the agent proposes/your rules decide/six rules, no exceptions/allowlist, never a blocklist/a pure function, not a model/same inputs, same verdict/edge cases are stamped, not hidden/then attack your own rules/non-custodial/the agent proposes/your rules decide/six rules, no exceptions/allowlist, never a blocklist/a pure function, not a model/same inputs, same verdict/edge cases are stamped, not hidden/then attack your own rules/non-custodial/
ai agent · tokenized stocks · solanait never decides

The agent
proposes.
your rules decide.

An interlock is the circuit that will not let a machine start until the guard is shut. This one sits between what the model wants and what it is allowed to do: the AI writes the proposal, a deterministic engine you can read in one sitting writes the verdict, and no amount of confidence in the thesis substitutes for passing the rules. Then you point a hostile model at your own rules and see what still gets through.

59 verified tokenized stocks · 3 allowed by default · non-custodial · the engine is a pure function.

$ILOCK contract
EwewmjWBEejYiQmbTaRy4SQXqQiuHhhvUTuLxUdqpump

published in two places only — interlock.lat and @interlocklat on X. any other address — a DM, a reply, a “team” account — is a fake, however early it appeared. match the address, never the ticker.

guard circuit online
proposals → gate → verdictallowrefuse
rules
6
decided by
math
model vote
0
stock menu
59
loading the stock menu…
how it works

three moving parts. only one of them is a model.

01

You write the rules

Which tickers it may touch, which side it may take, the most it may spend at once and per day, how sure it must be, and how much reasoning it owes you. Six numbers — that's the whole policy.

02

The agent proposes

A model reads the live market and writes exactly one trade, with a thesis. It's told your policy. It still doesn't get a vote — it can only ask.

03

Arithmetic decides

A pure function runs every rule and stamps ALLOW or REFUSE, with the exact line that decided it. No clock, no randomness, no model. Same inputs, same verdict, forever.

the protocol, in numbers

nothing here is invented.

reading chain…
verdicts signed
0
on-chain · none yet
allowed / refused
0 / 0
settled decisions
stock menu
0
real equities
rules enforced
6
no exceptions

pre-launch — signed verdicts and settled decisions read zero until real ones exist. the stock menu, its prices, and every verdict you run are live right now. $ILOCK figures appear once the mint is set.

the difference

every other agent
asks you to trust it.

A model can be confident and wrong, persuasive and manipulated, or quietly steered by whatever it just read. None of that reaches your money here, because the thing holding the trigger isn't a model — it's six lines of arithmetic you can read, re-run, and tighten yourself.

And when something does slip through, that isn't a bug in the engine. It's the floor of what you authorized — which is exactly why we ship an adversary that hunts for it in public.

what's real right now

shipped & testable

  • · the verdict engine (pure, 23/23 tests)
  • · live proposals from a real model
  • · the adversary that attacks your rules
  • · 59 verified tokenized stocks, priced live
  • · server-side verdicts anyone can call

not yet — and we say so

  • · execution: nothing is traded for you
  • · on-chain signed verdict log
  • · $ILOCK + fee routing

published before launch so it can be held against us.

what would your rules
let through?