Today we're opening the beta of Orbital AI, an agentic trading desk, to everyone. You describe a strategy in a sentence. An agent turns it into a workflow you can read node by node, tests it on Orbital's own backtesting engine, and, once you approve, runs it on your Alpaca paper account within limits you set. Every decision and order is logged with who made it and who signed off.

The beta is paper trading only, and anyone can sign up at app.orbitalaip.com.

How it works

Orbital has five parts: Chat, Agents, Workflows, Backtesting, and Trade. Together they take an idea from a sentence to a filled order in seven steps. Three of those steps wait for you: accepting the strategy, deploying it, and approving its orders. Orbital handles the rest.

Diagram of Orbital’s seven-step loop: Author, Test, Deploy, Decide, Admit, Send and Show, then back to Author. Three steps wait for a person’s click: Author, Deploy and Admit.
From idea to order in seven steps.

Chat

Ask about your workflows, backtests, and results in plain English. When Chat wants to change something, it proposes the change as a card, and nothing changes until you apply it. Every tool call it makes is shown as it ran, and you choose the model for each conversation.

Orbital Chat showing a proposed change as a card: the strategy graph with a new EMA 50 node marked as added, the entry rule marked as changed, and Apply and Discard buttons.
Chat suggests adding a 15-minute EMA(50) trend filter as a card. Nothing changes until you click Apply.

Agents

Agents work in the background: on a schedule, toward a goal, or when a signal fires. By default they ask before tracking a strategy or spending credits, and every run ends with a receipt. The beta ships with six, including an Investment Committee where five specialist agents argue the case and a chair makes the call.

Orbital’s Agents view. The Strategy Developer’s run shows two steps approved by Nathan, at 10:42 and 10:48, with a backtest started. The sidebar lists agents under Needs you, Working, Watching and Recent.
The Strategy Developer stops for your approval twice, to turn its draft into a strategy and to run the backtest, and each step records who approved it and when.

Workflows

Every strategy is a workflow you can open and read: a universe, indicators, an entry, an exit, and a position size, each one a node. It can make its decisions with simple rules, code, or an agent's judgment. When an agent edits a strategy, the changes show up dashed on the canvas until you accept them.

A strategy on Orbital’s Workflows canvas, drawn as connected nodes from Universe to Signals. A proposed EMA 50 node and an edited Entry Rule have dashed outlines with Apply and Discard buttons.
The same strategy on the Workflows canvas. The agent’s edits stay dashed until you apply them.

Backtesting

Orbital's own engine replays daily bars for every US stock since 2016 and minute bars since September 2023, with trading costs included. Rules fire at a bar's close and act at the next bar's open, so a test can't peek ahead, and the same test always produces the same result. Each report then asks how much of the return could be luck: it discounts the Sharpe ratio for every version you tried, runs 1,000 Monte Carlo resamples of your trades, and spells out what the sample is too small to tell you.

Orbital’s backtest report for an example SPY/QQQ strategy: stat cards for return, drawdown, Sharpe, trades and alpha, four Is it real? checks, monthly returns, and the worst drawdowns.
The report for an example strategy. Under the headline numbers, “Is it real?” weighs the result: only one version tried, 1,000 resamples, a short sample, and costs that took about 44% of the edge.

Trade

Trade runs your strategies on your Alpaca paper account and lays them out side by side, each with its evidence, budget, and results. Orders waiting on you sit in Activity, where you can approve, adjust, or refuse each one. You can also place orders by hand, and nothing shows as closed until Alpaca confirms the fill.

Orbital’s deploy review: a $10,000 budget, $5,000 max per trade, 2 open positions and a $300 daily loss pause, with trading permission set to Auto and the strategy deployed to a paper account.
Deploying is one review: budget, max per trade, open positions, a daily loss pause, and how much the strategy may do on its own.

You stay in control

Agents don't hold your broker keys. An agent can decide to trade, but Orbital places the order, and only after checking it against that strategy's permissions and limits.

