Atmee v2 is a real-time video model that puts a talking avatar inside a website or app, so a user can speak with an AI agent and watch it answer. It is meant for teams building tutors, sales coaches, support agents, companions, and in-app characters, and for creators who want a persona people can call. Atmee already runs live calls on an earlier model. This release opens the newer one, which Atmanity has spent two years building for live, face-to-face conversations. An avatar starts from one image and a short voice sample, plus a personality and any documents it should answer from. Atmee v2 renders the face, body, and reactions live, including while the user is still talking. Causeur, a separate behavior model, handles personality, emotion, timing, and active listening, so the motion follows the exchange. The character can be photoreal, stylized, or not human at all. A developer can paste one setup link into Claude Code, Codex, or Cursor and have that agent create the avatar and the page around it, or start each session from their own server through the API. V2 access currently requires a higher plan or an invite code. Atmanity, the lab behind Atmee, was founded in 2024 by Chen-Ping Yu and Yi-Hsuan Tsai. Yu previously founded Phiar, an AR navigation startup Google acquired in 2022, and Tsai led AI there before both later worked at Google. The same model can also generate ordinary short video clips. Atmee says those clips come back up to 50 times faster and up to 100 times cheaper than the same job on general video models such as Seedance.