The Challenges of Self-Sovereign Open Source AI - Marco Chiappetta (QVAC) - ETH Tbilisi 2.2

The full ETH Tbilisi Season 2 third event — QVAC: Self-Sovereign Open Source AI, the biggest challenge of this decade with Marco Chiappetta from QVAC.

Your AI Doesn't Belong to You. Here's How to Fix That. 🔥

Every time you use ChatGPT, Gemini, or Claude, your prompts, your data, your thoughts travel to a data center owned by someone else. A corporation decides what you can ask. A terms of service agreement decides what happens to your input. A centralized server decides if the service even stays online.

You don't own your AI. You rent it. And the landlord can change the rules whenever they want.

This is the default today: closed-source models, gated APIs, usage limits, content filters decided by committees you'll never meet, and a kill switch that sits in someone else's hands. Your most intimate questions, your business logic, your creative process, all of it routed through infrastructure you don't control and can't audit.

Now imagine the opposite. AI that runs on your phone. On your laptop. Offline, if you want. No API keys. No data leaving your device. No one between you and your own intelligence. Open-source, peer-to-peer, and unstoppable.

That's what QVAC is building.

QVAC is Tether's open-source AI platform, and it represents a fundamentally different philosophy: intelligence is not a service to rent, it's a resource to own. Built on a local-first, peer-to-peer architecture, QVAC lets developers run LLMs, speech-to-text, translation, vision, and more directly on consumer hardware across every major operating system, from a single codebase. No cloud dependency. No central point of failure. No gatekeeper.

The SDK launched in April 2026 and is already fully open-source on GitHub. It supports text generation, embeddings, multimodal inputs, OCR, and delegated inference through peer-to-peer swarms. Models are distributed through the Holepunch protocol stack, the same decentralized infrastructure behind Keet. AI agents built on QVAC can even transact autonomously using Bitcoin and USDt through Tether's Wallet Development Kit.

Why This Matters for the Agent Economy

Ethereum is building the infrastructure for autonomous AI agents. ERC-8004 gives agents on-chain identity, reputation, and validation through three registries, making them discoverable and trustable across organizational boundaries without pre-existing trust.

But here's the problem nobody talks about: what's the point of giving an agent an on-chain identity if you can't verify what the agent is actually doing?

If your agent's brain runs on a closed-source cloud, its on-chain identity is just a label on a black box.

Open-source, local AI fixes this. When the model is open, the code is auditable, and the inference runs on hardware you control, the on-chain identity actually means something. ERC-8004's reputation and validation registries become meaningful because you can verify not just who the agent claims to be, but what it's actually computing.

This is the full stack of self-sovereign AI: QVAC for the brain and ERC-8004 for the identity.

The Speaker 🧑‍💻

Marco Chiappetta — Software developer building QVAC at Tether. CS background with deep experience in backend infrastructure and distributed systems. Now working on Tether's open-source local AI platform, turning the vision of self-sovereign intelligence into working code. X: https://x.com/elchiapp

From QVAC by Tether (@qvac) — an open-source, cross-platform AI framework for building local-first, peer-to-peer AI applications. QVAC supports LLMs, speech, vision, translation, and more, running on any device without cloud dependency. https://qvac.tether.io | https://x.com/qvac

Timestamps:

00:00 - Intro
02:28 - The issue of TRUST in AI
06:51 - Main Challenges in building Open Source AI
13:11 - The role of ERC 8004 and Open Source LLMs
17:17 - How QVAC works
21:10 - What is the future of Open Source AI?
22:35 - Are local models competitive to Cloud-hosted LLMs?
30:00 - Is AI net positive for humanity?
35:44 - Real-world usages of QVAC
40:18 - Is AI better than doctors?
42:38 - Is Distributed Training possible?
45:23 - Have we reached AGI?
48:17 - How do you maintain the integrity of open source in the AI race?
50:26 - How do you incentivize people to use local AI?
52:08 - Can Open Source AI be captured by regulators?
53:20 - AI should be designed to solve practical problems or address existential concerns?
54:53 - Have QVAC plans to supports builders?
56:08 - How to switch to QVAC?
01:00:02 - What are potential use cases of AI in blockchain?

Join Our Community:

📱 Telegram: https://t.me/ethtbilisi
🐦 Twitter: https://x.com/ethtbilisi
📅 Upcoming Events: https://luma.com/user/ethtbilisi

Supported By 🤝
ETH Tbilisi is made possible thanks to:

Glue - A new DeFi Primitive to transform any Token and NFT into a "Digital Asset Treasury"
https://glue.finance | https://x.com/glue_fi

Видео The Challenges of Self-Sovereign Open Source AI - Marco Chiappetta (QVAC) - ETH Tbilisi 2.2 канала ETH Tbilisi