Each strategy runs in one of four permission modes. Ask is the default, and you can loosen it as a strategy's record earns your trust.

ModeNew tradesExits
DenyNever placedNever placed
Ask (default)Wait for your approvalWait for your approval
AutoWait for your approvalPlaced, never held
Always-approvePlaced within limits, shorts includedPlaced

If an order waits on you for 24 hours, a new trade is dropped and an exit goes out anyway, so a strategy is never left holding a position because you were away.

Every strategy also has a budget, a cap per trade, a cap on open positions, and a daily loss limit that pauses new trades. And each account has one switch that stops everything: it pauses every strategy, cancels open orders, and closes positions.

A real strategy

The Backtesting screenshot above is the most recent strategy from my own account: a regime-gated overnight long/short strategy on eight AI-infrastructure stocks, now on its sixth version.

The backtest replays January 2, 2024 to October 28, 2026 on 1-minute bars, 708 sessions in all, starting with $100,000, with Alpaca's costs, 0.03% slippage, and market orders filled at the next bar's open. It began forward testing on October 29, so from here on its results come from days the backtest never saw.

MeasureResult
Return+259.81% (+57.5% a year, compounded), against +68.17% for SPY over the same days
Max drawdown-13.76%
Sharpe2.45
Trades987 (8 still open), 59% won, profit factor 1.80
Alpha vs SPY+40.1% a year, beta 0.19

Numbers this strong need checking. It took six versions and eleven runs to get here, and the report discounts the Sharpe for that search. The deepest drawdown is also the most recent, at the very end of the test. And these are backtested results, not live trading.

Orbital’s backtest overview for Regime-gated overnight long/short, AI infra 8: equity rising from $100,000 to $354,381 between January 2024 and October 2026, above a dashed SPY line, with the deepest drawdown at the end.
A backtest of my most recent strategy, version 6, from January 2024 to October 2026. The dashed line is SPY. The deepest drawdown, at the far right, is also the most recent.

Pricing and access

Every plan includes the full market data layer: prices and bars, fundamentals, news, analyst ratings, earnings, insider and congressional trades, and Stocktwits sentiment. Credits pay for AI usage only; backtests don't use any.

PlanPriceAI created per monthModels
Free$0100Cost-efficient OpenAI and Claude models
Pro$303,000 ($30 of AI usage)Any model, frontier included
Max$10010,000 ($100 of AI usage)Any model, frontier included; bring your own keys at cost

This is beta pricing, and plans may change before general availability. To get started:

  1. Sign up at app.orbitalaip.com.
  2. Connect an Alpaca paper account.
  3. Describe a strategy, review the backtest, and deploy it with limits you're comfortable with.

Questions go to the Orbital Discord until the docs ship. Teams, press, and partners can reach us at hello@orbitalaip.com.

Limitations

  • No real money moves during the beta. Every order goes to an Alpaca paper account.
  • Market data may be delayed, and minute bars only go back to September 2023, regular trading hours only.
  • A good backtest can still mislead. The strategy above took six versions to find, and its deepest drawdown is its most recent.
  • Agents make mistakes, which is why they ask first by default.

Why I built Orbital AI

I had strategy ideas I could describe in a sentence and no quick way to turn them into something I could read, test against real data, and run inside limits. Each one meant a week of code. The tools that promised to skip that step either hid the strategy inside a model or hid the costs inside a backtest nobody could reproduce.

I've spent thirteen years building Udundi, an e-commerce agency I started in college, and I built Compose, an A/B testing platform for Shopify. Both taught me the same thing: a process you can write down is a process you can improve. Orbital applies that to trading.

Orbital is a solo project. I build it with coding agents, and I test my own ideas on it, like the strategy above.

If you try Orbital and it does something you wouldn't have done, tell me in the Discord, on X, or on Stocktwits.


Nathan Parker,
Founder of Orbital AI, Inc.