.png&w=256&q=75)
🤓 Latest Submissions
Countersign
When a call demands something urgent and irreversible, like a wire transfer or a credential reset, Countersign answers first. It is a real-time voice agent, built on the AssemblyAI Voice Agent API, that runs a calm, adaptive security interview before any request can move, checking what the caller claims to know, asking cross-checking questions, and tracking whether their story holds up turn to turn. Every claim is checked against structured evidence, and a deterministic engine, not the AI, decides and writes every word the agent speaks. Countersign never claims to detect a fake or cloned voice. It proves things behaviorally: what the caller knows, whether their story stays consistent, and what independent checks confirm. A fully verified call never releases money. It only stages the request for a second, independent human to approve. A failed call freezes the request, opens an incident, and leaves a hash-chained evidence export behind. Countersign runs on the AssemblyAI Voice Agent API: a stored agent bound per call, with our own LLM endpoint so the deterministic engine writes every word the agent speaks; single-use server tokens; max-accuracy transcription with the call's proper nouns as key terms; adaptive turn detection with real barge-in; and AssemblyAI's own two-voice session recordings, which power the no-microphone replay of real calls. Realtime primitives, not a chat wrapper. Try it: open the demo and pick a role card. Each card shows what to say and the facts you need. No microphone? "Watch a recorded attack" replays real recorded calls through the real engine.
30 Sep 2026
