This is the Friday product note - less about the mechanism in isolation, more about the decision an agent actually faces.
All week the argument in the agent-economy timeline was about settlement: when two agents strike a trade, does it complete through a custodian holding both sides, or atomically with no one holding either? Collateral is where that argument stops being abstract. The moment an agent has to post something to back an obligation - a forward to deliver, a loan to repay, a leg in a multi-step trade - it has to answer a very specific question: after I post this, who can move it?
For Bitcoin, the usual answer is "someone other than your agent." That is worth slowing down on.
The easy path quietly re-custodies your collateral
Bitcoin is the deepest collateral asset in crypto, and the obvious thing to want is to post native BTC. The problem is that Bitcoin the chain cannot run the contract logic a collateralized position needs - no expressive contracts, no objects, no state machine watching the position over time. So the collateral has to be usable on a chain that can reason about it, while the BTC lives on a chain that can't.
The standard fix is to wrap it: hand the real Bitcoin to a custodian, federation, or bridge contract, and receive a token on the destination chain. Now your agent can use that token as collateral anywhere that chain's contracts run.
It has also, without meaning to, done three things:
- Converted a bearer asset into a claim against an intermediary.
- Contributed to a honeypot - the locked BTC pile behind every wrapped token is a single standing target.
- Made its collateral's integrity depend on something it has no good way to audit: the ongoing solvency and honesty of whoever holds the real coins.
This is the same trap the week's settlement debate kept circling. You can remove the custodian from the trade and still let one walk back in at the collateral - and an agent that can't tell the difference is exactly the agent that gets hurt by it. A trustless settlement layer that sits on custodial collateral isn't trustless; it has just moved the trust somewhere the agent stopped looking.
The product: post the Bitcoin, keep the Bitcoin
A collateral vault takes the opposite trade. The BTC never leaves Bitcoin and is never represented by a minted token. It stays as native BTC, locked in a script on the Bitcoin chain. What crosses chains is one piece of information - the hash of a secret - not the asset.
From the agent's side, the shape is simple enough to reason about as a few tool calls rather than a pile of cross-chain plumbing:
- The collateral is locked into a P2WSH output on Bitcoin whose redeem script is a hash-time-lock: a hashlock path spendable by revealing a preimage, and a timelock path spendable by the depositor after a block-height deadline.
- The obligation it backs lives on Sui, where our Move contracts are deployed. That contract is written so the outcome of the obligation controls who learns the preimage, and when.
- Bitcoin Script can't read Sui state and Sui can't read Bitcoin's UTXO set, but both can agree in advance on one hash. The preimage of that hash is the single key that resolves the position on both chains at once.
A vault has exactly three ways to end, and all three are mechanical:
- Perform. The obligor delivers - reveal-to-claim publishes the secret, and the BTC settles to whoever the performance entitles.
- Default. The hashlock path routes the BTC to the counterparty as the agreed remedy. No chasing anyone through courts or reputation.
- Silence. If the position is abandoned, the timelock returns the BTC to the depositor after the deadline - enforced by Bitcoin consensus, not by anyone's cooperation.
In every path the BTC was native Bitcoin the whole time, and no third party ever had the unilateral ability to move it. That is the entire product claim, and it is a narrow, checkable one.
What it deliberately is not
A vault is not a margin account, and pretending otherwise would be the easy oversell. Bitcoin Script has no price feed, so a vault cannot do a continuous margin call - it resolves on performance, default, or a deadline, not on a price crossing a threshold. For discrete obligations (a forward with a fixed date, a fixed-term loan, a bonded commitment) that's fine, because the obligation is discrete too. For a position that genuinely needs real-time mark-to-market, this is the wrong tool, and we'd rather say so than smuggle an oracle into the design.
The honest status, stated plainly
Because this is a product note and not a launch, the status report matters more than the pitch:
- The Bitcoin P2WSH HTLC is validated on signet - Bitcoin's test network - with mainnet pending.
- The Sui contracts are deployed and CLI-tested, with gateway wiring in progress.
- The only place atomic settlement is live end-to-end today is Ethereum mainnet.
So a BTC collateral vault is a design we're building toward, exposed through our MCP server (hashlock-tech/mcp, scoped - six tools) so an agent reasons about the position instead of the script-level mechanics. MCP is the open protocol Anthropic introduced for connecting models to external systems; a vault is just another set of tool calls on that surface. But it is not a button you can press this morning, and the npm package is at a 0.4.x patch line - no new version to announce. The coarse part of the tradeoff is real, too: Bitcoin's ~10-minute blocks make timeout windows coarse, the cross-chain deadlines have to be conservative by construction, and collateral sits locked and idle for the life of the position. Those are the costs of deleting the custodian. We think they're worth paying for the class of positions a vault is built for. They are still costs.
The question
Wrapped BTC made Bitcoin usable everywhere by making it custodial everywhere. The bet behind a collateral vault is that an agent can keep most of that usefulness - Bitcoin backing an obligation on a chain that can actually reason about it - without the custodian, by moving a hash instead of the coins.
So, for anyone building agent-side collateral logic: when your agent posts Bitcoin as collateral, do you actually know who can move those coins? If the honest answer is anything other than "only the agent, and only along paths it agreed to in advance," what is that trust buying you?
Hashlock Markets - atomic settlement for the agent economy. Sealed-bid RFQ + HTLC settlement, fused into one operation. No bridges, no custodians.
- Protocol: https://hashlock.markets/?utm_source=devto&utm_medium=referral&utm_campaign=2026-06-26-btc-collateral-vaults
- MCP server (source): https://github.com/Hashlock-Tech/hashlock-mcp
- The underlying mechanism design is written up in a working paper on SSRN: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6712722