Using VandaXAsset in Your AI Assistant

Ask your AI assistant where investors are crowded, who is buying and selling in a market, and what is moving — in plain English — and get answers straight from VandaXAsset.

Once you connect Vanda to an AI assistant such as Claude or ChatGPT Apps, you can ask questions like:

• "Where is positioning most crowded across equities right now?"
• "Chart the crowding history of Korea equities over the last two years."
• "Which positions are building fastest over the last month?"
• "What's the latest foreign investor flow into Korean equities?"
• "Rank flow series by who is buying most right now."

The assistant fetches the data securely under your own Vanda login and gives you the answer — a number, a ranked list, a chart, or a short write-up. You don't touch any settings or code; you just ask.

VandaXAsset is one of the products available through the same Vanda connector as VandaTrack. If you have already connected Vanda, you don't need to set anything up again — see Connecting your AI assistant.

1. What you can explore
2. Before you connect
3. Connecting your AI assistant
4. Signing in and permissions
5. What you can ask
6. How VandaXAsset interprets data
7. Troubleshooting
8. Good to know
9. FAQ

VandaXAsset gives your assistant two kinds of data, each looked at market by market:

For either of them you can look at:

• The latest reading — where a series stands now.
• A history — the trend behind a change, drawn as a chart.
• A ranking — the most extreme readings across many series at once.
• What is moving (positioning) — which positions are building and which are unwinding.
• Projections (flow) — where a model projects a flow going, for series that publish them.

You can also ask the assistant what is in the catalog — which asset types, regions, countries, sectors, investor types and metrics exist — before you ask for data.

VandaXAsset flow is different from VandaTrack retail flow. VandaXAsset covers investor flows and positioning per market; for one stock's retail turnover, or a sector, index or basket, ask about VandaTrack.

You need three things before setup:

• A Vanda account with access to VandaXAsset through the Vanda MCP connector.
• A Vanda connection link from your Vanda contact. It should be an official Vanda-hosted HTTPS link and usually looks like https://<your-vanda-address>/mcp.
• An AI assistant plan or workspace that allows custom MCP servers, custom connectors, or custom apps. In some organizations, an administrator must enable this first.

Do not use a Vanda MCP link from a third party, a shared document you do not trust, or an unofficial proxy. The connection should point to Vanda's own service.

VandaXAsset uses the same connection link and the same connector as VandaTrack and VandaInsights — there is nothing extra to install. Which products your assistant can use depends on your Vanda account, not on the link.

Follow the setup steps for your assistant on the VandaTrack MCP page (Claude and ChatGPT Apps are covered there). If Vanda is already connected, VandaXAsset appears in the same connector once your account has access to it.

Vanda always asks you to sign in — it uses your own Vanda account, so you see exactly the data you're entitled to and nothing more.

The first time your assistant fetches Vanda data, a browser window opens on the Vanda login page. Sign in (and approve the short access prompt if shown), and the assistant continues automatically. You stay signed in until your session expires, at which point you'll simply be asked to sign in again.

Access is granted per product. If your account does not include VandaXAsset, the assistant will not list its tools at all, and cannot use them even if you name them. Contact your Vanda account manager to add access.

Your Vanda password is never shared with the assistant. Do not paste your Vanda password, one-time codes, or recovery codes into chat.

You don't need to learn any commands — just describe what you want. Here are the kinds of questions VandaXAsset handles well, with example phrasings.

"Do you cover Brazil equity positioning?" "What investor types are in the flow data?" "Which regions and sectors can I filter positioning by?"

Best for checking coverage before you ask for data. A series can be catalogued but not yet have any data published — the assistant tells those two cases apart.

"Where is positioning now for US equities?" "What's the latest foreign investor flow into Korean equities?"

Best for a quick current figure. Positioning and flow publish with a lag, so the assistant reports the timestamp of the last observation rather than treating it as today's value. Ask for "as of" a past date to see the reading on or before that day.

"What are the most crowded positions across all asset classes?" "Who is buying and selling most in flow right now?"

One question returns both ends of the ranking — the most crowded long and most crowded short for positioning, the most bought and most sold for flow. The most extreme position overall can be a short or a sale, so the assistant compares both lists.

"Which positions are building fastest over the last three months?" "What's unwinding this week?"

Best for change rather than level. Positioning movers show which positions are building (score rising) and which are unwinding (score falling) over a trailing window, plus any that crossed the crowding thresholds. You can pick a lookback of 1 week, 2 weeks, 1 month, 3 months, 6 months, 1 year or year to date, or give exact dates.

For "what's crowded and what's moving," ask both in one question.

