Technology•

Tether's QVAC Project Grows Into an On-Device AI Toolkit as an 800 Million Parameter Agent Model Arrives

Tether is expanding QVAC, its open-source platform for running AI on phones and laptops, adding the 800 million parameter BotPsy agent model released November 11.

Written By :MINRK
Tether's QVAC Project Grows Into an On-Device AI Toolkit as an 800 Million Parameter Agent Model Arrives

Photo by Sumaid pal Singh Bakshi on Unsplash

Mentioned in This Article

USDT
USDT
Tether
$1.00
0.01%
BTC
BTC
Bitcoin
$59,027.00
2.12%

1. Stablecoin issuer pushes deeper into local AI

Tether, the company behind the USDT stablecoin, is building out QVAC, an open-source artificial intelligence ecosystem designed to run models directly on phones, laptops and desktops rather than in remote data centers. The latest addition to Tether QVAC is BotPsy, an 800 million parameter model for agentic tasks released on November 11, 2026, which scored 44.0% on the BFCL v4 tool-calling benchmark.

The project brings together a developer toolkit, a framework for customizing models on consumer hardware and a growing family of compact models. Its unifying idea is that useful AI does not have to depend on a connection to a cloud provider, and that the devices people already own can do much of the work themselves.

For a company best known for issuing the most widely used dollar-pegged token in crypto, the effort represents a substantial bet on a different technology. It also links the two businesses, because Tether is connecting its AI tools to software that can send and receive digital money.

2. What is Tether QVAC?

Tether QVAC is an open-source platform for running and customizing artificial intelligence models locally on consumer devices. Instead of sending a request to a server farm and waiting for an answer, an application built with QVAC processes the request on the user's own hardware, whether that is a smartphone, a laptop or a desktop computer.

Tether first presented the project in 2025 with a clear emphasis on local-first AI, and since then it has gradually turned that concept into concrete developer tools. The company has released core components under permissive open-source licenses, which means outside developers are free to inspect the code, modify it and build their own products on top of it.

That openness is meant to encourage a community of builders around the platform. An ecosystem where independent developers can extend the tools has a better chance of spreading across many applications than one controlled entirely by a single company.

3. Why a stablecoin company is building AI tools

Tether's core business is issuing USDT, a token designed to hold a value of one US dollar and backed by reserves that consist largely of short-term government debt and other liquid assets. USDT is the largest stablecoin by market value and is used heavily across crypto exchanges, payments and cross-border transfers, particularly in markets where access to dollars is limited.

The interest earned on those reserves has made Tether one of the most profitable companies in the digital asset industry. In recent years it has used that income to invest well beyond stablecoins, backing ventures in areas such as Bitcoin mining, energy, communications and artificial intelligence.

QVAC fits into that pattern of diversification, but it is also closely tied to the company's original business. Tether's long-running pitch for USDT emphasizes giving people direct control over their money without relying on intermediaries. Local AI applies a similar philosophy to computing, keeping models and data on the user's own device instead of routing them through large technology platforms.

4. The QVAC SDK and its inference runtime

A central piece of the platform is the QVAC SDK, which Tether launched earlier in 2026. A software development kit, or SDK, is a bundle of code libraries, documentation and tools that lets developers add specific capabilities to their applications without building everything from scratch.

At the core of the SDK sits a cross-platform inference runtime. Inference is the stage at which a trained AI model actually produces an output, such as answering a question, summarizing a document or deciding which action to take. A runtime is the engine that carries out that work efficiently on a particular piece of hardware.

The goal is to let a single model run on laptops, desktops and phones without developers having to rebuild it separately for each type of device. Anyone who has shipped software across several operating systems knows how much effort that kind of portability can save. Tether has highlighted personalization and local execution as the main design priorities for the toolkit.

5. QVAC Fabric brings fine-tuning to consumer hardware

Another component, QVAC Fabric, focuses on fine-tuning, the process of taking a general-purpose model and specializing it for a particular task, domain or user. Fine-tuning is how a broadly capable model can be adapted to handle, for example, a company's internal terminology or an individual's writing style.

Fabric is designed to perform that customization on consumer hardware, including the graphics processors found in mobile devices. Historically, fine-tuning has required powerful servers equipped with expensive data center chips, which put it out of reach for many developers and users.

The framework relies on a technique known as LoRA, short for Low-Rank Adaptation. Rather than retraining every parameter in a model, LoRA adjusts a small set of additional parameters layered on top of the original. That dramatically reduces the computing power and memory needed, which is what makes fine-tuning on a phone or laptop practical in the first place.

