An Agent with a boss.
Ask in plain words. It reads the market and your accounts through typed tools, shows every call, and hands you anything that would trade.
In the app
Ori, Alpha’s companion, made as an object for this site.
Ask about your own trading.
Five questions it can work through, each with the steps it took and a chart you can explore. They play by themselves; choose one whenever you like.
Agent examples
Example questions
Examples on simulated trades and a labelled model. Nothing here calls a model or places an order.
Backtest an NQ opening-range breakout over every mix of range length and stop size. Is there a stable area, or one lucky setting?
- research.draft{"strategy":"opening-range breakout","target":"2× stop"}Strata draft
- research.validate{"grid":"orBars × stopTicks","sessions":60}100 backtests
- research.review{"walkForward":"40/20"}choose in sample, test out of sample
- artifact.create{"recipe":"parameter-surface"}3D surface
28 of 100 settings made money over 60 simulated sessions. The best, a 75-minute range with a 48-tick stop, made +$2,387; its neighbours averaged +$1,383.
- Chosen on the first 40 sessions (75 min, 48 ticks: +$1,741), it made +$646 on the last 20.
- Simulated sessions; a surface describes this sample, not the future.
One order, four steps.
Recorded in the app on a simulated account. The Agent works on your side of the desk; the core sits between it and the market.
Every tool call is on the page.
The Agent does not get a private window into your accounts. It calls named tools (market.read, order.submit, artifact.create) and each call is printed in the conversation with its arguments and result. The tier tells you what it can touch: read, prepare, or trade.
When it trades, the order is an Agent order, attributed to the Agent, held by the core until you answer.
Or let it trade, inside a grant.
When you are away from the screen, a grant lets the Agent act without asking, only on the account, size and time you chose. On simulated accounts today.
Authority is set per account.
You choose what the Agent may do on each account. Only you can change it. The Agent has no action for it, and the core refuses it from the Agent’s connection.
- ObserveReads and navigates only. It can answer, chart and explain; it cannot prepare an order.
- AssistMay prepare orders that wait for you. Each one appears under Needs you with Confirm and Decline.
- Standing grantYou issue it for one simulated account, a size and a number of minutes. Inside it the Agent trades without asking; every risk check still runs.
- Delegate and Full Authority (in progress)Grants the core enforces for longer autonomy on simulated accounts: which instruments and actions, how many trades, a risk cap per trade, a loss budget, a time window, a pause around scheduled news, and a dead-man switch that holds or flattens if the Agent stops answering. The core enforces them today; the controls to issue them in the app are in progress.
Whatever the level, it cannot
- confirm its own orders,
- issue or widen grants,
- change risk limits,
- trade a live account on its own.
Your model, or none.
Bring a model key and the Agent plans with it; leave it empty and an offline local planner answers the common requests. Keys are kept out of logs. Models and spending has its own page in the Agent module.
The conversations on this page use the offline planner, so what you see is what the app does without any model key.
Research you can reproduce.
Ask for a correlation, a return distribution, a volatility cone or an equity simulation of your own trades, and the Agent makes a research card from the core’s closed bars or your round trips. Each card records its inputs and their hash, reproduces exactly from them, reads out the value under the pointer and exports as CSV or JSON.
Cards say what they are not: descriptive history, not a forecast or proof of an edge.
What is built, and its limits.
Recordings on this page are the Alpha app on a simulated account.
- Typed toolsIn the app
The Agent reads markets and accounts through typed tools and shows every call it makes.
- Authority levels and grantsIn the app
Per account: observe, or assist with orders that wait for you. A standing grant you issue can let it trade within a size and a time; risk checks still apply.
The Agent cannot confirm its own orders, issue grants or change risk limits.
- Trades on its own, inside a grantIn the app
Issue a grant for a simulated account, a maximum size and a number of minutes, and the Agent may place orders without asking until it ends. Validated strategies can be deployed to a simulation account and trade by themselves. Every risk check still runs, and you can stop either at any time.
Simulated accounts only. Alpha refuses agent autonomy on a live account until live broker routes, safety checks and legal review are complete.
- Your model, or offlineIn the app
Plans with a model key you supply, or with an offline local planner if you supply none.
- Research cardsIn the app
Correlation heatmaps, return statistics and distributions, volatility cones and equity simulations from your own trades, each reproducible from a hash of its inputs and exportable.
- Strategy and indicator draftsIn the app
Drafts strategies and custom indicators that you preview and save; display changes come with a restore point.
- The island and OriIn the app
The title bar’s island shows the clock, day P&L, feed health, risk room or the next event, and opens Home, Ask, Activity, Needs you, Risk and Copy.
Questions
Can the Agent trade without me?
Yes, inside a grant you issue, on simulated accounts: one account, a maximum size and an end time, Delegate and Full Authority grants, with a loss budget and a time window, are enforced by the core; issuing them from the app is in progress. Every risk check still applies, and you can revoke a grant from the island. Without a grant, its orders wait for you. Live accounts are not available for agent autonomy.
Which model does it use?
The one whose key you provide, or an offline local planner if you provide none. Alpha does not bundle a cloud model.
Does the website demo call a model or a broker?
No. The films are recordings of the app on a simulated account. Nothing on this site calls a model or a broker.
Can it change my risk limits if I ask it to?
No. Changing limits is not an action the Agent has, and the core refuses it from the Agent’s connection. You change limits yourself in Accounts and risk.
Want to see the Agent work on your own routine? Tell us what you would ask it first.