Trading is temporarily paused. More info here. THORChain has no active refund, airdrop, or compensation program. Be cautious of fraudulent websites impersonating the protocol and report them back to us.

NanoGPT Coming to THORChain: Every AI Model, No KYC, Paid in Crypto

Raynalytics logo
Ray

2026-06-13 — 10 min read

    Podcast
THORChain x NanoGPT Ecosystem Podcast #208 thumbnail featuring Milan Dereede discussing NanoGPT's integration with Kenton and Patriotsounds.

THORChain x NanoGPT Podcast #208 ft. milan_dereede, KentonC137 & Patriotsounds | June 13, 2026 | Watch the full episode on YouTube

By Raynalytics

TL;DR

  • THORChain is paused while a fix ships. v3.19.1 patches a bug in the new key-signing and key-share process, and once node operators adopt it, trading can resume within days, with the churn as the next hurdle.
  • NanoGPT, an AI aggregator that routes to roughly 87 providers and effectively every major model across text, image, and video, has integrated THORChain. No accounts, no KYC, pay per use in crypto.
  • The team came to THORChain after a centralized exchange froze their account over a small Dogecoin payment whose coins had once touched an Iranian exchange.
  • Co-founder Milan Dereede is a former Dutch Central Bank crypto lead who grew NanoGPT from a Telegram bot. The THORChain integration was finished by his co-founder on the flight home from Bitcoin Vegas.
  • On the table: a possible Nano ($XNO) pool, which turned into a clear walkthrough of THORChain's shift toward protocol-owned liquidity. Privacy coins (Monero, Dash, Zano, Zephyr) remain the near-term roadmap.

A small Dogecoin payment, around $32, was all it took to freeze NanoGPT's business. A centralized exchange suspended the account because those particular coins had, at some point in their history, touched an Iranian exchange. Bills kept arriving while the money sat locked. For a no-KYC AI service run by a former central banker, it was the clearest possible argument for permissionless settlement, and it is a big part of why NanoGPT integrated THORChain.

This Saturday episode brought together Milan Dereede, co-founder of NanoGPT, with hosts Denny and Kenton. The three first met at Bitcoin Vegas, and the conversation ran from what NanoGPT actually does, to how the THORChain integration came together on a plane, to whether Nano itself could one day get a pool. First, though, the hosts opened with the update the community is waiting on.

1. Where THORChain Stands: The Restart

At recording time, THORChain was paused, with no trading on the network. Kenton walked through the path back. v3.19.1, which fixes a bug in the new key-signing and key-share process, was expected to ship at the start of the week. Once it is out, the expectation is that node operators adopt it quickly and trading could resume within a few days. Chad Barraford had hoped to have it ready by the previous Friday, and the work has continued since.

The next hurdle is the churn. Churns on THORChain have a mixed history: sometimes they wrap up in hours, sometimes they take days. Once the network churns successfully, things should move quickly toward resumed trading, with a hopeful target of the end of the following week. (NanoGPT had been routing some swaps through sister protocol Maya Protocol as a backup, though that is paused on their side for now too.)

Individual chains have been paused or halted before, but a full-network pause is rare.

"This is definitely a first. And so hopefully this is the last." (Kenton)

In roughly five years of mainnet, THORChain has paused single chains, but never the entire network.

2. Meet NanoGPT: Every Model, No KYC, Pay As You Go

NanoGPT is an AI aggregator. Rather than running its own models, it routes to roughly 87 providers and gives users access to essentially every model on the market across text, image, and video, from closed-source labs like OpenAI, Anthropic, Google, and xAI through to the full open-source field. There are no accounts and no KYC, you pay per use in crypto, and you can stay anonymous. Monero is the most-used coin on the platform.

Two features stood out. The first is the auto-model: you type a prompt, a fast model (currently Gemini Flash) classifies what kind of request it is and how hard it is, and NanoGPT routes it to the cheapest model that can handle it. The second is a coding router, where you set how capable the model needs to be and the service picks the cheapest option scoring above that bar, so the same level of intelligence gets cheaper over time as new models arrive. Beyond model access, the team ships constantly, with recent additions including on-device PII redaction, verifiable trusted-execution models, one-click local models, the x402 accountless payment protocol, and document workspaces.

Because NanoGPT pays per call through provider APIs rather than splitting subscriptions, Milan said pricing generally lands at or below what you would pay a provider directly. That also turned out to appeal to businesses, who get one integration, centralized billing, and access to every model for every department. The multi-provider setup is about resilience too. Milan pointed to the recent Claude Fable 5 situation, where access was pulled abruptly with no warning for everyone outside Anthropic, with the longer-term picture unclear and possibly limited to the US. Fable 5 had quickly become the most-used model on the service, and because NanoGPT routes the same model across providers like AWS, Azure, and Google Vertex, a single outage usually does not take a model offline, even if in Fable's case every provider pulled it at once.

What stuck with Milan most was a smaller story. One early user in Kenya used the service for medical questions because the realistic alternative was not a doctor.

