture. We’re building toward GDPR-ready infrastructure from day one: DPIA, consent, data minimization, encryption, deletion rights, auditability — with AI processing designed to stay on-device. The goal: protect kids without turning their private conversations into a dataset. Privacy by design. Safety by default.

While the full parent dashboard is in development, I'm shipping something parents can use right now. @SafeCircleBot — a Telegram bot that brings Horizon alerts directly to your phone. Here's what it does: 🔴 Real-time push notifications when a risk is detected "Emma — grooming detected, 91% confidence, WhatsApp, 2 mins ago" One tap to view the full reasoning. One tap to mark it read. /alerts → your 5 latest unread alerts across all children /status → all devices, online status, unread count at a glance /children → paired devices and last-seen times /quiet → toggle quiet hours so you're not woken up at 3am No app to install. No new account. If you use Telegram, you're ready. Link your SafeCircle account once from the dashboard → Settings → Notifications → Connect Telegram. That's it. The full dashboard is coming. But parents needed something today. So I built it over the weekend. $SC @SafeCircleAI — Tomas

Most AI safety systems assume they can throw compute at the problem. We can't. Horizon has to run 24/7 directly on a child's Android phone — not in the cloud. So we're optimizing for something very specific: high-accuracy threat detection under extreme resource constraints.

Horizon NX is here. We could just call ChatGPT or another general-purpose model. Instead, we’re building our own AI specifically for safety detection. Why? General models are built to answer questions. Horizon is built to detect risk — understanding context, patterns and signals that matter for SafeCircle, while being small enough to run where we need it. More control. More privacy. Less dependency on someone else’s API. A model we can actually optimize for our problem. And we’re building it in public. → github.com/safecircleia/hori… Horizon NX. This is our model. This is just the beginning. @RobinhoodApp @vladtenev

We don’t do “coming soon.” Orbit is being built in public. Source code: open. Builds: public. Progress: verifiable in real time. Watch us cook ↓ github.com/safecircleia/orbi…

SafeCircle AI Playground is now available on our web. Test our in-house models with offline, no-limit inference powered by @Google LiteRT-LM. Help us improve, share your feedback. ai.safecircle.tech

We built Horizon to run entirely on-device. Here's every architectural decision that made that possible — and what we'd do differently 🧵