The Vault Decade: How On-Chain Asset Management Evolved

Contents+

1. The right idea, the wrong constraints (2016 to 2022)
2. The standards: one made vaults composable, the other made them funds
3. The institutional wave: power, with a sales call attached
4. The permissionless answer
5. The ecosystem fills in
6. Vaults inside lending protocols: institutional, but bounded
7. What is still missing
8. Where T3tris fits

For ten years, the same idea has been rewritten in Solidity: pool capital, let a manager run a strategy, issue shares, let the share price carry performance. A fund, in other words. The idea was never the hard part. The plumbing was.

What follows is the short history of how on-chain vaults went from an elegant dead end to the infrastructure layer that now holds tens of billions of dollars and what remains unsolved.

The right idea, the wrong constraints (2016 to 2022)

The first serious attempt was Enzyme, founded in 2016 by Mona El Isa as Melonport, building the Melon protocol. It is one of the oldest live protocols on Ethereum, and it got the shape of the problem right years before anyone else: a vault contract, a manager who cannot touch the assets, policies that constrain what the manager may do, fee modules that pay them, and investors holding transferable shares.

The idea is sound because it gives a manager everything it needs to run a fund except the ability to run away with the assets.

The constraint was architectural. In the original design, a vault could only trade where the protocol had built an adapter, and only hold assets the protocol could price on-chain. Every venue was an integration, and before the v4 Sulu release in Q1 2022, adding one required a council vote.

That model is perfectly coherent, and it excludes most of the asset management industry. If your edge is a basis trade on Binance, an OTC book, a private credit deal, or anything that settles T+2, there was no vault for you. You could not price it on-chain, and you could not reach it through an adapter.

A real business, but never the rails for the industry it was designed for.

Meanwhile, "vault" acquired a second, louder meaning. Yearn launched in February 2020 and turned the word into a retail yield product: deposit, receive yTokens, let a strategy farm. It peaked at $6.91B in December 2021 and holds $192M today.¹ Yearn proved the share-accounting UX at scale.

The standards: one made vaults composable, the other made them funds

ERC-4626, created on 22 December 2021, standardized the tokenized vault: one underlying asset, one share token, a predictable deposit and redeem interface.² It ended the era of every protocol writing its own adapter for every yield token.

But 4626 encodes an assumption that no real fund satisfies. As ERC-7540 puts it plainly: "The standard is optimized for atomic deposits and redemptions up to a limit. This limitation does not work well for any smart contract system with asynchronous actions or delays as a prerequisite for interfacing with the Vault."³

ERC-7540, created on 18 October 2023, added the missing piece: requests. You request a deposit or a redemption, the manager settles the batch at a valuation point, you claim. The EIP names its intended users explicitly: real-world asset protocols, undercollateralized lending, cross-chain lending, liquid staking, insurance safety modules.

That is not a technical detail. Request-then-settle is the subscription and redemption cycle of a fund, expressed in a contract: a valuation point, forward pricing, no forced unwind to honour a withdrawal. Everything built after 2024 descends from it.

The institutional wave: power, with a sales call attached

Veda shipped BoringVault in March 2024 and became the biggest thing in the category. The design is minimal: a custody contract of roughly 100 lines, a Teller for deposits, an Accountant that computes the share price off-chain and pushes it on-chain, a Manager that checks every strategist action against a merkle tree of pre-authorised calls, and a solver-driven withdrawal queue.⁴

It is the strongest institutional product in the category, and it is not a tool a manager picks up alone. The entry points on Veda's site are "Contact sales" and "Looking for a vault design partner?", and terms are not published.

For an exchange that wants a yield product without building an operations desk, that is exactly right. For a manager who already has one, it means a partnership to negotiate and economics to share. The institutional wave solved the technical problem and left a commercial one: you could run a real strategy on-chain, provided you were large enough to be worth the conversation.

The permissionless answer

Lagoon went at exactly that gap. Same async machinery (ERC-7540, silos for pending capital, settlement by a curator) but a different distribution model: a permissionless vault factory, no approval, no minimum, and terms published rather than negotiated.

Those terms deserve stating plainly, because they are what the category actually charges: the protocol takes 10% of the fees the curator charges its own investors, with an on-chain ceiling of 30%.⁵ On a 2 and 20, that is 0.2% of assets and two points of performance every year.

And the rate is not the curator's to set: it lives in a protocol-level registry where changes take effect immediately, with no cooldown. It is cheaper and far more legible than a negotiated partnership. It is still a tax levied on the manager's economics rather than on the service the protocol provides.

One nuance matters: the factory is permissionless, but the front-end is curated. Only vetted vaults appear on the main interface. Deployment is open, distribution is not.

The ecosystem fills in

Upshift went after the custody-side institution: permissioned vaults that let Kraken Institutional clients put idle treasury to work without leaving qualified custody.

The trade-off is control, since Upshift computes or validates the NAV, the operator settles each epoch, and in an emergency an owner multisig can pause everything and sweep the funds.⁶ The curator does not hold the last word on its own book.

Midas does not sell infrastructure, it sells the finished product: a strategy wrapped into a regulated debt certificate that trades as a plain ERC-20, under a BaFin licence with MiCA passporting.⁷ mTBILL carries T-bill exposure, mBASIS a market-neutral basis trade.

The buyer is a creditor of the issuer rather than an owner of the assets, and redemption runs on the issuer's schedule behind a KYC gate. What that buys is distribution, since mTokens are accepted as collateral on Morpho and Pendle where a vault share is not.

