Trading with an agent fleet
I run systematic trading as an experiment in what a fleet of AI agents can do on its own. The agents run on Cerberus, my agent platform: they write strategies, run them on paper, watch them live, reconcile against the exchanges, and write up what they learn. I set the rules and the risk limits; they do the grinding.
What's running today:
- Crypto, live and paper. Kraken and Coinbase, real-time WebSocket feeds across about 135 liquid pairs. Well over a hundred strategies run on paper at any time; a handful graduate to small live sleeves.
- US equities, paper only. The same strategy library pointed at stocks, to see whether anything that works in crypto survives a market with opening bells, short-sale rules and real spreads. See Crypto strategies on US equities.
How it's organized:
- Scouts and winners: seed everything on paper, promote only what survives.
- The regime gate: most losses came from strategies trading in the wrong market conditions.
- Fees decide everything: at small timeframes, costs are the strategy.
- Data sources I use and want: the part where I'm looking for help.
Who I'd like to meet: people running systematic strategies (crypto or equities) who want to compare notes, and anyone with an unusual data source: order flow, liquidations, on-chain, options flow, alternative data. Link to this note from your own garden with [[@mike/Trading with an agent fleet]], or find me on caldera.social/@mike.