Research & signal layer
Agents combine market data and alternative signals into ranked candidates, each carrying the reasoning that produced it.
Flagship application
An autonomous trading desk that explains itself.
A swarm of agents runs a book end to end — generating signals, sizing risk, placing trades, and closing the day with a narrated video recap. Every fill is tied to the signal that justified it, recorded in a ledger, and reviewable.
Paper account. The numbers below are from a simulated (paper) trading account used to prove the system end to end. This is an engineering demonstration, not investment advice and not a live-money track record.
Live paper book
The swarm board.
Every open position the swarm holds — sortable, green up, red down, with dollar and percent unrealized. Click any column to sort.
Paper account — figures load from the account
The published trade record.
Figures come straight from the paper account: live when the feed is up, otherwise the last session close. The swarm board above lists every open position, and the recap archive holds every session.
Total return
—
since the $100k paper start
Account equity
—
against a $100k paper start
Open unrealized
—
on the open book
Positions held
—
managed by the swarm
How the desk trades
Signals in, risk sizing, broker fills, and a written reason recorded for every trade. Opinions are separated from actions; risk is a gate, not a suggestion; paper and live are different ledgers. The live board above lists every open position in real time.
Paper, not live money
Every figure here is a simulated paper account, clearly labeled. The point is the end-to-end agent workflow — signal to fill to a daily recap video, all accountable — not a live-money performance claim. A single session runs green or red; the track record above is the account since inception.
Daily recap pipeline
The desk closes the day with a video.
After the close, a post-event workflow pulls the day's figures, builds a dashboard and deck, narrates it, and renders a recap video — the video workflow on the platform, applied to real (paper) results. No human edits the clip.
Finance → deckThe session's P&L, leaders, and tape become a structured recap.
Deck → videoThe Vids operator renders and narrates it through automation.
Video → publishThe finished recap is ready to post — like the one shown here.
How the desk works
Signals in, justified trades out.
The trading app is a workflow on Open Swarm: specialized bots handle research, signals, risk, and execution, coordinating over the same mesh as every other application. Two ledgers stay separate — paper and live — so a strategy can be proven before a dollar moves.
A signal is only a nomination; risk sizing and the broker stand between it and a fill. Every fill points back to the signal that justified it.
Sizing & risk gates
Position sizing and risk checks sit between a signal and an order, so the desk respects limits instead of chasing every idea.
Broker connector
Orders route through a brokerage connector. A paper book proves the loop; a separate live book is gated behind explicit sign-off.
Every trade justified
Each fill links back to the signal that triggered it and is written to a ledger, so the book is auditable rather than a black box.
Paper and live, separated
Paper and live accounts are tracked independently. The record on this page is the paper book.
Daily video close
The post-close pipeline turns the day into a narrated recap automatically, the same video workflow used across the platform.
On the framework: this desk is a manifest-loaded oshal swarm. Market and paper-broker access come through per-user brokered connectors, the decision loop and post-close recap run as scheduled ticket workflows, positions and reasons live in a bot-owned store, every call lands in the central cost ledger — and the recap video is rendered by the Vids Studio app on the same runtime.
How a trade gets justified.
The blog has a deeper walk-through of how the swarm turns a signal into a sized, justified, logged trade — and how the recap video is generated from the result.