Giving devices a voice, as long as you call home first

What it does

EVA OS provides open-source iOS and Android SDKs that pipe real-time voice, video, and agent commands between edge devices and AutoArk’s cloud platform. It handles WebRTC streaming, multi-voice synthesis, avatar lip-sync, and tool use, aiming to turn phones, toys, and eventually ESP32-class hardware into conversational terminals. The project is MIT-licensed client code for a hosted multimodal backend.

The interesting bit

The README pitches EVA OS as a hardware “nervous system,” but the actual neurons—the EVA-flash model, memory layer, and reasoning—run remotely. That makes this less of an operating system and more of a well-equipped client framework for a proprietary AI service, complete with built-in MCP tools and multi-agent routing.

Key highlights

• Full-duplex audio/video with barge-in via WebRTC/LiveKit
• Open-source iOS and Android SDKs with React Native examples
• 10+ TTS voices and a real-time lip-sync avatar engine
• Built-in MCP tools (weather, search, maps) and configurable agent templates
• Embedded SDKs for ESP32/Rockchip/MCU listed as “coming soon”

Caveats

• Requires an AutoArk platform account and API key; not self-hostable
• Several promoted features (custom avatars, embedded SDKs, edge-cloud collaboration) remain on the roadmap and are not yet available
• The “EVA-flash” model is free but cloud-hosted, so devices need persistent connectivity

Verdict

Useful if you’re shipping consumer IoT or mobile apps and want to outsource LLM hosting, real-time audio pipelines, and avatar rendering. Look elsewhere if you need fully offline, air-gapped intelligence or immediate embedded support.