Autonomous AI trading agents, explained.
An autonomous trading agent watches markets, forms a decision, and acts on it without a human approving every step. Tradewink's version is Paper Autopilot: it runs that loop fully automated in a simulator or a connected paper/sandbox account. Public subscriptions are paper-only; separately approved private beta accounts may submit live broker orders.
Updated September 2026
Paper Autopilot workflow
Scan
Price, news, filings, options flow, and regime.
Debate
AI specialists review timing, catalysts, and risk.
Gate
Daily loss cap, position limits, trade-rate caps, and user rules.
Simulate
Paper orders fill in the simulator or your paper/sandbox account. No real money.
Paper or sandbox accounts at supported brokers only. See published signals or compare plans.
How an autonomous trading agent evaluates a trade
An autonomous trading agent chains the steps a trader would otherwise do by hand: gather data, form a view, check it against risk rules, simulate, and review the result. Tradewink's Paper Autopilot composes concurrent async loops that each specialize in a slice of the market — price, news, SEC filings — feeding a central event bus that reacts in real time.
The reasoning layer uses AI models routed per subscription tier, and the roster changes over time. Each candidate gets a conviction score on a 0–100 scale, and higher-scoring setups can get a deeper multi-agent review in which separate specialist roles — technical, risk, and execution — assess the trade, with a disagreement acting as a veto. Lighter models run the fast work: sentiment scoring, headline summaries, and exit-strategy updates. The AI score adjusts the rule-based ranking rather than replacing it, and only candidates that clear both the AI gate and the rule-engine gate open a paper position, in the simulator or your paper/sandbox account, and only after you have turned Paper Autopilot on. Public subscriptions are paper-only; separately approved private beta accounts may submit live broker orders.
The agent also learns. Every closed paper trade is reviewed by an AI reflection loop that extracts lessons — what the agent got right, what it got wrong, what regime it was wrong in. Those lessons are embedded into a vector memory and pulled into the conviction score on future trades. Read the deep dive in What is AI trading?
Paper Autopilot capabilities
Multi-agent AI review
Conviction scoring on every candidate, plus specialist review for higher-scoring setups. Models are routed per plan tier.
Regime-aware sizing
HMM-based regime detection gates paper position sizing and exits.
Risk manager with teeth
Daily loss circuit breaker, position limits, trade-rate caps, confidence gate.
Real-time event streams
Alpaca WS bars, Finnhub news, SEC EDGAR filings — all flowing through priority queue.
Adaptive strategies
Momentum, mean-reversion, breakout, VWAP, ORB — picked by regime and ticker profile.
Self-improving loop
Paper-trade reflections feed future conviction scoring — the agent learns from its own simulated trades.
Non-custodial by design
Tradewink does not hold funds. Paper/sandbox broker keys are Fernet-encrypted and revocable. Public subscriptions are paper-only; separately approved private beta accounts may submit live broker orders.
Paper-only orders
Orders fill in the simulator or a paper/sandbox account. Public subscriptions are paper-only; separately approved private beta accounts may submit live broker orders.
Set up Paper Autopilot, research first
- 1
Build a watchlist and review signals
Start in the web dashboard or Discord DMs: add the tickers you follow, review each signal's reasoning and risk levels, and paper-track ideas. No broker connection is required.
- 2
Connect a paper or sandbox account (optional)
Paper Autopilot can run entirely in Tradewink's simulator. If you prefer a broker's own paper environment, register API keys for a paper or sandbox account at a supported broker, such as an Alpaca paper account. Broker connections are limited to paper or sandbox accounts. Keys are Fernet-encrypted with PBKDF2 and revocable — Tradewink does not hold funds. Public subscriptions are paper-only; separately approved private beta accounts may submit live broker orders.
- 3
Set your preferences
Configure 50+ trading preferences: max position size, daily loss cap, excluded tickers, excluded sectors, max trades per hour, micro-account mode, and which strategies Paper Autopilot is allowed to run.
- 4
Turn on Paper Autopilot
Paper Autopilot stays off until you turn it on. On public plans, its loops watch price, news, SEC filings, and market regime, routed AI models score each candidate, and orders go only to the simulator or your paper/sandbox account. Any real trade under a public plan stays your own decision, placed by you at your broker. Separately approved private beta accounts may submit live broker orders.
