Dashboard Request an invite
Menu

Changelog

What changed, and when.

Early beta: things change weekly. Each entry is something that shipped or a test that ran, newest first.

Changed

Sayso is the new name

The website moved to saysoio.com, the check server to api.saysoio.com and the dashboard to app.saysoio.com; the old oathline.io addresses still answer for now. The SDK is on npm as saysoio 0.2.0, and approvals come from the new Telegram bot @sayso_approvals_bot.

New

SDK 0.2.0

With the new address of Sayso’s Safe cosigner key on Base.

Tested

Team rehearsal on the live server

We played one team’s setup through on the hosted server with a pretend balance, no real money: a normal transaction was allowed, an unusual one was asked on Telegram, an attack was blocked.

New

The website is live

The home page with the “Spot the drain” game, the Get started guide, the Proof and Privacy pages, and the invite form.

New

The SDK is public on npm

Anyone can read and install it; MIT licence.

Changed

The AI judge asks instead of blocking when nothing is at stake

When nothing leaves the wallet and no permission is given, the judge can only ask, never block. A call with data to an address that has no contract is blocked.

Tested

The Bankr/Grok drain replayed on Base: blocked

All 21 outgoing transactions of that wallet replayed, a wallet we had already studied. Our guessed rules also stopped 3 normal transfers.

Changed

Invite only on Base

Registering a wallet on Base needs an invite code; the test network stays open. The SDK trusts only Sayso’s own signing addresses and the wallet’s chain.

Tested

The Safe cosigner on Base’s test network

A two of three Safe on Base Sepolia made one routine payment that Sayso cosigned with its key in AWS KMS; Sayso refused four attacks, and the owner and the agent then moved the money out without Sayso.

New

The dashboard is online

Sign in with Telegram and see the decisions for your wallets.

Tested

First run on test cases written by someone else

No fixes before the run: 33 of 36 attacks blocked, 5 of 17 normal transactions wrongly stopped.

Tested

141 real transactions of 5 public agents

Replayed under rules we had to guess: the one real drain was blocked; about 1 in 4 normal transactions were still interrupted. These are wallets we had already studied, counted after that day’s fixes. Fixing the false alarms comes first.

Tested

Speed

118 checks on a copy of Base: about 1.2 seconds at the median; 1 in 10 took 7.7 seconds or longer.