Telegram community security · Private beta
The free anti-spam bot for Telegram — classical, CPU-only, and explainable. Catches homoglyphs, edit-after-post, admin impersonation, and obfuscation at $0 per message; every action carries a reason. The paid M2 tier adds LLM-grade detection for AI-driven attacks — conversational scam funnels, hyper-personalized phishing.
The threat → the stack
Attackers now use jailbroken LLMs, homoglyph obfuscation, and slow-pivot conversational funnels that read as on-topic until they aren't. Keyword filters and static blocklists don't see it coming. Targa answers with a coordinated stack — not one trick.
Every decision carries a structured reason — /why <id>. Opacity is the universal admin pain point. Not here.
Deterministic detectors screen every message — homoglyphs, edit-after-post, link density. CPU-only, $0 per message.
Smart-routed LLM judges only the ambiguous 1–5%; each verdict is reused fleet-wide. Catches what classical can't.
Who is sending — not just what. Avatar pHash vs admin set, join-age, account heritage, impersonation lockout.
Classical, deterministic anti-spam — no LLM, $0 per message, free forever. Closes the gaps that defeat rule-based bots today.
Adds smart-routed LLM detection — most messages resolve cheaply, the model only judges the ambiguous fraction. Catches what classical can't.
Homoglyphs, edit-after-post, image-text scams, admin impersonation, aged-account swarms — most caught in the free tier.
Hosted & multi-tenant: a campaign judged once is caught fleet-wide. Detection gets better and cheaper as the network grows.
Most security companies publish at intervals. We run a live, public Spam Index and ship original threat research — The State of Spam in Digital Communities is the first deliverable. And it feeds straight back into detection.
Open the research arm →