Meet the LiquidityScan AI Agent: Ask for Setups in Plain English

What Is the LiquidityScan AI Trading Assistant?

The LiquidityScan AI trading assistant is a full-page chat that answers plain-English questions using live platform data. It calls internal tools to fetch active signals, market data, and strategy docs, then replies from what it retrieved, never from a guess. It is analysis tooling, not financial advice.

You reach it at /ai-agent, type a question the way you would ask a colleague, and it works the problem in the open, showing each tool it calls and what came back before it writes the answer.

LiquidityScan is deep by design: ten-plus detection engines, a Core-Layer confluence stack, a Confluence catalog with Sequences, a War Room cockpit, X-Ray dossiers, Pulse, and a Daily AI Brief. The cost of that depth is navigation. Asking one plain-English question and letting the assistant fetch and stitch the pieces beats assembling the answer across three or four screens by hand.

How the AI Trading Assistant Works: Tools, Not Guesses

The core design decision is that this assistant does not answer from memory. It answers by calling tools. When you ask a question, the model decides which internal functions to invoke, the platform runs them against live state, and the results are fed back to the model before it composes a reply.

The toolset maps onto the platform's real data surfaces. Broadly, the assistant can reach:

• Active and recent signals across the scanners, so it can list what is currently detected on a symbol or across the market.
• The Pulse feed, the RSI-confluence quality filter, to answer "which of these also has momentum confluence?"
• The Core-Layer snapshot, the multi-timeframe alignment stacks, for "where do the timeframes agree?"
• Ticker and market-overview data for live price, 24-hour change, volume, and a broad read of the tape.
• Funding, open interest, and long-short data for positioning context.
• Strategy docs, so it can explain what an Order Block or a Fair Value Gap (FVG) is, grounded in the platform's own definitions.

It can also take account actions on your behalf when you ask: add a symbol to your watchlist, set an alert, apply a tag, add a journal note, or run a position-size and trade-plan helper. Those are the same actions available in the UI, reachable now through a sentence instead of a series of clicks.

The practical effect is that the AI trading assistant behaves less like a chatbot with opinions and more like an operator sitting at the terminal, pulling up exactly the panels you would have opened yourself and reading them back to you.

The Anti-Hallucination Design: Why It Cannot Invent Data

This is the part that matters most, and it is worth stating plainly rather than dressing up. Large language models, left unconstrained, will confidently produce a price, a signal, or a link that looks right and is entirely fabricated. In a trading context that is not a quirk, it is a hazard. The whole assistant is built to make that failure mode structurally hard.

The rule is simple: every factual claim must come from a tool call. It will not state a price it did not fetch, describe a signal it did not retrieve, or produce a link it was not given. If the data is not in a tool result, the honest response is "I do not have that."

Three properties reinforce this in practice:

1. Retrieval before assertion. Facts about the live market originate in a tool result, not in the model's training. Ask for BTC's price and it calls the ticker tool rather than recalling a stale figure.
2. Transparency of steps. The assistant streams its reasoning and its tool calls as it works. You see which tool ran and what it returned, so a claim is auditable rather than taken on faith.
3. Bounded scope. The tools define what the assistant can know. It is a window onto the platform's data, not an oracle with independent beliefs about where price is going.

None of this makes the assistant infallible, and the honest framing is not that it is always right. It is that when it states a market fact, that fact was pulled from the same live pipeline the scanners use, and you can watch it happen. Grounding reduces confident fabrication; it does not eliminate the need for you to sanity-check the output.

What You Can Ask and Do

The assistant covers two broad modes: querying live state, and taking actions. On the query side, useful requests look like normal questions:

• "What setups are live on ETH right now?"
• "Give me a market overview, what is the broad tone today?"
• "Are there any OB+ zones forming on majors?"
• "What does the Core-Layer stack look like on SOL?"
• "What is funding and open interest doing on BTC?"
• "Explain the difference between a CHoCH and a BOS."

On the action side, you can drive your account without leaving the chat:

• "Add SOL and ETH to my watchlist."
• "Set an alert for that setup."
• "Tag this one as a London-session idea."
• "Add a journal note: waiting for the 4H retest."
• "If I risk 1% on a $10k account with a stop 1.2% away, what is my size?"

The position-size and trade-plan helpers are worth calling out precisely because of what they are not. They do the arithmetic of risk and geometry from inputs you provide. They do not tell you to take the trade. A trade-plan helper that lays out entry, stop, and target from your parameters is a calculator, not a recommendation.

A Worked Example: "Show Me Live OB+ Setups on Majors in the London Kill Zone"

Type that sentence and watch the assistant work top to bottom. It does not answer from a hunch; it decomposes the request into retrievals and shows each one.

