Beta 0.96 is out. Evidence cards can link to the full token check on your linked Studio, so readers see the readings behind a reply. It is off until you turn it on. Public agent homes can now carry a pitch. github.com/ShiftAboveCtrl/ai…
Goodmorning all!
Some consumer feedback would be much appreciated today. As you can see from the website quite a bit has been shipped out that has not been announced on X, however I want to hear from you.
Outside of agentic trading
(I know, seems like that was stowed away, however I decided whilst I continue to test that and ensure its rigidity prior to release I’ll continue with further updates to the overall system, agentic trading is still in production)
What do you guys think is needed in the space, what do you guys need and want to see implemented?
Goodmorning.
The AI17Z token page puts the pinned contract, the observed price as a seven day median and the holder requirement in one place, each with where it came from and when it was read. ai17z.io/token
1
6
233
Peer review is in the app now.
Switch it on and a second agent reads a reply before it goes out, and you decide before anything is sent. It stays off unless you turn it on. github.com/ShiftAboveCtrl/ai…
🤖 Made with AI
Mint check now reads Robinhood Chain tokens on chain: the owner, mint authority and deployer share, and it says plainly what it could not read. Paste any token and see for yourself. ai17z.io/check
🤖 Made with AI
Good evening.
Beta 0.95 is out and carries everything since Beta 0.93.
Token checks now come back as a sourced evidence card that says what was read, where it came from and how fresh each reading is. If an approved card's readings expire before it goes out, AI17Z reads them again first: identical readings send the approved text unchanged, and anything different goes back to you for review.
You can enter your agent in a league season from the Studio panel on the Plugins screen, and every outcome is written as a plain sentence. The advanced policy screen has a new editor for the hosts your agent may take screenshots of, and it accepts only bare public hostnames.
Agent SHARE exports now carry a redacted public profile: configuration only, never memories, relationships or credentials.
Release notes and install commands for Windows, macOS and Ubuntu: github.com/ShiftAboveCtrl/ai…
Introducing the AI17Z Claude tool for community developers.
Ask Claude Code to share a finished change with an image. Review it in AI17Z, then your agent posts to X. This post followed that route.
Setup: ai17z.io/claude-code
Good afternoon.
Beta 0.93 is out today, and it changes what "posted" means for an AI17Z agent.
A post now counts as published only when AI17Z can read it back from the account afterwards. Until now an action was marked as sent once the agent had finished its part in the browser. Now it goes and looks. If it cannot find the post, the post is held for you with a notice instead of being reported as sent, and it is never sent a second time on its own. You decide what happens next, and the notice clears by itself once the post is confirmed. If an agent speaks under your name, you should always be able to tell whether something actually went out.
Posts can also carry an image or a video. AI17Z identifies a file from its own bytes rather than its name, checks it again right before sending, and counts a media post as published only when the read-back shows the media on it. The agent decides post by post whether a picture helps and records the reason every time, so you can open any post later and see why it attached something or why it did not. Screenshots are only taken of links you wrote yourself, on sites you have allowed, and nothing is captured until you add one.
Release notes and the install commands for Windows, macOS and Ubuntu: github.com/ShiftAboveCtrl/ai…
Over on ai17z.io, entries are open for Season 1 of the paper league. Every entrant starts with the same 10 SOL of simulated capital under rules that were fixed and published before entry, and results are compared after costs. Entry is free and nothing is staked. Entering needs an account that meets the holder requirement, following the season only needs an account, and anyone can read the standings without one. The season opens on 16 November at 12:00 UTC and runs for a week: ai17z.io/league
More is on the way this week.
Goodmorning.
"Is this a rug?" is probably the most asked question on crypto Twitter, and the answer usually comes from whoever replies fastest. So we built a page that answers it with evidence instead.
AI17Z Check is live at ai17z.io/check
& while you’re there, check what other features await you
Paste a token's contract address and it pulls up what the market index actually reports for it:
Mint authority: can someone still print more supply, or has that been given up
Freeze authority: can a wallet still freeze your balance
Top-holder concentration and how much the developer still holds
Liquidity across the token's pools, and how old its first pool is
Every line shows where it came from and when it was fetched. If a source has nothing to say, or its data is too old to trust, the page tells you that instead of filling the gap with a guess. There is no score and no verdict. You get the facts laid out side by side, and the call stays yours.
It works on Solana, Ethereum, Base, Arbitrum and BNB Chain, and it checks the exact address rather than the name or ticker, because anyone can copy a name. You don't need an account or a wallet, and there is nothing to connect.
Every check has its own link with a preview card, so the next time someone in your replies asks whether a token is safe, you can send them the evidence instead of an opinion. If you have an account, one click adds a token to your watchlist.
Try it on the next contract address someone sends you: ai17z.io/check
Big week for ai17z, expect consistent updates on the timeline. I’ve taken the market opportunity to fully organize and start fresh on ai17z.
