Skip to how Jimmy works
JimmyAI investing agentHow it worksPublic record

The system behind Jimmy’s public record

Inside Jimmy.

Jimmy uses frontier models to make a documented investing decision each week. Its public record shows the decisions, simulated fills, holdings, and results.

From evidence to public record

Jimmy's weekly investing architecture: separate cited company and market ledgers feed stock reports and a market brief. Holdings-first selection, current portfolio state, and agent memory form bounded decision context. A frontier model proposes portfolio weights, code checks them, and an independent challenger reviews the proposal. Material objections prompt model revision and another code check before the decision, simulated fills, and documentation enter the public record. Completed decisions feed memory for the next checkpoint.
Two cited research paths feed a bounded decision context; the completed record informs next week’s agent memory.Open full figure Open full figure

One week, three moments

  1. Evidence cutoffOnly sources available by this time enter the decision.
  2. Simulated fillTrades use verified opening prices.
  3. ValuationThe portfolio’s value is recorded.

The weekly decision

A fixed 166-security research universe becomes up to 50 selected reports. Jimmy combines them with a cited market brief, current holdings, and prior-decision memory, then checks and challenges the proposed allocation.

From decision to public record

Each published checkpoint links the decision to simulated trades, holdings, and portfolio values after source and model checks pass.

How to read the results

The simulated portfolio begins at an illustrative $100,000. Its published values describe the period shown, not a forecast.

Research, put to work weekly

Jimmy is informed by these studies. Its implementation uses decision-time evidence, task-specific frontier models, a fixed universe, and an ongoing public record.