AI Trading

The Analyst

An AI analyst that hunts your desk for setups and posts marked-up setup cards — the chart, the levels, an explicit invalidation and two targets — with every card graded on the record. Your desk is its mandate; standing instructions steer how it hunts.

Run Interval

In Setups, click the interval beside Start/Stop. New analysts follow the saved desk interval by default: Desk · 5m means the analyst checks at each five-minute candle close, and changing the desk interval changes its schedule too. Choose Every 1m, 5m, 15m, 1h, 4h or 1d to use a fixed schedule. Choose Desk interval to follow the desk again.

  • The choice is saved per desk and works with the browser closed. A fixed schedule changes how often the analyst looks; it still reads the desk’s signals and chart timeframe.
  • Changing the interval does not start a stopped analyst. Start/Stop remains separate. Runs are checked once a minute after candle closes; a setup card appears only when a setup qualifies.
  • Existing analysts retain their saved intervals. Select Desk interval to make one follow the desk.

Sound For New Setups

Click the speaker beside Start/Stop in the Analyst header to hear a preview and enable the alert chime. Click again to mute. This browser remembers your choice across desks and reloads. After reopening VYX, click or press a key once to allow browser audio.

  • A chime announces new setup cards on the desk you have open, including cards filed by your connected AI. It still works while the Analyst feed is folded or Workbench is selected.
  • Existing cards, desk switches, reloads and outcome updates stay silent. Several setups arriving together produce one chime.
  • Keep VYX open and your device awake. Sound is an in-browser cue, not an email or push notification; a closed or suspended tab cannot play it.

What The Analyst Is

On desktop, Analyst fills the left column beneath the desk navigation. Match Log, the Analyst/Match Log tabs, and the desktop Catalysts column are temporarily hidden. Hover a desk tab and click its gear to open the wider signal editor. Analyst temporarily hides while you edit; Done restores your place in the feed. The research-panel toggle in the top bar collapses or restores the column. Collapsed setup cards keep their header and marked-up chart visible so you can scroll through the charts. Click a header to expand the signals, thesis, price table and catalyst beneath it. Collapse the Analyst section with its chevron; Start and Stop remain in the header. The analyst finds setups; it never trades. Where an AI Trader authors plans and places orders, the analyst reads the same board and posts setup cards — so you can run one on a desk you trade entirely by hand. There is nothing you have to write: the signals lit on your desk define what a qualifying setup looks like, and the analyst hunts for exactly that from the moment you press Start.

Setup Cards

On the rare pass where the board shows a clean setup — price at a real level with the flow reads agreeing — the analyst posts a card: a marked-up chart of the setup with the range, the leg into it, and the level lines drawn on, a short thesis citing your desk's own signals, and an explicit frame — entry, invalidation, first and second target, with the reward-to-risk each implies. Every price on the card comes from levels actually visible on the board; a proposal that reaches for an invented number is dropped before it ever becomes a card.

  • A few cards a day at most, one live card per symbol. Most passes produce none; that is the point.
  • A card is decision support with an explicit invalidation — research and opinion, not a signal to buy or sell.
  • A card grades only from the bar its entry prints — a bar that trades through the entry, or one that opens past it when the entry was resting at or on the near side of the last price (a bid under the market fills on a gap down, at a better price than the card claims). If the stop, the first target, or the 48-hour deadline arrives before that, the card is void — a move that happened without the trade is not a result, win or lose.
  • Every frame has to pay: the second target sits at least twice the risk away from the entry. A proposal whose second target is closer than that is dropped before it becomes a card, however hard the board fires.
  • Each card says why its entry and its invalidation sit where they do, and lists two to four conditions that would end the call — a close past a level, a regime row crossing a value, the desk's opposite read firing on the name — so you can check the call against the board yourself on every bar. The reward-to-risk of each target and the risk as a share of entry are computed for you from the frame; the analyst never does that arithmetic.

The Read

At the top of Setups sits the analyst’s read of your desk’s regime strip from its last pass: each setup family with its permission — hunt, selective or stand — on one line, with how long ago the strip was read. Hunt is lit; stand is dim. Click the line to unfold the environment in the analyst’s words, the tile reading that decides each family, what would flip the read, and what moved since the last pass.

  • Stand is a normal answer. When every family reads stand, the feed is empty on purpose — the line says why — and no card is ever posted for a family the read marks stand.
  • Every word is the analyst’s own from that pass, and it cites only your desk’s regime tiles. The line updates each time the analyst runs; a stopped analyst keeps its last read, with its age.
  • The same read is available to your connected AI over the API and MCP.

The Workbench

Workbench sits beside AI chat and keeps the research for your desk: charts, market comparisons, order-flow studies, and signal reads, all drawn from venue data. A signal read plots one signal’s score bar by bar for a market, scored by the same engine that ranks your board — either a signal already on your desk, or one the AI composes for the question and does not save. A drafted one wears DRAFT and offers Keep as signal: nothing reaches your signal rail unless you keep it, so you can see whether an idea would have fired before you own it. Bars the signal could not score show as gaps rather than zeros, and a shaded run at the left means it was still warming up. Ask chat to show its evidence; a new study opens beside the conversation with a brief highlight. Each study names its author and is a saved snapshot, not a live chart. Chat links reopen the study later, even after a reload. Choose All desk research or This conversation, and use Discuss to bring a study back into chat. On mobile, switch between Chat and Workbench without losing your message. Users with AI Analyst access start on Setups when the app loads, with Workbench and Analyst beside it in the same column. Background research adds a new-item count without interrupting an instruction draft or older research you are reading.

