A $25M vote of confidence in a primitive
This week, Portal to Bitcoin raised $25M to build an "atomic OTC desk" — HTLC-based, non-custodial, cross-chain settlement aimed at institutions and large block trades, with Coinbase Ventures, OKX Ventures, and Arrington Capital among the backers.
If you've been building in this space, the interesting part isn't the competitive framing. It's the validation. The thesis that atomic, trust-minimized settlement is real infrastructure — not a whitepaper, not a research curiosity — just got capitalized by people who price these things for a living. That's a good day for everyone working on the same primitive.
It also makes this a good moment to be precise. "Atomic" is one of the most over-used words in crypto, and a funded category is exactly when the word starts getting stretched. So: what does atomic actually require, and what does an agent-native version of it add?
What "atomic" actually requires
Atomic has a hard technical meaning, borrowed from databases: a set of operations either all complete, or none of them do. There is no intermediate state where the system is half-done.
For a cross-chain trade, that means: either both legs settle — you get the asset you bought and the counterparty gets theirs — or neither does, and both sides keep what they started with. Critically, there is no window where one party has delivered and the other hasn't. That window is where counterparty risk lives, and "atomic" is precisely the claim that the window doesn't exist.
A Hashed Timelock Contract (HTLC) enforces this with two ingredients:
- A hashlock. Funds on both chains are locked against the hash of a secret. Revealing the secret to claim one side mathematically exposes it, which lets the counterparty claim the other side. One reveal unlocks both legs.
- A timelock. If the secret is never revealed, both locks expire and both sides refund automatically. The failure mode is "everyone gets their money back," not "someone is stuck."
The result: the trade either fully completes or fully unwinds. No bridge minting a wrapped IOU. No custodian holding both sides "for a moment." No relayer you have to trust not to run. The guarantee is structural, not reputational.
This is the part Portal and Hashlock Markets genuinely agree on, and it's worth saying plainly: HTLC-based settlement is the honest version of cross-chain trading. Most "atomic swaps" you see advertised are actually optimistic bridges — lock-and-mint with a challenge period — and the difference matters the first time a challenge period gets gamed.
Where the word gets stretched
Three things commonly get marketed as "atomic" that aren't:
- Optimistic bridges. Funds are locked, a wrapped asset is minted, and a fraud window protects you — until it doesn't. The honeypot is the locked collateral.
- Custodial settlement. A trusted intermediary holds both sides and releases them together. The atomicity is real, but it's enforced by the custodian's solvency and honesty, not by cryptography.
- Fast sequential routing. Two transactions that happen quickly are still two transactions. If the second reverts, the first already happened.
A real atomicity claim should survive the question: what happens if the most adversarial party in the trade does the worst possible thing at the worst possible time? HTLCs answer that with "everyone refunds." Most of the alternatives answer it with "you file a support ticket."
Two directions from the same primitive
So Portal and Hashlock Markets start from the same place — HTLC, non-custodial, cross-chain — and then point it at different users.
Portal is building for institutional desks. Humans, large block trades, a BTC-centric federation model, an L3 for scaling. That's a coherent bet: there's real money in moving large blocks without a custodian, and institutions are the buyers who feel custody risk most acutely.
Hashlock Markets points the same primitive at agents. The settlement layer is exposed as an MCP server — six tools — that an AI agent calls directly. There's no desk, no portal to log into, no human in the settlement loop. An agent constructs a request-for-quote, receives sealed bids, and the winning bid settles over an HTLC — RFQ and settlement fused into one operation.
The difference isn't the cryptography. It's the caller. Portal's caller is a trader at a terminal. Our caller is a process — a LangChain or CrewAI or AutoGen agent that needs to exchange value with another agent across chains and cannot, by construction, trust it.
That's why "agent-native" isn't a marketing skin on top of an OTC desk. The interface is the product decision. An OTC desk assumes a human reads quotes, judges counterparties, and clicks. An MCP tool assumes a model does all of that programmatically, which means the counterparty model, the quote format, and the settlement call all have to be machine-legible from the start.
What agent-native settlement adds
Three things, concretely:
- A callable interface. The MCP server (hashlock-tech/mcp, scoped on npm) gives an agent the settlement primitive as a tool, not a website. This is the difference between "an agent could theoretically use this" and "an agent can use this in the next token."
- Sealed-bid RFQ. Price discovery without front-running. The agent broadcasts what it wants, makers bid blind, and the best bid settles atomically. No orderbook to snipe, no mempool to watch.
- Multi-chain scope. ETH is live end-to-end on the backend today. SUI contracts are deployed and CLI-tested; BTC is validated on signet with mainnet pending. The roadmap adds Base, Arbitrum, Solana, and TON. The cross-chain examples in our docs — say, a three-leg BTC→ETH→SUI route — are illustrative of the primitive's shape, not a claim that every leg is customer-ready today. We'd rather be precise about that than stretch the word "live."
The honest limitations
Agent-native settlement is not free of tradeoffs, and a funded competitor is a good reason to be candid about them:
- HTLCs cost latency. The hashlock/timelock dance is slower than a custodian saying "done." For large institutional blocks where settlement finality in minutes is fine, that cost is invisible. For high-frequency agent activity, it's a real design constraint.
- Liquidity has to show up. Sealed-bid RFQ only works if makers bid. A settlement primitive doesn't manufacture counterparties; it makes the trade safe once they exist.
- The agent economy is early. We'll say the quiet part plainly: the rails are ready, the trains are still arriving. We're not going to claim agent customers we don't have. What we can claim is that when agents do need to settle cross-chain trustlessly, the primitive is built, audited-hardened, and callable.
Portal's $25M says the institutional-human version of this market is real enough to fund. Our bet is that the agent-to-agent version is the larger surface over a longer horizon — and it needs the same atomic guarantee, just exposed to a different kind of caller.
The open question
Here's what we're actually unsure about, and we'd like the answer from people building agents in production:
As atomic settlement becomes a funded category, who ends up calling it most — a human on a desk, or an agent through an API?
If you're building agent infrastructure and have a view, reply here or open an issue on the repo. We read all of it.
Hashlock Markets — the atomic settlement layer for the agent economy. Sealed-bid RFQ plus HTLC settlement, no bridges, no custodians.
- Protocol: https://hashlock.markets/?utm_source=devto&utm_medium=social&utm_campaign=2026-05-15-atomic-otc-validation
- GitHub: https://github.com/Hashlock-Tech/hashlock-mcp
- Full protocol spec: papers.ssrn.com/sol3/papers.cfm?abstract_id=6712722