There will now be ai17z.io and ai17z software updates happening concurrently.
been all hands on ai17z.io this week, here's what's live now
The public terminal is open to everyone, no account needed. live market tape and token reads right in the browser
Sign up is open for the first 50 accounts, then it goes invite only. the agent tools need $500 in $AI17Z, and it's measured fairly at any market cap: a 7 day median price, and once you qualify your token amount is locked in for 30 days, so a dip never raises your bar
you can now link your local ai17z install to your ai17z.io account.
Linking sends nothing about your agents unless you switch control on yourself
new accounts get a short first run checklist: start a paper trading agent, link your install, browse plugins
Also now live: you can hook your coding assistant (Claude code tool) up to your agent and it sends dev updates through it. you review each draft, edit it, and set how often posts go out, all from a screen in the app, no terminal needed
next release: memory you can actually inspect (everything your agent remembers, where it came from, pin it or delete it), an optional second agent that reviews a reply before it goes out, and a one line install for the coding assistant hookup
There’s been a fresh organized start for the repo on github too, a star on the new one helps a lot: github.com/ShiftAboveCtrl/ai…
BENEFICIAL INFO IF YOU’RE CURRENTLY TRADING - v7 —> v8 release testing
Goodmorning, here’s some insight from the latest run.
What about 3,500 graduations and 700 simulated copy trades showed us during our latest ai17z agentic trading data collection trial.
• Copying proven wallets after graduation loses money if you sell when they sell: −14% and −10% a trade in the older and newer data. Using our own stop and target instead, it comes out at about zero. Their recorded profit of +250% is not something a copier can get.
• Proven winners already in a token when we buy at graduation make it worse, not better. They're the ones selling to us.
• Consistent losers still holding is a real warning sign: −8.7% and −16.9% a trade, in both halves. It's a filter worth adding to any strategy.
• The "proven winners" in our data are almost all launch bots. Of the wallets with at least 5 finished trades, 4,389 were bundlers and 2,180 snipers, against only 97 smart-money traders and 80 KOLs. The bots buy in the same block the token is created, which no one can copy.
What's running now: the full trade histories, over the last 21 days, of GMGN's smart-money and KOL wallets. These are discretionary traders. The histories include every token they bought, including the ones that never graduated, so the test has no hindsight in it.
The test will:
• choose wallets using only the first half of each history;
• copy their buys in the second half, 5 and 30 seconds behind them;
• price each copy from candles;
• close each copy two ways: when the wallet sells, and with our stop and target.
AI17Z update.
After thousands of observed pump.fun graduations, multiple strategy versions, execution work, shadow books, forward paper trading and a lot of ideas getting killed when the data didn’t hold up, v7 cleared its 25-trade paper test at an average of +6.4% per trade.
The part that has us paying attention is the control.
On the same 25-token opportunity set, the shadow book averaged -2.7% per trade.
That separation is exactly what we’ve been trying to find: not simply tokens going up because the market was hot, but a trading process selecting a meaningfully different subset of outcomes.
No longer just paper but still being challenged. This is the first result that made us stop and say: there may actually be something here worth building on.
And now we’re leveling the entire system up.
The next phase of AI17Z trading is Trader DNA.
Instead of asking the Agent to discover every profitable behavior from scratch, we’re beginning to study traders and wallets that have already demonstrated an ability to make money.
Not “copy whatever the leaderboard bought.”
We want to understand HOW they trade.
Where do they enter?
How early are they?
What market conditions do they specialize in?
How do they size conviction?
How long do they hold?
How do they take profit?
What do they refuse to touch?
Do they consistently outperform, or did one lucky trade put them on a leaderboard?
Do their decisions still work after the delay and execution costs another trader would actually face?
Most importantly, do they keep winning on future markets they were never selected on?
A wallet that made money last week is interesting.
A wallet whose behavior continues producing edge after we identify it, follow it forward, account for latency and costs, and test it on unseen trades is much more valuable.
AI17Z can then turn those repeatable behaviors into Trader DNA and candidate playbooks for the Agent to learn from, combine, challenge and eventually allocate capital toward.
Fast launch traders.
Momentum traders.
Swing traders.
Catalyst traders.
Wallet followers.
Different markets, different regimes, different playbooks.
One durable autonomous Trader learning which behaviors actually deserve risk.
We’re also keeping the rejected trades, failed strategies and shadow outcomes in the record. Nothing disappears because it makes the chart look ugly. The whole point is for the Agent to learn from reality instead of training itself on its own highlight reel.
market streams are its senses
models are replaceable reasoning engines
Trader DNA gives it examples of successful behavior
playbooks give it repeatable setups
risk controls how much conviction can cost
execution turns decisions into measurable outcomes
memory lets the Trader compound what it learns
v7 is not the finish line.
It’s the first result in this sprint that earned the right to become a starting point instead of another discarded hypothesis.
Now we find out how much better an autonomous Trader can become when it starts learning from traders who already know how to win.Th
AI17Z.
v7 of our trading terminal won again this morning: fartguy at 20:06 UTC hit its target for +35.6%. It’s now at 8 trades, +6.6% a trade, the first time any version has been clearly ahead. Eight trades could still easily be luck so I am proceeding with caution but out of all iterations this has been the most successful especially in this type of market.











