I designed a non-custodial wallet for the Hyperliquid ecosystem from scratch — then used the data to make crypto safe enough for real people to hold.
The challenge
The Hyperliquid ecosystem was growing fast, but it had no native wallet built for normal people. Self-custody is brutal: a lost seed phrase means lost funds, a wrong network means money gone forever, and gas and slippage read like a foreign language. Existing wallets dumped all of that risk on the user and called it “freedom.”
My brief: a non-custodial wallet that keeps users in full control of their keys — yet feels safe enough that a first-timer can hold, send and swap without fear of one irreversible mistake.
Discovery
I’ll lose the seed and lose everything
Seed phrases are the single biggest source of dread — and most people skip backing them up entirely.
One wrong send and it’s gone
Irreversible transfers — wrong network, wrong address, wrong amount — are the errors users fear most.
What even is slippage?
Gas, commission and slippage are opaque. Hidden costs at the final step destroy trust in a swap.
From 11 interviews with crypto users (novice to degen), a teardown of 7 wallets, and analysis of the support tickets and on-chain errors that cost people real money.
How I decided
Anything that can lose funds gets friction, confirmation and clarity.
Network, fees and slippage shown in plain terms, before you commit.
Sensible defaults up top; real control one tap away.
Key decisions
I shipped the first version, watched where people stalled or lost funds, and redesigned the risky moments. Every call below is mine — and each one moved a number.
Home
History
Hidden
Backup gate
Send
Confirmed
Swap
Slippage
Control
Receive
Self-custody doesn’t have to mean self-destruct. Good design is the safety net.
Impact
Measured over the first 8 weeks via Amplitude funnels, on-chain transaction analysis and support volume. The seed-backup gate and Send guardrails were the biggest movers on both safety and trust.
Reflection
Hard-coding one network kept v1 simple but bit us when Arbitrum landed. I’d model chains as first-class from day one.
Great against shoulder-surfing, annoying on every revisit. I added Face ID to keep security without the tax.
A pre-send transaction simulation would catch the scariest mistakes before they’re irreversible. I scoped it for v2.
Design system
Palette
Typography — Inter display, DM Sans body, Space Mono for numbers
Key components
The number, front and centre — with 24h change.
High-contrast mint CTA for the three core moves.
Icon, price, change and holding in one scan.