Inter-1 is Interhuman's omni-modal model for social intelligence, served through the Signals API. It watches video of a person communicating and reports social signals as they occur, so an AI product can react to how someone is speaking, not only to the words. Target uses include sales coaching, tutoring, user research, meeting copilots, AI interviews, communication training, and healthcare support.
The API detects ten signals, each returned with a timestamp, a probability, and a rationale, alongside an engagement state (engaged, disengaged, or neutral) and conversation-quality metrics with a CQI score, dimension scores, and event metrics. Cues that feed the signals include gaze shifts, pre-speech exhales, throat clears, hedging language, and emphatic overcompensation.
You can analyze a recorded file through a single upload request or stream live media over a WebSocket, receiving typed events per chunk. TypeScript and Python SDKs wrap the endpoints, and a playground lets you try the API before integrating it.
Features
- Ten social signals: agreement, confidence, confusion, disagreement, frustration, hesitation, interest, skepticism, stress, and uncertainty
- Engagement state: engaged, disengaged, or neutral, reported separately from the signals
- Conversation quality: a CQI score with dimension scores and event metrics
- Timestamped output: each signal carries a timestamp, probability, and rationale
- Upload endpoint: POST /v1/upload/analyze for recorded video
- Streaming endpoint: wss://api.interhuman.ai/v1/stream/analyze with MediaRecorder or file segments and per-chunk events
- SDKs: @interhumanai/sdk for TypeScript and interhumanai for Python
- Playground: an interactive console for testing calls against the live API
