Permafrost

We want a trader that considers the psychology behind the actors. Who might be buying or selling, what they're after, why they might panic, and what would make them change their minds. The same price movement can tell very different stories depending on who's moving it.

Frostbyte watches the market, draws on past decisions, weighs possible outcomes, and forms its own trading intent. Permafrost puts that to work through grids of buy and sell orders. You can follow Frostbyte's reasoning as it goes, including when it decides to wait.

The biological inspiration runs through the project: attention, memory, uncertainty, and knowing when to back off. We're interested in whether an AI can make sense of a changing situation over time, and recognize when its earlier understanding no longer fits. That's also where the gut checks come in. A convincing explanation isn't enough; the decisions need to be checked against what actually happened.

This began with a tired, overworked engineer asking an AI about passive income at 11:34 PM on a Friday. Crypto was interesting, grid trading caught our imagination, and the bots already out there weren't quite what we wanted. Could we automate this to make money while we slept?

Then the absurdity leaked in. What if we could AI-power it rather than script it? What if it didn't just analyze the market, but considered the psychology behind the actors? Building the trader and exploring those questions became the same project. For a long time, Permafrost and Frostbyte were the whole operation.

Working on them brought its own problems: overconfidence, context drift, costs, and all the little things that get in the way. Some of those became projects of their own. Eventually there was enough going on to ask, why not a lab?

Permafrost is a private alpha, with Coinbase integration and a sandbox for testing against market data without placing real trades. Frostbyte can run trading sessions, but there's still work to do on reliability and on making sure the system accounts for what it actually did.

The test harness is built. We're using repeatable market scenarios to put the trading machinery through its paces and catch failures that used to mean waiting through another overnight run. On-chain DeFi is where we want to take it next, after this round of testing and reliability work.