← Back to studio
📈

Paper Trader

A simulated trading desk where AI agents make the calls — inside hard, non-negotiable guardrails. It is never connected to a real brokerage. Every decision is logged, supervised, and reviewable.

Status · Simulation only Guardrails · Hard-coded Views —
Safety architecture

Guardrails first, opinions second.

The agent proposes; a deterministic supervisor decides. Risk rules live in code the model cannot rewrite.

🧱

Hard risk gates

Position size, daily loss limit, and per-ticker caps are enforced by a deterministic supervisor. Orders failing any gate are rejected — not warned, rejected.

🔌

No brokerage link

Paper-only ledger. There is no API path to a real account, by design. The app cannot move real money.

🧾

Audit ledger

Every signal, decision, guardrail outcome, and fill is appended to an immutable JSONL ledger for replay and review.

🤖

Agent theses

Research agents produce scored, cited buy/sell theses. The supervisor converts accepted theses into sized orders — never the reverse.

⏸️

Kill switch

One flag halts all autonomous activity instantly. State is flushed cleanly; nothing fires while it is set.

👁️

Human review

A live dashboard shows open positions, exposure, and agent reasoning for every trade, in real time.

The desk

A day on the simulated desk.

Illustrative demo of the agent loop — synthetic data, real mechanics.

BUY · gate ✓BUY · gate ✓TRIM · gate ✓ STOP · loss limitHOLD · below size
⚠️ Educational simulation only. Synthetic chart, not investment advice. The live product never connects to a real brokerage and cannot execute real trades.
Risk model

What the supervisor enforces.

≤2%
risk per position
≤5%
daily drawdown halt
20+
guardrail checks / order
0
real-money connections