QVAC

The QVAC team published a developer guide on July 22, 2026 orienting builders on local AI trade-offs and how to use coding agents with the QVAC SDK via a QVAC.md context file. The guide frames the constraints of local AI (hardware-limited model size, weaker frontier-model performance on hardest reasoning and longest co The guide introduces QVAC.md as the single file that enables coding agents to write QVAC code correctly on first attempt, covering the trade-offs of running AI on personal hardware, what the SDK can do, and how to set up an agent properly. First-run model downloads require an internet connection, after which everything

The QVAC team published an application showcase on June 16, 2026 demonstrating a local AI football match predictor built entirely on the QVAC SDK. The app pulls team data from a public football dataset, runs a Monte Carlo simulation of the match 10,000 times to produce win/draw/loss probabilities and a scoreline heatma The architecture illustrates a practical pattern for local AI applications: cheap statistical simulation produces a probability distribution, then a local model interprets that distribution into a readable decision with transparent chain-of-thought reasoning. A secondary tab lets users browse team statistics and simula

A WIRED sponsored-content piece dated August 27, 2026 details QVAC's architecture as a free, open-source, decentralized AI platform that lets developers run, train, and fine-tune AI models directly on local hardware with no cloud fees, subscriptions, or API costs. The article introduces QVAC Fabric, launched in March 2 CEO Paolo Ardoino is quoted arguing that centralized AI is a dead end due to physics-imposed latency, single points of failure, and concentrated control. The piece frames QVAC as a modular SDK small enough to fit inside a light bulb, stackable and infinitely scalable across devices, and operating without internet conne

Tether's AI Research Group announced on June 1, 2026 the production release of its open-source implementation of TurboQuant, Google Research's memory compression algorithm, integrated into QVAC Fabric, an evolution of llama.cpp that forms the core of Tether's local/edge AI engine. The release includes a full quantizati The technical motivation is concrete: at roughly 262,000 tokens (several hours of conversation or a few hundred pages of text), the KV cache for a 4B model can consume about 8 GB of memory, with four such sessions pushing cache alone to around 32 GB. By integrating TurboQuant, QVAC Fabric aims to let laptops, phones, e