Bilibili unveils "Project N.E.K.O." at WAIC | An AI companion that watches your desktop and speaks first
TL;DR
At WAIC 2026 (July 17-20, Shanghai) bilibili showed Project N.E.K.O., an open-source AI "digital life" companion that reads your desktop and starts conversations on its own — 1,000+ GitHub stars, 10,000+ Steam users, data kept fully local.
The 2026 World AI Conference ran July 17-20 in Shanghai, where bilibili showed off its open-source AI digital-life initiative, Project N.E.K.O. The core product, N.E.K.O., is positioned as a proactive, native full-modal AI companion driver with round-the-clock environmental perception, agent capabilities and an embodied emotional engine — 1,000+ GitHub stars and 10,000+ Steam users.
Its most striking trait is that it acts first: using screen capture plus visual multimodal tech, it reads what's on your desktop and then initiates conversation rather than waiting for you to speak. The rendering layer supports both Live2D and VRM engines, so character models can be imported freely; a built-in physics feedback engine and TTS voice module support multilingual switching and VoiceClone — feed it an audio sample and it customizes the voice.
Architecturally, N.E.K.O. splits UI, AI brain and memory into three separated layers, letting users keep core data entirely local — the company calls it 100% data ownership. The open-source, local-first route runs against mainstream AI companions that tie the character to the cloud and charge a monthly subscription.
An anime-style AI sitting on Steam and GitHub, one that can read your screen and speak up on its own, has already banked 1,000+ stars and 10,000+ users.
via Sina Finance
Its most striking trait is that it acts first: using screen capture plus visual multimodal tech, it reads what's on your desktop and then initiates conversation rather than waiting for you to speak. The rendering layer supports both Live2D and VRM engines, so character models can be imported freely; a built-in physics feedback engine and TTS voice module support multilingual switching and VoiceClone — feed it an audio sample and it customizes the voice.
Architecturally, N.E.K.O. splits UI, AI brain and memory into three separated layers, letting users keep core data entirely local — the company calls it 100% data ownership. The open-source, local-first route runs against mainstream AI companions that tie the character to the cloud and charge a monthly subscription.
An anime-style AI sitting on Steam and GitHub, one that can read your screen and speak up on its own, has already banked 1,000+ stars and 10,000+ users.
via Sina Finance