6. BotPsy targets agentic tasks with a small footprint

On November 11, 2026, Tether released QVAC BotPsy, a model with 800 million parameters tuned specifically for agentic work. Parameters are the internal values a model learns during training, and their number is a rough measure of a model's size. At 800 million, BotPsy is small compared with the largest models in use today, which is precisely what allows it to run on everyday devices.

Agentic AI refers to systems that do more than hold a conversation. An agent can call software tools, take actions on a user's behalf and complete tasks that involve several steps, such as finding information, filling in a form and then confirming the result. Doing that reliably requires a model to understand a request and translate it into the right sequence of function calls.

In Tether's initial benchmark results, BotPsy scored 44.0% on BFCL v4, a test that measures how well a model selects and executes software functions. The company also reported that the model outperformed a rival roughly 2.5 times its size. These figures come from Tether's own early testing and have not been independently verified, so they are best treated as a first indication rather than a settled verdict.

7. Why smaller models matter for on-device AI

The release of BotPsy reflects a broader argument that bigger is not always better. Much of the attention in AI has focused on ever-larger models trained on vast clusters of chips. Those systems can be extremely capable, but they are costly to run and generally require a connection to a remote server.

Compact models trade some raw capability for efficiency. They need less memory and less processing power, which allows them to fit on a phone or laptop and respond without a network connection. For many specific jobs, such as operating an app or handling a defined set of tools, a well-tuned small model may be good enough, and sometimes better suited, than a much larger general-purpose system.

Local execution also has implications for privacy. When a model runs on the device, the data it processes does not need to leave that device. Users and developers who are wary of sending sensitive information to third-party servers may find that approach appealing, particularly for personal assistants that handle messages, documents or financial details.

8. Linking AI agents to Tether's wallet tools

QVAC is connected to Tether's Wallet Development Kit, a set of tools for building cryptocurrency wallets that supports transactions in Bitcoin (BTC) and Tether's USDT. The integration means that an AI assistant built with QVAC could, in principle, hold funds and send payments directly from a user's device.

The strategy is to embed payments into AI agents from the outset. If software agents are going to buy services, pay for data or settle with other agents, they will need a way to move money. By pairing its AI tools with its wallet software, Tether is positioning its own stablecoin as a natural choice for that role.

Tether has described an ambition of trillions of agents running on everyday devices, and it has floated potential uses ranging from robotics to brain-computer interfaces. Those goals are long-term aspirations rather than products available today, but they show the scale at which the company is thinking about the intersection of AI and digital money.

9. What the QVAC push means for Tether and developers

For Tether, QVAC is a way to extend its reach beyond stablecoin issuance into the infrastructure of AI. If agents that act and transact on behalf of users become common, they will need a settlement asset, and Tether wants USDT built into the toolkit developers reach for by default.

For developers, the open-source licensing and cross-device design lower the barriers to building local AI applications. A team can experiment with the SDK, customize models with Fabric and deploy compact agents such as BotPsy without paying for cloud computing every time a user makes a request. Applications built this way can also keep working in places where network coverage is weak or unavailable.

The next tests for the project will be adoption and independent validation. Developers will look for evidence that the tools perform as advertised across a wide range of hardware, and outside benchmarks of models like BotPsy will show whether Tether's early results hold up. How many applications choose to build on QVAC, and whether any of them make use of its built-in payment capabilities, will indicate whether the company's bet on on-device AI gains real traction.

Related Articles

NEWSLETTERS

Don't miss another story.

Subscribe to the MINRK Newsletter today.

Tick the newsletters you want. No selection? You'll get MINRK Headlines.

By signing up, you will receive emails about MINRK products and you agree to our terms of use and privacy policy.

MINRK Daybook

Market analysis for crypto traders and investors.

EVERY WEEKDAY

MINRK for Advisors

Defining crypto, digital assets and the future of finance for financial advisors.

EVERY THURSDAY

MINRK Protocol

Exploring the tech behind crypto one block at a time.

WEEKLY

MINRK Long & Short

A must read for institutions. Insights, news and analysis delivered weekly.

EVERY WEDNESDAY

MINRK Headlines

The biggest crypto news and ideas of the day.

EVERY WEEKDAY

MINRK Policy

Examining the intersection of cryptocurrency and government.

WEEKLY

MINRK Research

Join thousands of readers who rely on MINRK for data-driven insights on the latest digital asset trends.

MONTHLY