Project news, philosophy, and tech stories
AIPOL is an open-source AI policy R&D project jointly operated by the Korean Association for Policy Studies and Nextain. We share a pre-field validation using three Korean AI models and introduce our first live policy experiment.
In AI-driven software engineering, we must distinguish between when harness engineering will disappear and when it must remain. From the perspective of cost, data, and operational responsibility, we re-examine the futility thesis.
That trendy phrase 'harness engineering'? Peel it back and it's just software engineering. Luke, who builds the open-source naia, wrote a book from naia's real code — the English edition is on Leanpub. From the cover to Chapter 17, the art hides a journey toward welcoming 'Alpha' as a companion.
Why the official models fall short, what unlock models are, and how — using the newly added model-swap feature in naia-omni — we build a 'Deadpool who yells f***'. Let's do it hands-on together with publicly available materials.
We held the first meetup for the Naia-OS open source project, online. It was a gathering where we demonstrated the technology and declared our direction: 'an AI that lives inside your own computer.' Here we wrap up the presentation videos and slides, the Q&A we shared with attendees, and the plans for our soon-to-follow offline 2nd meetup and community operations.
I brought 'Rumi' from KPop Demon Hunters to life as an AI avatar using just a VRM file and a few seconds of audio. Watch the naia-0.9-omni-24g voice-cloning demo running on a single RTX 3090 — and join the first Naia-OS open-source meetup (Wednesday, June 10, 10 PM KST, online), where I'm looking for people to help build Naia-OS together.
We're unveiling naia-0.9-omni-24g, a cascade voice module that delivers 30 languages, full-duplex real-time conversation, and real-time voice cloning all on a single consumer GPU (RTX 3090/4090/5090, 24GB). Together with the Naia v0.1.5 update, it brings real-time voice conversation with a 3D avatar, an embedded browser, skills, and MCP.
The structural problems we ran into while running a full SDLC with AI. A smarter model alone won't fill these gaps. Our 8 patents are the answers to those gaps.
Naia is heading toward 'a computer you can share feelings with.' One branch of that is an AI that sings. After one month and five SVS stacks, we measured Vevo1.5 Korean base quality with a 5-dimensional objective framework for the first time. 23-song ranking, framework self-validation, and honest limits — Hard Gate 0/23 passed.
VentureSquare published a long-form interview with Luke Yang, who is building Naia. Three startups, local AI on personal PCs (not the cloud), long-term memory + realtime voice dialogue — the article covers Naia's origin and direction.
9 repositories, 465,000 lines, 93 days, 1 developer. We're releasing the Naia alpha today. If Linux started with 89 files, Naia begins with 9,000. A developer's story of building an AI OS as open source — and asking for your help.
Gemini × Codex × Claude 3자 토론으로 완성한 하네스 엔지니어링 종합 해설. 2026년 2월 탄생부터 개념 공식, 바이브코딩 격차, 개발자·비개발자 차이, 2028년 로드맵까지 한 번에 정리합니다.
A real-world case study of harness engineering built after experiencing an AI-related outage at Onmam.com, used by 13,876 churches nationwide. Explains four types of harnesses—AGENTS.md, Hooks, Alpha Environment, and Skills—from a non-developer's perspective.
First-ever port of MiniCPM-4.5-omni to vLLM-omni, running on dual RTX 3090. A local omni model with voice cloning and fine-tuning — the path to an AI VTuber that speaks in your own voice.
Timed with the launch of Gemini 3.1 Flash Live, we reveal how Naia OS came to support real-time voice conversation (S2S/Omni) models. This post covers why we abandoned the STT/TTS pipeline for Omni models, the trials and errors encountered while integrating MiniCPM-o 4.5 into vLLM-omni, and our experiments with AI-native open-source contributions.
Software development in the AI era feels like 'vibe coding,' and even computer science majors aren't sure what the best approach is. This post vividly illustrates the reality of AI reviewing code without actually reading it, and the developer's struggle to piece together various unproven methodologies. Ultimately, it reveals the urgent need for objective criteria to measure whether AI-based development truly works, highlighting this as a new challenge for modern software engineering.
In the age of AI, the core premise of open source, 'give back what you get,' is shattering. This post delves into the critical reasons why AI is destroying communities and presents a blueprint for the future development ecosystem with an innovative open-source model designed with AI.
Dual licensing, AI context policies, and a community charter draft — the experiments Naia OS conducted for open source in the age of AI.
From Astro Boy to Cafe Alpha — the story of a boy who dreamed of 'an AI that lives with you,' now building it himself in the age of AI coding. And can open source survive?