"My alternative is a shaman in my village." (a user in Kenya, via Milan)

3. From the Dutch Central Bank to NanoGPT

Milan's path into crypto started around 2012, through an online strategy game called Battle Dawn, where players needed to move money across borders to pay each other and someone suggested Bitcoin. He later joined the Dutch Central Bank, eventually becoming its crypto lead and working on crypto policy, before leaving, frustrated, after years of arguing two incompatible worldviews.

"They liked the idea of blockchain and not the idea of actually decentralized crypto." (Milan)

For much of that time he was anonymous on crypto Twitter as Mira Hurley, a persona he kept separate from his day job. Then, in late 2022, ChatGPT launched and he was hooked. To get around the subscription required for the best models, he asked ChatGPT itself to help him build a Telegram bot where anyone could deposit a little crypto and send a few messages. The bot found an audience (Erik Voorhees was an early user), his co-founder Huggy reached out from the Nano community, and the two built it into a website. That was the origin of NanoGPT.

4. The $32 That Froze the Business

NanoGPT still pays many suppliers in fiat, since OpenAI and Anthropic do not yet accept crypto, so the team routinely sends crypto to a centralized exchange to convert it. One of those payments, a small amount of Dogecoin, triggered an immediate account suspension. The exchange eventually explained that the specific coins had, somewhere in their history, touched an Iranian exchange, and the team had to prove they were not laundering Iranian money.

"It's literally $32 in Dogecoin." (Milan)

The account stayed frozen for a while, with money locked inside and bills still arriving. It is exactly the failure mode THORChain is built to remove: there is no central party that can suspend an account over the prior history of a coin. That is what made the fit click for the NanoGPT team.

5. Vegas, the F1 Arcade, and a Plane-Wi-Fi Integration

The partnership traces back to Bitcoin Vegas. Invited to a black-tie rooftop speaker event, Milan and Huggy instead went to THORChain's side event at the F1 Arcade, where, by all accounts, Milan won the racing. The dinner that night put NanoGPT at a table with Mark Mason of Bitcoin Magazine, Joel from Dash, the Edge Wallet team, and Chad.

"Everyone there actually cared about crypto and having it work." (Milan)

Kenton handed over a THORChain API key, and Huggy started integrating on the flight home, over the plane Wi-Fi. By the time the teams spoke a couple of weeks later, it was already done. Milan said the documentation was clear and the integration was straightforward, which prompted Denny to joke that "you can integrate our API on a plane using the plane Wi-Fi" belongs in the marketing. The takeaway for the hosts: side events work.

6. Could Nano Come to THORChain?

A good chunk of the conversation explored whether Nano ($XNO), the coin NanoGPT was originally built around, could get its own THORChain pool. Nano is unusual: it is account-based rather than UTXO, has no smart contracts and no memo field, settles in roughly 400 milliseconds, and is feeless. Milan said running a node costs about $10 a month, so it is extremely lightweight. It also has no decentralized-exchange presence today, only centralized exchanges, which is part of the appeal of changing that.

"You don't want to be reliant on a centralized exchange for people to get into Nano or out of Nano." (Milan)

Two hurdles came up. The first is who seeds the initial liquidity: Nano is fully decentralized, with no foundation or developer fund, so a pool would likely need to be community-funded. The second is the development work to support the chain, which Kenton noted other teams have handled in a few weeks once they fit into the pipeline.

That led into a clear explanation of where THORChain liquidity is heading. Under the protocol-owned liquidity model Kenton described, liquidity providers stop receiving the yield, and that revenue instead flows to the protocol, which becomes a permanent liquidity provider itself and directs capital toward the pools with the highest fees relative to their depth. A small, high-volume Nano pool could attract exactly that kind of protocol-owned liquidity over time. In Kenton's framing, the community could seed something like $25,000 to $50,000, accept some impermanent loss, and later withdraw, leaving behind a permanent pool on a decentralized exchange. For users, rather than navigating multiple centralized exchanges, getting Nano would be roughly one click on the swap interface.

As for the no-memo problem, memoless swaps, which router v6 is expected to extend, offer a path. Nano already identifies payments using tiny amounts appended to a transaction, much the way THORChain's memoless swaps use a precise amount to tag a swap.

7. What to Watch

  • The restart: once v3.19.1 is adopted and the network churns, the team is hopeful trading resumes by the end of the following week.
  • Privacy coins: Monero is already NanoGPT's most-used payment coin and a THORChain priority, Dash is expected to be one of the next main chains, and Zano (finishing its gateway nodes) and Zephyr are also in the conversation.
  • A possible Nano ($XNO) pool, pending community-seeded liquidity and the development work to support the chain.
  • NanoGPT's Partners product (nano-gpt.com/partners) lets any business embed AI models and outsource the payment flow. Denny floated point-of-sale apps tapping it, the same lane THORChain-settling apps like Bior Labs and Moca are building in.
  • A new year-long marketing deal with the Sal and Mark Show (hosted by Sal the Agorist and Mark Edge), aimed at a privacy-minded audience, is set to begin once trading is back.
  • September in Amsterdam: DashCon and Common S3nse (formerly ETHDam) run back to back, by which point the team hopes Monero swaps will be live.

More THORChain data, check out raynalytics.net

Follow Raynalytics for more Weekly Analytics and Podcast recaps.

Try the World’s Leading Bitcoin DEX

No sign up required. Easy to use.