The product · Private beta
From idea
to order.
Describe a strategy, let an agent build and test it, deploy it with limits, and watch every order — five surfaces that close one loop, with a person’s permission wherever it matters.
00 · The loop
One loop, seven steps.
Three are yours.
Describe it in plain English. Chat proposes each change as a card you apply; an agent builds a draft workflow and asks before it tracks a strategy or spends on a backtest.
In the product__create_workflow__track_strategy__run_backtestOrbital’s own deterministic engine replays daily or minute bars with explicit costs. The verdict arrives with its evidence — every trade, the drawdown, 1,000 resamples — and a Sharpe that says how many configurations were tried to find it.
In the productEquity vs. benchmarkHow much of this is luck?In Trade you pick the paper account, set four limits — budget, max per trade, open positions, a daily loss pause — and choose how much it does alone: Deny, Ask, Auto or Always-approve. Ask is the default.
In the productDeploy reviewBudget · Max per trade · Open positions · Daily loss pauseEach tick in market hours, the strategy decides — by its rules, its compiled program or an agent’s judgment, every 5 minutes or once a day. On paper today, agent-decided strategies trade.
In the productEvery 5 minutesEvery trading dayBefore an order leaves, the platform checks the permission mode, the four limits, the market clock and which strategy owns the position. Under Ask or Auto, a new trade waits in Activity for your Approve — and expires if nobody answers.
In the productWaiting for youApprove and placeOne order goes to your Alpaca paper account, placed by the platform — deciding agents never hold your broker credentials.
In the productSentFilledPositions, orders and one Activity timeline — every decision, hold, order and fill, with who acted and who approved it. Nothing is shown as closed until Alpaca confirms the fill.
In the productFilled · Strategy · Approved by Nathan
- A person’s click, by default
- The platform or an agent
01 · Orbital Chat
Ask in plain English.
Apply what you like.
Orbital Chat reads your workflows, runs and results and answers in plain English. When it wants to change something, it proposes the change as a card — nothing is written until you apply it.
Proposal cards
Every write comes back as a card you Apply; every backtest is a card you Run.
The work, shown
Raw tool calls like
backtest_getandedit_workflow, exactly as they ran.Your model
Pick a frontier model per conversation; usage shows as credits on every answer.
02 · Agents
Hire agents with jobs.
Not prompts.
An agent is a named, configured model with instructions, triggers and permissions. It works in the background and stops for you wherever you have asked it to.
Jobs, missions, monitors
Named agents that run on a schedule, work toward a goal, or wake when a signal fires.
Ask by default
Tracking a strategy or spending on a backtest stops at an Approve / Refuse card in your Needs-you inbox.
Receipts
Every run ends with what it made, what it spent and a
Show worktrail.
03 · Workflows
Strategies you can read.
Node by node.
A strategy is a workflow typed as a strategy: a universe, indicators, an entry, an exit and a size — each one a node you can open and read, and every run of it on the record.
Rules, program or agent
Decide with rule cards, compiled code or an agent’s judgment.
Proposals as ghosts
Agent edits appear dashed on the canvas until you apply them.
Every run audited
Per-node status, durations and logs, run after run.
04 · Backtesting
Evidence before exposure.
On the whole US tape.
Orbital’s own deterministic engine replays the tape with explicit costs, then reports what the result is worth — including how much of it could be luck.
The whole tape
Daily bars for every US stock since 2016, minute bars since September 2023.
Byte-reproducible
The same spec on the same data produces the same bytes.
Luck, measured
Deflated Sharpe with its configuration count, Monte Carlo resamples and an honesty ledger.
Every run reports
Found, or fitted? Every backtest tells you.
- Daily bars start: All US stocks, from 2016-01-04.
- Minute bars start: From 2023-09-05, regular trading hours only.
- Reproducible: Same spec, data snapshot and engine version produce byte-identical results.
05 · Trade
Every strategy on one ledger.
Every order on the record.
Run your strategies on your Alpaca paper account from one place. Activity records every decision, hold, order and fill — and who approved it.
One ledger
Every strategy on the account with its evidence, its budget beside every other budget and how it is doing. Going over the account warns; it never refuses.
Approvals in Activity
Held entries and exits wait in one list. Approve opens a review, Manual… opens a prefilled order, and Refuse takes two clicks.
New order
Place a trade yourself — Buy or Sell — and the ticket says when a sell would open a short.
06 · Control
Autonomy
you can dial.
Permissions are enforced by the platform the moment an order is admitted — never by a prompt. Deciding agents hold no broker credentials.
Asks you to approve every trade it makes, including exits.
- Entry signal11:35 ET
- Waits for your approvalDropped if nobody answers within 24 hours.
- Exit signal12:10 ET
- Waits for your approvalPlaced anyway if nobody answers within 24 hours.
Ask. Asks you to approve every trade it makes, including exits. Entry signal: Waits for your approval. Exit signal: Waits for your approval.
Deploy review · one strategy
One review before anything runs.
It settles what a strategy may do with your paper account — and only a person deploys it.
The proposal
Strategy, account and limits arrive together — filled in by you or proposed by an agent — and nothing runs until you deploy it.
Kill switch · one account
Stop everything. Two clicks.
Pause every strategy, cancel open orders and close positions at market — in a safe order, each step waiting for the one before it. Other accounts keep running.
Nothing is shown as closed until Alpaca confirms the fill.
07 · Under the hood
The rest of the desk. Built in, not bolted on.
Deterministic engine
Rust, bar replay, byte-reproducible results.
The whole tape
Daily bars since 2016, minute bars since September 2023.
Frontier models
Pick the model per chat and per agent.
Market-hours automations
Schedules and events gated to US market sessions, in ET.
Notifications
Approvals waiting for you and finished runs, as they happen.
Dashboards
62 widget types for everything else you watch.
Watchlists
One list feeds every workflow that reads it.
One timeline
Every decision, hold, order and fill, with who approved it.
Get early access.
Orbital is in private beta. Join the waitlist and we’ll let you know when a seat opens.