  • Setup cards now draw the next timeframe up behind their bars as hollow candles — a 5-minute card shows its 15-minute structure — so the higher-timeframe context is on the card, not in your head.
  • The AI you connect over the MCP can file setup cards of its own. They are fenced and graded exactly like the analyst's and marked as its author on the card, so its record on your desk is as real as the analyst's.

Standing Instructions

The desk is the mandate; instructions are the fine print. On Setups, Add analyst opens the Analyst view. Its status row says whether the desk has an analyst and whether it is hunting, with Start, Stop and the run interval right there. The box below holds the standing instructions — optional — what to hunt, what to skip, and how entries should be timed. A mean-reversion desk can demand the turn before the card: "after an extreme, wait for the flush to stall and hold — never catch a falling knife." A breakout desk can demand the opposite. Create analyst makes it and starts hunting on the next pass; Apply saves a change to a running or stopped analyst without starting or stopping it. One AI button is always there and reads by state: Draft with AI reads your desk and writes a first draft in a new chat; Improve with AI sends the draft in the box for a tighter version; Revise with AI reviews an existing analyst’s instructions against the desk. Every AI suggestion opens in this same box, labelled as drafted by AI, for you to edit and apply — nothing is saved until you do. Drafts stay with their original desk when you switch boards, and if the saved instructions change under an open draft you review them before applying. Remove analyst at the bottom deletes it with its cards and record. Setups filed by your own connected AI over the MCP show on the same tab, with or without a desk analyst.

  • Instructions steer the hunt; they never loosen the rails — levels must be real, catalysts are never invented, and every card is still graded.
  • The analyst reads a play-by-play of recent tape action — sweeps, absorption, liquidation cascades, level tests — so timing instructions act on evidence, not guesswork. Where a market cannot show a kind of evidence (there is no public liquidation feed, for one), the analyst is told so and never waits for it.
  • It also reads each of your desk's regime rows over their last 100 bars, bar by bar and oldest to newest, so an extreme, a fade, or a crossing back through zero is evidence it can time against — and a regime scoped to one of your watchlists reads over that watchlist for it, exactly as on your desk.
  • Before each pass the analyst also reads its own graded record on your desk — which archetypes hit, which were invalidated, and the thesis it wrote — so a losing call is not repeated in new words.
  • On every pass the analyst also keeps a running narration of each name on its board — what it is doing against its levels, which of your desk's reads carry it, which gates are open, and what would have to print for it to become or stop being a setup. A name that has left the board is no longer narrated, and when the board is crowded the names with cards and the names forming a plan keep their place ahead of the ones the read has ruled out. Every number in it is one your desk shows; a sentence with any other number is cut. Once a card is filed, the narration follows it — the entry printed, the first target printed, resolved — with the desk, not the analyst, deciding those states. On a live card the latest line for that name sits under the frame with its state — forming, entry printed, first target printed — and the same narration is available to your connected AI over the API or MCP.
  • Each pass the analyst also manages its own live cards: it tests every condition the card said would end it against the current board and either holds, withdraws the card citing the condition that tripped, or tightens the frame — a revision may move the invalidation toward entry or re-place a target, never widen the risk. A price-level condition ("close above 98.36") is tested by the desk against the last closed bar, and a condition that names one regime row and a threshold ("Sweep Direction under −30000") against that row's latest value, whatever the analyst answers. Every decision is logged on the card, and a withdrawn card stays on the record as withdrawn.

From Card To Chart

  • Hover a card and the real chart follows it: the market switches, the view travels to the moment the call was made, and the setup is drawn the way your own trades are — a plate from the bar of the call with the entry ruled and the risk and reward bands around it. A card from days ago loads that window in one read.
  • Click a live card to commit it. The market locks, the ticket opens with the analyst's entry, stop and two targets already staged, and size is the only thing left to set. The levels are yours from there — drag them, or send as-is.
  • Click a graded card to see the setup on the chart and what price did after it; nothing is staged for history.
  • Click a committed card again to release it. Leaving a hovered card returns the chart to where you were.

Every Card Is Graded

Cards are graded by the system against their own frame — first target, second target, invalidated, or expired — never by the analyst itself, stamped where price actually touched. The running record sits beside the feed, losses included, so you always know what the calls have been worth. The feed holds every card the analyst has posted on that desk, not a recent window: the newest open claim sits at the top, open by default and foldable like any other card when you want the feed's height back, and every collapsed card keeps its marked-up chart visible for scanning, so the record and the calls it counts are always the same set.

Catalysts Are Never Invented

When a card says the move is behind a headline or a macro print, it is pointing at one that actually exists, with timing that fits the move. When a market is moving and nothing on the wire explains it, the card says that instead — "no visible catalyst, flow move" is a real answer, and a far more useful one than a plausible story.

Limits

  • The analyst never places, changes, or cancels an order, and it never signs anything.
  • Its cards are research and opinion, not personalized advice.
  • It only sees the signals lit on its desk — light a signal to put a behavior in front of it. Change the desk and you change what it hunts.
  • One analyst per desk. Its feed follows the selected Trade Desk; an automated trader keeps its own execution controls and trade record.
  • Start, Stop, run interval, and standing instructions are its controls — there is no chat with the analyst; its voice is the card.

More docs

Try it on the live map

Open the desk and put this to work across 300+ Hyperliquid markets — free, no install, no password.

Open VYX