Tether offers open-ended grants for local AI and self-custody

Tether launched an open-ended developer grants program that will pay $1,500 to $4,000 per completed task to build local-first artificial intelligence, self-custodial wallet tools and components for its QVAC platform. Payments will be made in USDT or Bitcoin after specific technical deliverables are met. The company said there is no cap on total program payouts and it has set aside roughly CHF 5 million (about $5.38 million) for support through 2030.

The program will fund work on local AI, payments infrastructure and decentralization research. Priority areas include building core libraries for QVAC, producing technical documentation, developing applications on Tether’s open stack and researching edge AI. A central target is a Wallet Development Kit intended to let developers embed self-custodial wallets into apps so users can manage accounts and transact without relying on custodial services or hosted APIs.

Paolo Ardoino, Tether’s CEO, described his shift from cloud-based AI to local, self-sovereign systems and warned about security risks from emerging AI agent tools. He framed the grants with the statement: “If you can build something that runs locally, holds value directly, and doesn’t rely on external providers, we’ll fund it.”

Tether has previously funded related projects, including consecutive $100,000 grants to the BTC Pay Server Foundation and a $250,000 donation to OpenSats for Bitcoin development. The company said it has also issued more than 500 student education grants. The new program is presented as part of ongoing support for offline-first AI and payments tooling and for research into smaller models and edge computing.

Tether’s in-house AI Research Group released QVAC MedPsy, two medical language models designed to run on smartphones and wearables without internet access. The smaller model, QVAC MedPsy-1.7B, scored 62.62 across seven closed-ended medical benchmarks. The larger QVAC MedPsy-4B scored 70.54 on the same benchmarks and scored 58.00 on a clinical reasoning test called HealthBench Hard, compared with 42.00 for a roughly 27 billion-parameter model. Tether said the QVAC models use up to 3.2 times fewer tokens than comparable systems, which it said reduces response time and device compute demand.

Under the program, developers are paid per completed task and may choose payment in USDT or Bitcoin. Tether said the grants will run in multiple phases with funding reserved through 2030.