AI trading agent versus rule-based bot
| Capability | What to check in a trading bot | Tradewink Paper Autopilot |
|---|---|---|
| Decision model | Which rules and models inform the decision? | Rule-based checks and routed AI review |
| Context awareness | Which data sources provide context? | Price + news + filings + regime |
| Regime adaptation | How are changing market conditions handled? | HMM regime detection + re-debate |
| Risk management | Which limits and failure handling are available? | Risk manager, circuit breaker, per-user limits |
| Learning | Are review and learning methods documented? | Post-trade reflection, lessons stored, fed forward |
| Account traded | Is execution paper, live, or both? | Public plans: simulator or paper/sandbox account; live access requires separate approval |
| Custody | Who holds your funds and can withdraw them? | Non-custodial — Tradewink never holds funds |
Safety, control, and regulatory questions
What is an autonomous trading agent?
An autonomous trading agent is a software system that continuously monitors markets, identifies opportunities, forms a decision, and acts on it without a human approving every step. Most run a loop of scan, review, risk check, order, and reflect. Tradewink's version is Paper Autopilot: it runs that same loop fully automated, but every order fills in Tradewink's simulator or in a connected paper or sandbox broker account. Public subscriptions are paper-only; separately approved private beta accounts may submit live broker orders.
Which AI models does Paper Autopilot use?
Models are routed per subscription tier using explicit model IDs, and you can also bring your own LLM keys (BYOK). The roster changes over time, so no single model is the product. The AI layer is a conviction check on top of rule-based screeners, not a standalone signal: it cannot see live order books or predict price, but it can weigh many noisy inputs (technicals, news sentiment, SEC filings, market regime) and surface inconsistencies. Paper Autopilot opens a simulated position only when the conviction score clears the configured threshold and the rule engine agrees, and only after you have turned it on.
How does Tradewink use AI models?
AI powers the reasoning layer on top of rule-based screeners and strategy engines. Each candidate gets a 0–100 conviction score from a routed model, and higher-scoring setups can get a deeper multi-agent review in which separate specialist roles (technical, risk, and execution) assess the trade and a disagreement acts as a veto. Lighter models handle high-frequency work such as sentiment scoring, headline summaries, and exit-strategy updates. Which model serves each task is resolved per subscription tier.
Can Tradewink's agent trade my real brokerage account?
No for public subscribers. Paper Autopilot runs in Tradewink's simulator, or in a paper or sandbox account at a supported broker that offers one (for example Alpaca, Tradier, Interactive Brokers, TradeStation, or OANDA). Public subscription broker connections are limited to paper or sandbox accounts. Keys are encrypted with Fernet/PBKDF2 and can be revoked at any time, and Tradewink does not hold funds. Any real trade under a public plan is your own decision, placed by you at your broker outside Tradewink. Separately approved private beta accounts may submit live broker orders.
Is autonomous AI trading safe?
Safety is a property of the guardrails, not the model. An agent that trades real money can compound a bad model call or a software bug quickly. Public subscriptions are paper-only; separately approved private beta accounts may submit live broker orders. Paper Autopilot still runs the guardrails you would want: it stays off until you turn it on, and it applies a daily loss circuit breaker, position limits, per-user max trades per hour, and an audit log of every action, so you can watch how an automated strategy behaves without putting money at risk.
How is an autonomous agent different from a trading bot?
A traditional trading bot follows a fixed rule set — buy if RSI < 30, sell if price crosses SMA. An autonomous agent reasons over context. Tradewink's Paper Autopilot weighs technicals, news sentiment, regime detection, and SEC filings, then runs an AI review before opening a paper position. It also learns from closed paper trades through a post-trade reflection loop.
Is autonomous AI trading legal?
Generally, yes — in the US, trading your own brokerage account with software you control is legal, and the SEC and FINRA do not prohibit automated or AI-assisted execution, though brokers set their own automation and account rules. Tradewink itself doesn't include live trading in public plans: it is a research and paper-trading tool, not a registered investment adviser or broker-dealer. It publishes signals and runs Paper Autopilot in a simulator or paper/sandbox account; it does not give personalized investment advice or hold funds. Nothing here is investment advice.
Start with signal review
Build a private watchlist, then inspect the thesis, counter-case, and risk context behind an available signal. Paper Autopilot runs in a simulator or a paper/sandbox account. Public subscriptions are paper-only; separately approved private beta accounts may submit live broker orders.
Important disclosures
Informational purposes only
Tradewink is published by Tradewink LLC, which is not a registered investment adviser, broker-dealer, commodity trading advisor, or financial planner. All data, signals, and analytics on this page are general, impersonal, and for informational purposes only. They do not constitute investment advice, financial advice, or a recommendation to buy or sell any security or other instrument.
Trading risk
Past performance does not guarantee future results. Trading involves substantial risk of loss, including the possibility of losing more than your initial investment. You are solely responsible for your own trading decisions.