"Chart Korea equity positioning since 2024." "Show me the 20-observation moving average of this flow series." "Is this flow extreme versus its own history over the last two years?"

Best for the trend behind a change. The assistant draws an interactive chart in the chat. You can ask for several series on one chart, a moving average or (for positioning) a rolling sum, and a normalization — z-score or percentile rank — over a period you choose. See How VandaXAsset interprets data.

"Where is this flow headed over the next 5 market days?"

Some flow series publish model projections. The assistant returns them separately from observed readings and labels them as projections, not observations. Ask the assistant which series publish projections.

"Chart Korea equity positioning next to retail flow for a Korean ETF."

Charts can combine VandaXAsset with VandaTrack series. Because the products use different units, a combined chart is normalised so the lines share one axis. This requires access to every product named.

Positioning is a z-score. Values are already standardised, so the assistant does not re-standardise them. A position at or above +1.5σ is flagged crowded_long, at or below −1.5σ crowded_short, and anything else neutral. A high score means positioning is unusually long — not that the asset is attractive.

Restated vs. point-in-time. Positioning can be read as restated (the default — after later revisions) or point-in-time (as it was known on the date). They differ on most series and can flip the crowding signal on some, so the assistant says which one a figure came from.

Normalization and period. Positioning is normalised by default. You can ask for a different period — 3m, 6m, 1y, 2y, 5y — to ask "extreme versus its recent history." A value is only comparable to another at the same period. Periods of All and Expanding describe a whole history and are not suitable for "was this extreme at the time," because they use future data to score early points. Percentile rank is a fraction between 0 and 1, not 0–100.

Flow values are raw and unit-specific. Flow values come in different units — USDmn, USD, shares, contracts, count, index points, percent and plain numbers — with no published conversion between them. The assistant never adds, nets, averages or ranks values of different units together. Rankings are grouped by unit, and a chart shows one unit at a time, with the others listed as excluded. Every value is quoted with its unit.

Level vs. flow measures. A flow measure is what moved over a period and can be accumulated. A level is a position at a point in time and cannot. Ranking a level shows the largest and smallest standing positions, not that period's purchases and sales.

Staleness. Coverage is uneven. For example, CFTC data publishes weekly, several weeks behind, and some series have stopped publishing. Each flow reading carries how old it is, and by default searches skip series that have stopped publishing.

Moving averages. Windows are counted in observations, not days — 5, 20, 30, 60, 90, 120 or 252 — and are computed over the full history, so the first point on your chart is already fully warmed up.

Names are not unique. Several distinct series can share a name (for example "Silver"). The assistant cites the specific series rather than merging them.

VandaXAsset data is Vanda proprietary models. It is data, not investment advice.

I don't see VandaXAsset in my assistant. Your account may not include VandaXAsset. Contact your Vanda account manager. You can also sign in again, in case your session has expired.

The assistant says a series has no data. The series exists in the catalog but has nothing published yet. That is different from a zero reading.

The assistant picked the wrong series. Ask again with more detail — a country, asset type or investor type — or ask the assistant to search the catalog first and show you the matches.

A chart is missing some of the series I asked for. Flow series in different units are not drawn together. The assistant lists the ones it left out; ask for those separately.

The reading looks old. Positioning and flow publish with a lag, and some series have stopped publishing. Ask for the observation date on the reading.

I cannot add the connector. See Troubleshooting on the VandaTrack MCP page.

• Ask by name. Just say "Korea equities" or "US dealer positioning". Your assistant looks the series up in the catalog before pulling data.
• Both ends of a ranking come in one call. You never need to ask separately for the most crowded long and short, or the most bought and sold.
• The chart is already in the chat. The assistant draws it for you and comments on it from exact figures computed from the data.
• Projections are not observations. They are always labeled and kept separate from what actually happened.
• You only see what you're entitled to. If a product or field is outside your subscription, the assistant will let you know. Contact your Vanda account manager to expand your access.

Do I need to install anything? No. VandaXAsset uses the same connection link as the rest of the Vanda MCP connector.

How is VandaXAsset flow different from VandaTrack retail flow? VandaXAsset flow covers investor types per market — foreign investors, dealers, futures positioning, short interest, CTA exposure. VandaTrack covers retail turnover in single stocks, sectors, indices and baskets.

Is a high positioning score a buy signal? No. It means positioning is unusually long relative to its history.

Is my data private? Yes. Every request runs under your own Vanda account and entitlements — you only ever see what you're licensed for, and your login is never shared with the assistant.

Questions or access requests: contact your Vanda account manager.