IPOR came at it from another angle. It began as an on-chain interest rate benchmark, then turned that work into Fusion, a vault infrastructure for managers who do not write Solidity: one-click factory deployment, a Python SDK, around 40 integrated protocols.

But the allocation model is the older one: an Alpha "can only execute transactions for which the necessary fuses have been connected to the vault",⁸ assets never leave the vault for an external address, and positions are valued through price feeds.

It is Enzyme's original design, rebuilt properly, and it draws the line this whole decade turns on: either the vault holds the assets and constrains what can be done with them, or the curator holds them and reports what they are worth.

And Enzyme itself pivoted. Enzyme Onyx abandons the closed-universe model entirely. The vault keeps tokenisation, accounting and the investor registry, while execution moves to an external Management Wallet: any custody, any protocol, centralised venues included, with NAV reported at checkpoints rather than derived from live price feeds.⁹

And the field keeps filling in: Aera, Blok Capital, Concrete, Ember, Makina, Midas, Nashpoint, Orion, Rethink and Accountable are each building some layer of the same stack, from vault engines and issuance wrappers to the reporting and verification tools an allocator needs around them.

Vaults inside lending protocols: institutional, but bounded

The largest pools of vault capital today aren't in asset-management infrastructure at all. They're inside lending protocols.

Morpho turned curation into a product. First MetaMorpho vaults, where a curator allocates depositor capital across isolated markets under caps and timelocks, then Vaults V2 on 29 September 2025, adding adapters, flash-loan-powered in-kind redemptions, segregated duties and gates for accredited investors.¹⁰ Morpho sits at $10.41B TVL, an all-time high. Spark runs $6.86B, deploying DAO capital through vaults as an allocation venue.¹

These are institutional in governance and in size. But they are allocation layers: a curator decides how capital is spread across markets the protocol already supports. That is a different job from running a strategy on a CEX, in OTC, or against an RWA book, and it is why "vault" in a lending protocol and "vault" in a fund context remain two different products.

That distinction deserves its own piece, and it is the one we will take apart next: what separates a lending vault, which allocates inside a protocol's own markets, from a generalist vault that runs a fund.

What is still missing

Three things, after a decade.

1. Cost structure. Vault infrastructure is paid out of the manager's own economics, and the terms are rarely comparable.
2. Access. Enterprise deployment models select for size. A curator with $5M and a real track record is not a prospect for a sales team, and that curator is precisely who the industry needs to onboard next.
3. Strategy neutrality. Async settlement is what allows a vault to hold a CEX position or a private credit book without unwinding it to honour a withdrawal. Many "vaults" still assume the strategy lives on-chain and prices itself.

Where T3tris fits

T3tris starts from a different technical premise: its vaults are a hybrid of ERC-4626 and ERC-7540, so a curator can settle synchronously or through the request-and-settle cycle, depending on what the strategy needs.

The business model. Curators keep 100% of their management and performance fees: T3tris has no fee switch, and deploying a vault, pooling capital and running a strategy cost nothing. The protocol earns on idle capital instead, routing funds into money markets during the settlement window. A curator who wants that capital elsewhere can buy a premium and plug in its own connector, a Morpho vault or any ERC-4626, and T3tris then shares in the yield that connector generates, never in the fees. Everything else is optional: on-chain KYB with a verified curator badge, a white-label suite of SDKs and APIs to run your own front-end, and a marketplace of custody, compliance, NAV and risk providers paid by the providers rather than the manager. Staying out of the manager's fee chain is not only cheaper for the curator, it changes what the protocol is inside a regulated structure, which opens options that a fee-taking intermediary does not have.

Permissionless includes the interface. Most open factories are permissionless at the contract and curated at the front door: anyone can deploy, only approved vaults get listed. On T3tris a vault is public in the explorer the moment it exists. Discovery is not a favour the platform grants.

And the operator details that decide daily life:

• Fees are claimable out of deposit and withdrawal flows, so a curator settles without a separate transaction, without topping up a wallet, and without selling into slippage to cover them.
• NAV movement bounds are enforced by the contract, not by the reporting process, so a bad valuation is rejected on-chain.
• Deposit disclaimers live in the vault itself, shown and acknowledged before capital goes in, once or every time, at the curator's choice.
• The activity log sits with the performance chart: settlements, valuations and role changes are visible where the price is, each linked to its transaction.

Enzyme proved funds belong on-chain, then spent nine years removing its own constraints. ERC-7540 gave everyone the settlement cycle. Veda proved institutions will pay for the rails. Lagoon proved the factory can be open. What is left is the question none of them had a reason to ask: whether infrastructure needs a cut of the manager's economics at all. T3tris is built on the answer being no.

Sources

1. TVL series from DefiLlama, read on 18 September 2026. ↩ ↩²

2. ERC-4626: Tokenized Vaults, created 22 December 2021. ↩

3. ERC-7540: Asynchronous ERC-4626 Tokenized Vaults, created 18 October 2023. ↩

4. Veda documentation, BoringVault architecture. ↩

5. Lagoon documentation: fees, upgradability and opt-in proxy, v0.6.0. ↩

6. Upshift documentation: curator fund flows, risk and emergency procedures. ↩

7. Midas documentation. ↩

8. IPOR Fusion documentation: fuses and Alpha permissions. ↩

9. Enzyme, Onyx, and the Enzyme documentation. ↩

10. Morpho, Vaults V2: a new standard for asset curation, 29 September 2025. ↩