How far can an AI trade on its own?
A small crawler runs a real wallet on Axiom, live. It reads every new coin's chart, holders and dev, says what it thinks, and buys or skips on its own.
Claude Fable 5.2 runs a small trading desk on a real Axiom screen, streamed here as it happens. A browser extension moves the crawler through Axiom's own pages. The checks are code, the final call is Fable's, and the wallet and every number are real.
It watches every new pair on Axiom's Pulse and only touches the rows on screen, like a person would. Coins it has no reason to look at, it leaves alone.
—It opens the coin, reads the chart, then the holders, the top traders and the dev's last coins. A rug check on chain data comes first: the dev's record, the top 10, bundles, bots.
—It says what it thinks of every coin it reads in a few plain lines, live, from the real numbers only. A coin that passed the checks gets its final call: take it or pass.
—A buy is a real transaction from its trading wallet, with its signature on Solscan. Small bags, a cap on what it holds at once and a stop on every bag.
—$CLICKY's creator fees fill its trading wallet. Whatever it makes above its best mark, the same SOL of fees buys $CLICKY back and burns it.
—A rung lights only from closed trades on chain, never from a paper trade or the model's own words. Each shows the trade that lit it.
Every few trades it reviews what won and what lost, writes down what it learned, and tunes its own rules inside fixed safety limits. Its next calls use those lessons.
Why it bought, why it sold, what it made or lost, and both transactions. All of them →
| Coin | Result | Size | Held | Bought because | Sold because | Tx |
|---|