A trader's self-built AI trading harness uses Jev for confidence scoring
A trader doing $50m–$100m in monthly volume built a read-only AI trading system where Opus 5.5 orchestrates agent swarms and Jev handles deterministic data and confidence scoring.
Working on a big article about the landscape of AI trading harnesses and agentic trading. As a discretionary trader doing $50m–$100m volume each month and an AI power user, I created my own AI trading harness with research agents and news trackers running 24/7, backtesting, and reading all venues where I hold positions. It proposes trades and suggests adding/trimming positions based on sentiment, PA, TA, onchain, and GEX flows. It has all the required skills, connectors, and APIs. The main model is easily swappable - currently Opus 5.5 as an orchestrator running agent swarms with lesser models and a mix of agents for double-checking bullshit, plus @typesafeai JEV for deterministic data and confidence scoring. My hard rule is that the system is read-only; I still click the buttons myself. But to fully remove yourself and your emotions (the biggest risk factor for me), you need to break the barrier of “I don’t fully trust agents with real money.” So I started digging - from AI research and research co-pilots to fully autonomous agents trading onchain, in the trenches, and onchain perps and options. I’ve been testing and using different products: @nansen_ai @get_truenorth @minara @gmgnai @CircuitAI @paradex (dimeterminal) @pear_protocol (agent pear) @DXRGai If you know someone or are building in this space, reply below or tag teams/products/founders. I’ll cover the best products/teams in my article.