Today we release VisionPsy-Nano: state-of-the-art vision-language models at 460M parameters, built to run on-device. It leads every model in its weight class on 16 of 17 benchmarks with the highest overall normalized score in its class, and it tops all four capability areas: document understanding and OCR, visual perception, reasoning, and instruction following. Open weights, Apache 2.0.

Introducing VisionPsy-Nano: state-of-the-art vision-language models at 460M parameters, small enough to run on the phone in your pocket. VisionPsy-Nano-460M leads every other ~0.5B vision-language model tested & compares favourably on 16 of 17 benchmarks, tops all four capability categories, and outperforms models up to 2.3x larger. Two variants, one tuned for maximum accuracy and one for real-time responsiveness. Open weights, Apache 2.0, and no cloud in the loop.

@thisonedevv built a code school for our stack: around 70 QVAC samples, runnable on your own machine, fully offline, with instant feedback on each lesson. Good to see the community growing the ecosystem!

QVAC SDK 0.16 is live. 🚀 - Ideogram 4 image generation, on-device - LTX-2 video with synchronized audio - NVIDIA's GR00T robotics model, running local - OpenAI compatible server improvements for better coding agents integration - Speech packages improvements and TypeScript-native migration Full release notes:

The mechanic behind every story like this: the conversation lives on a server, so it can be linked, crawled, and indexed. On-device AI removes the first step. No server copy, no URL, nothing to index. Privacy you do not have to configure.

Your laptop's AI forgets long conversations because it runs out of working memory, not intelligence. QVAC team shipped an open-source fix: TurboQuant, a 5x memory compression algorithm for local AI, available in QVAC SDK. Read how it works:

Have you ever tried building with the QVAC SDK? This dev just shipped a fully functional local AI second brain running entirely on QVAC, weekend build, open source. This is exactly the kind of app the SDK is for.

Physical AI will run on billions of devices. Centralized infra means someone else controls the robot in your house. Local-first changes that equation, same logic we build into QVAC. Full piece on why decentralization is the safety requirement, not a nice-to-have:

"Chat with your documents" usually means uploading them to someone else's cloud. The QVAC SDK does the whole thing on-device. It ships native RAG: ragIngest your files, embed turns them into vectors locally, ragSearch finds the relevant passages, and completion answers from text the model can actually see. No external vector database. What you can build with it: a knowledge base over years of your notes, contract and research Q&A, a support bot grounded only in your manuals, a private second brain that cites its sources. Your private documents never leave the machine. No cloud vector store, no API key, no leak surface. npm install @qvac/sdk

Cloud AI forces users to rent intelligence from centralized, high-cost data centers. Tether is breaking this model with local, edge-first AI. By enabling on-device inference and training on everyday hardware, AI becomes a portable capital asset owned entirely by you. Read more:

Generative images and video, with no cloud bill and no content filter you do not control. The QVAC SDK runs both on-device. diffusion runs FLUX.2-klein and Stable Diffusion locally, blends reference images for style, and upscale takes the result to high resolution. And video generates short clips from a text prompt, all on the device. What you can build with it: avatar and sticker makers, design and mockup tools, wallpaper and art generators, short animated clips and video effects, character creators. Every image and frame is free and private. No per-image cost, nothing uploaded, works offline. npm install @qvac/sdk

50+ languages. Zero servers. QVAC now powers fully on-device translation inside @keet_io. Every word is translated locally. Nothing sent anywhere. No one listening. Test it on - the chat app that knows nothing about you.

Rome. Le Corsie Sistine. The Last Question. @paoloardoino took the stage to make the case for local AI over centralized AI and for putting intelligence back in the hands of the people who use it. Not your AI, not your intelligence. #TheLastQuestion# #QVAC#

QVAC is built by a team working at the frontier where AI, biology, and computing converge, a convergence that is compressing decades of progress into years. That means real research, in the open. Tether just had four papers accepted at ICLR, and open-sources the whole stack for millions of developers and researchers to scrutinize.

QVAC now has a dedicated community channel on @keet_io, and the team is in there with hundreds of AI builders and enthusiasts. It is the easiest place to ask us anything and talk on-device AI directly. New to Keet? It is the most private messenger out there: fully peer-to-peer, end-to-end encrypted, no servers, no phone number, no metadata.

QVAC SDK 0.14.0 is live. This release makes the on-device stack faster on mobile, ships the developer-agent path, and takes local text-to-speech to 31 languages. Main highlights: - OpenCode and OpenClaw. The first official OpenCode plugin, plus a maintained OpenClaw compatibility path, both built on managed mode and qvac serve. Point a coding agent at a local model with far less setup and far fewer surprises. - Brain-computer interface transcription, on the SDK. Take recorded neural signal data and decode it into text, fully on-device, no cloud. Stream it in chunks through a simple API. In 0.14 it runs GPU-accelerated on iOS. - Text to Speech in 31 languages with our Supertonic3 upgrade. VOICE AND SPEECH - Supertonic3 multilingual TTS, 5 languages to 31. - Chatterbox and Supertonic now run on the Android GPU, with lower memory use (especially on iOS), quantized s3gen Chatterbox support, and a fix for Chatterbox occasionally emitting random speech. - Whisper transcription now runs on the iOS GPU. Parakeet runs on the Android GPU, with steadier real-time streaming. VISION AND OCR - VLM multi-tile batching: high-resolution Pan and Scan images are encoded in one pass instead of tile by tile, for faster vision throughput. - OCR on ggml (EasyOCR and DocTR) reaches full speed parity with the onnx path, across Metal, OpenCL, and Vulkan. PLATFORM AND RELIABILITY - Dynamic compute backends on Linux: one build picks the right backend at runtime, and opens the door to ROCm and CUDA support without per-backend builds. - Thinking tokens are kept out of the model context, so reasoning no longer fills the KV cache. SDK 0.14.0 is now leaner and faster to start. Let’s build.

The Tether Developers Cup is live: a global hackathon that runs like a football knockout tournament. Teams of 1 to 4 build a football-themed project on the Tether open-source stack, pick a nation to represent, and survive five rounds until one team lifts the cup. Three tracks: - @qvac for on-device AI - @Pears_p2p for peer-to-peer apps - @WDK_tether for self-custodial wallets 8,000 USDt in prizes. Free, worldwide. Register June 28 to July 6, live Final July 15. Register with the link below ⬇️

You probably know @openclaw, one of the fastest-growing open-source projects around. An OpenClaw harness lets you build, code, analyze, and take real control of your machine. But the more access an agent has to your machine, the more it matters where your data goes & who can see it. With QVAC, you can run your OpenClaw agent fully on your own machine in a few minutes. Nothing leaves your computer. Full setup guide on the blog, see first comment below. 💡 Pro tip: already running OpenClaw? Send the blog article to your agent and let the it do the set up.

Your local AI just got up to 5x more memory. Same model. Same device. Nearly zero accuracy loss. QVAC SDK 0.12.0 integrates TurboQuant - Google Research's latest memory optimisation algorithm. What is TurboQuant? The KV cache is the memory your model uses to track a conversation. As context grows, it fills up fast. 32K tokens. 64K. Game over. TurboQuant compresses it up to 5x with no accuracy loss. What does it unlock for you? Your app had a 16K token ceiling? It's now 96K. On the same device. Just update the QVAC SDK to get up to 5x more efficiency. No code changes. All from one SDK. The TurboQuant integration unlocks sovereign intelligence for more people, on more devices. Learn more →