Why you can sign what Yeetful builds

What non-custodial means here

• No deposits. There is no Yeetful balance to fund. Your assets stay in your wallet until a transaction you signed moves them.
• No keys. Yeetful cannot spend on your behalf. Every transaction, order, and vote is offered to your wallet and moves only with your signature. Even the email sign-up creates a Coinbase non-custodial wallet that you control, not Yeetful.
• One narrow exception, by explicit delegation: Guardian uses a Hyperliquid agent key you authorize with one signature. That key can only reduce the position you told it to protect — it cannot withdraw, cannot open or grow positions, expires on its own, and can be revoked at any time.

The model never writes calldata

The language model that understands your ask never writes calldata, amounts, or addresses. Money asks are claimed by deterministic parsers — the same sentence hits the same code path every time — and handed to per-venue builders that derive the transaction from live venue state: real quotes, real pools, real balances, fetched at build time. The full venue-by-venue doctrine is on Native venues & guards.

Every build is re-checked, fail-closed

Before your wallet sees an artifact, an independent guard re-derives or decodes it and checks it against policy: pinned contract addresses and function selectors, exact amounts, verified recipients. Any mismatch kills the build. When a check fails you get the reason — “wallet holds only 2.52 USDC on Arbitrum” — never a guess, and never a transaction that hopes for the best.

Nothing stale ever reaches your wallet

Quotes and calldata expire. Yeetful treats that as a safety property: artifacts carry their validity window, multi-step chains advance automatically and re-quote steps at offer time, and anything past its deadline is rebuilt, not re-offered. You are never handed dead calldata that could settle at a stale price.

Priced, receipted, public

Every artifact carries its USD value, and every decision — built, signed, or refused — lands as a receipt on your dashboard, with build decisions traced live on /activity. When Yeetful charges a fee (0.20% on fee-bearing swap venues, below Uniswap's 0.25% interface fee), it appears in the artifact as its own labeled transfer step — never hidden inside slippage.

Caps protect the autonomous part

Your account's spend policy draws one line: anything an agent initiates without you in the loop is capped at $200 per action and $200 per day by default, while anything you sign yourself is consented by that signature and is not walled by the caps. Money you receive — sale proceeds, filled listings — is never gated at all. You can curate the allowlist down from the dashboard whenever you want.

The kill switch outranks everything

Freeze your account from the dashboard and every payment and build is refused — agent-initiated or not, inflow or outflow, policy on or off. Pause a single agent to stop just that key. Both are reversible; nothing is deleted.

Standing intents follow the same rules

The things Yeetful does between your visits never escape the model above:

• Jobs— a compound intent compiles into steps, but each transaction is built fresh when it's offered and you sign every one. Settlement between steps is verified server-side, never assumed.
• Recurring buys (DCA)— “buy $10 of AAPL every week” creates a schedule, not an authorization. Each due period compiles a one-step job you sign; miss a week and it lapses rather than buying behind your back. See Jobs.
• Guardian — the only autonomous signer, under the delegated reduce-only key described above, re-guarded fail-closed at every trigger.

What you are trusting, exactly

Not prices — they are fetched live and re-checked. Not the model — it cannot write a transaction. Not Yeetful with your funds — it never has them. You are trusting one thing: that the guard doctrine holds — artifacts verify or die — and every build argues its case in public on /activity so you can watch it hold.