Build a Paper-Trading Agent with Separate Execution and Strategy Layers

A trading agent should separate the ability to execute trades from the logic that decides which trades to consider. In this architecture, Co-invest provides the programmatic trading environment, while Quiver Quant provides strategy and disclosure data. Hermes coordinates the two, schedules checks, records activity, and publishes dashboards.

The system should begin with paper trading. This allows the user to observe the agent's behavior without immediately exposing real funds.

The execution layer is connected through the Co-invest MCP. After authentication, Hermes can inspect a portfolio and interact with supported markets. Paper trading is enabled so positions and balances are simulated.

The strategy layer is connected through the Quiver Quant MCP and API access. One use case follows a selected congressional-trading strategy. Hermes checks for disclosed activity, translates relevant updates into the paper portfolio, and maintains a record of what it has done.

The workflow includes several scheduled elements:

1. A daily check during trading hours for new strategy data.
2. Execution of corresponding paper trades when the strategy produces an update.
3. A weekly Friday review.
4. A here.now dashboard showing the starting balance, current positions, performance, and strategy

being followed.

1. A Multica project and autopilot that expose the recurring work.

The example allocates a defined paper balance to one strategy and notes that disclosed trades may appear with a delay of one to three days. Hermes is instructed to keep checking and to mirror updates when they become available.

Separating execution and strategy makes the system easier to reason about. A brokerage connection can execute actions but does not determine whether the strategy is sensible. A strategy feed can provide ideas but cannot place trades. Hermes becomes the orchestration layer between them.

Paper trading also exposes operational problems. The agent may misread an update, size a position incorrectly, miss a schedule, or produce a confusing report. Those problems should be found before real money is enabled.

Dashboards make the activity reviewable from a phone. A project board creates a persistent record of what the agent checked and what it changed.

Create a dedicated trading profile rather than placing financial access in the general assistant profile.

Connect Co-invest and explicitly enable paper trading. Ask Hermes to report the current paper portfolio and confirm the account mode.

Connect the selected strategy provider. Ask Hermes to retrieve a known strategy or set of disclosed trades and explain what data is available before any simulated execution occurs.

Define the experiment:

• The paper balance allocated to the strategy.
• The exact strategy or person being followed.
• The frequency of checks.
• The handling of disclosure delays.
• The information required in the weekly report.

Create the daily cron job and Friday dashboard job. Require the dashboard to show the initial balance, current holdings, changes made, and the strategy label. Create the matching Multica project so every action can be inspected.

Review several cycles before considering any change in account mode. Corrections to position sizing, timing, or reporting can be saved as updates to the trading skill.

The system does not remove strategy risk. A strategy can perform poorly even when the execution is technically correct. Disclosure data can arrive after the original trade, so the copied result can differ materially from the tracked result.

A self-improvement loop can refine how Hermes interprets and executes the chosen procedure, but it cannot make an unprofitable strategy profitable by definition. Real-money access also changes the risk level substantially. The paper-trading setup uses simulation as the starting state and treats live execution as a separate decision.

Keep strategy, execution, scheduling, and reporting as separate layers. Validate the complete loop in paper trading, inspect every action through dashboards and project history, and do not confuse operational automation with a guarantee of investment performance.