WasmEdge, the lightweight, high-performance WebAssembly runtime under the Cloud Native Computing Foundation (CNCF), just wrapped up a whirlwind tour that took us from the vibrant Open Source Summit Korea in Seoul to the massive gathering at KubeCon + CloudNativeCon North America 2025 in Atlanta.
Across both continents, one theme was undeniable: the future of AI is open, portable, and running at the edge. Here is a look.
WasmEdge at OSSummit Korea 2025: Real-Time AI Voice Agent at the Edge with Rust
Miley Fu, WasmEdge founding member and CNCF Ambassador from Second State, took stage at Open Source Summit Korea on November 4-5, presenting “Orchestrating Real-Time Multimodal AI Agents with Rust”
The emphasis on multimodal AI is particularly significant in 2025, as organizations increasingly demand AI systems that can understand and respond to voices and call MCP and different tools to interact with users with low latency. A significant development announced during the event was the collaboration between WasmEdge and UNICC (United Nations International Computing Centre).
KubeCon NA 2025
Session 1: What's New in WasmEdge - Michael Yuan
Video https://www.youtube.com/watch?v=3hOcBe9H648 and ppt slides: https://sched.co/27d4K
During the opening day of KubeCon, Michael Yuan, WasmEdge maintainer and co-founder of Second State, delivered a project lightning talk titled “What's New in WasmEdge?” This presentation provided a comprehensive update on the latest developments within the WasmEdge runtime, particularly focusing on the emerging role of WebAssembly as the runtime of choice for AI and machine learning workloads. Yuan emphasized that over the past two years, WasmEdge has undergone a significant evolution from being positioned as a “container alternative” to becoming recognized as the premier AI runtime for cloud-native environments. Key highlights of this talk included:
- Expanded GPU Support: WasmEdge now supports a growing array of accelerators including NVIDIA CUDA, AMD GPUs, Apple Metal, Intel GPUs, and specialized AI processors.
- Multi-Backend Driver Support: The runtime now provides abstraction over multiple inference engines including TensorRT (NVIDIA), OpenVINO (Intel), Core ML (Apple), and emerging quantum computing frameworks.
- Broader Model Support: WasmEdge can now execute a vast array of AI models ranging from large language models (LLMs) like Llama to multimodal vision models, embedding models, automatic speech recognition (ASR) models like Whisper, and text-to-speech (TTS) systems.
- Component Model Implementation: WasmEdge is among the leading WebAssembly runtimes with full support for the upcoming WebAssembly Component Model.
- Community-Driven Innovation: The presentation highlighted contributions from developers worldwide, including LFX mentorship participants.
Session 2: Rust Is the Language of AGI - Miley Fu
Video and ppt slides: https://sched.co/27FV4
On the second day of KubeCon, Miley Fu presented “Rust Is the Language of AGI,” a thought-provoking session exploring the intersection of Rust programming language capabilities and artificial intelligence development. This presentation connects to broader industry conversations about language design for AI workloads.
Rust's emphasis on memory safety, performance, and concurrency makes it exceptionally well-suited for building production-grade AI infrastructure like LlamaEdge and EchoKit. The Rust language is a perfect fit for AI agents — as the powerful Rust compiler can provide real-time and autonomous feedback to the agent to ensure the validity of the generated code. The RustCoder project is an open source effort to provide such tools. It consists of API and MCP services that generate fully functional Rust projects from natural language descriptions. These services leverage LLMs to create complete Rust cargo projects, compile Rust source code, and automatically fix compiler errors.
Session 3: Open Source at the Edge: Hardware, Firmware, and AI Stacks - Miley Fu & Saiyam Pathak
Video and ppt slides: https://sched.co/27FZH In the talk, we walked through:
- EchoKit hardware: ESP32 / Raspberry Pi, microphone, speaker, and optional display
- Firmware in embedded Rust, streaming audio over WebSocket
- A modular AI pipeline (VAD → ASR → LLM → TTS → MCP tools) where every component is swappable
- Kubernetes-at-the-edge patterns so a classroom, lab, or team can manage fleets of devices from a central control plane
Why people got excited:
- Full-stack open source: hardware, firmware, server
- Run everything locally for privacy, or mix in cloud models when it makes sense
- Great for education and DIY builders to really learn how modern AI agents are built end-to-end
If you are interested in building your own voice AI lab at home, in a classroom, or for your team, you can explore the project here: https://echokit.dev and the server code here: https://github.com/second-state/echokit_server
The WasmEdge Project Booth: Hands-On AI at the Edge
Beyond the formal presentation sessions, WasmEdge maintained a dedicated project booth in the CNCF Project Pavilion at KubeCon NA 2025. This booth served as a gathering place for users, developers, and enterprises interested in exploring WebAssembly for their cloud-native and AI workloads. Visitors had the opportunity to see live demonstrations of WasmEdge running AI inference models in real-time, engage with maintainers about technical questions and use cases, and explore integration patterns with Kubernetes and other cloud-native tools. Notably, the WasmEdge team received on-site orders for echokit, an exciting indicator of commercial interest in Rust-based AI solutions. Make your purchase here: https://echokit.dev/
EchoKit: Voice AI Goes Portable
The presence of echokit orders at KubeCon signals growing commercial traction. EchoKit is an open-source AI service orchestrator designed specifically for real-time voice interactions. It coordinates multiple AI services—including LLMs, automatic speech recognition (ASR), text-to-speech (TTS), voice activity detection (VAD), and the Model Context Protocol (MCP)—to create intelligent conversational agents.