First it resolves scope. "Majors" and "London kill zone" are parsed into a concrete set: the large-cap symbols and the London session window, part of the platform's ICT Kill Zones framework. It streams that interpretation so you can correct it if it guessed wrong.

Next it calls the signals tool, filtered to the OB+ / OB++ strong order-block engine, for those symbols. The tool returns whatever is genuinely live, the fresh, unmitigated zones the scanner is currently tracking, with their real state: formed, approaching, or touched. If nothing qualifies, it says so, rather than inventing a zone to be helpful.

Then it may enrich the result. It can pull the ticker for each hit to show current price relative to the zone, or check the Core-Layer snapshot to note whether higher timeframes align. If you asked for it, it could add those symbols to your watchlist in the same turn.

What you get back is a compact, sourced summary: the OB+ zones live on majors, where price sits relative to each, the session context, and the tool calls that produced every line. What you do not get is a directive. The assistant surfaced and organized the setups; whether any is a trade is your call.

Guessing AI vs a Tool-Grounded AI Trading Assistant

The category "AI trading assistant" spans two very different architectures that happen to share a chat box. The difference is not cosmetic, it determines whether you can trust a stated fact.

The takeaway is not that grounding makes an assistant correct about the future, nothing does. It is that grounding makes an assistant honest about the present. A tool-grounded system can still misread your question or organize data awkwardly, but it will not manufacture a BTC price out of nothing, because it has no path to a price except calling the tool.

Honest Limits: What This Assistant Is Not

Being useful and being honest about scope are the same requirement here. A few limits are non-negotiable and stated on purpose.

It is not financial advice. The assistant retrieves and organizes what the platform already detects and helps you plan around it. It does not tell you to buy or sell, and most of the engines it queries do not even output entries, they surface zones, biases, and structure.

It does not predict or guarantee. Nothing it returns is a forecast. Live signals describe what has been detected on confirmed, closed candles, not what price will do next. There are no win rates, because the platform does not compute or publish one.

It is not a signal service, and not infallible. It can misinterpret a vaguely worded request, and a tool can return an empty or partial result. The mitigation is exactly the transparency described earlier: because it shows its steps, you can see when a retrieval was thin and ask again rather than trust a shaky summary.

Finally, access is gated per user. The assistant is a controlled surface, not an open free-for-all, which is appropriate for a tool wired into live account actions like alerts and watchlists.

Used within those limits, an AI trading assistant like this earns its place: it collapses the navigation cost of a deep platform into a sentence, pulls the answer from live data, shows its work, and hands the judgment back to you, where it belongs.

Frequently Asked Questions

Does the AI Agent give trading signals or tell me what to buy?

No. It retrieves and organizes what the scanners already detect and helps with planning arithmetic like position size. It never issues buy or sell calls. Most engines it queries output zones, biases, or structure rather than entries, and nothing it returns is financial advice or a prediction. You make every decision.

How does it avoid making up prices or signals?

By design it must source every factual claim from a tool call. It fetches prices, signals, funding, and other data from live platform tools instead of recalling them, and it streams each tool call so you can verify. If the data is not returned by a tool, it says it does not have it rather than inventing a plausible answer.

What can it actually do besides answer questions?

It can take account actions when you ask: add symbols to your watchlist, set alerts, apply tags, add journal notes, and run position-size and trade-plan helpers. These are the same actions available in the interface, reachable through plain-English requests instead of clicking through menus. Alert and journal availability follow your plan.

Is the data it returns real-time and non-repainting?

The market facts come from the same pipeline the scanners use, which detects on confirmed, closed candles only, so signals do not repaint. Prices and ticker data are live at query time. The assistant is a window onto that data; it is only as current as the tool result it just fetched, which it shows you.

Related query paths

The AI Agent is a conversational layer over the whole platform. These guides explain the data it retrieves and how the underlying detection is built and validated.

• What is LiquidityScan? — the full platform the assistant sits on top of.
• How the LiquidityScan Scanner Works — how raw candles become the signals the assistant reads back.
• Order Block Scanner: Real-Time OB & FVG — the engines behind an "OB+ setups" query.
• ICT Kill Zones Complete Guide: A Pro Trader's Framework — the session windows used in the worked example.
• LiquidityScan vs Manual Scanning — the coverage case for letting software do the cross-referencing.
• Is LiquidityScan Reliable? — how the detection the assistant reports is validated.
• The LiquidityScan Daily AI Brief: Your Asia, London and NY Pre-Market Edge — a related angle on liquidityscan ai brief.