Screen Stocks with an AI Assistant

What the assistant can do

TradingLead's stock screener lives in the Scan sidebar of the chart workspace. With TradingLead Desktop and its AI integration, a terminal assistant can operate that sidebar through the public tl-screening skill. These are the same controller and API calls the buttons use, not a hidden scanner, so you watch every step happen in the app.

The skill offers three surfaces:

Most traders want the first row.

Before you start

• Sign in. The Scanner needs a real account. Assistants never substitute sample data for a failed sign-in.
• Enable Help > AI integrations in Desktop and start your assistant in a new terminal. See control TradingLead with AI.

Step 1: Discover what is available

Ask: "Read screening.getState, screening.catalog and screening.listUniverses and tell me which presets, fields and lists I can use."

The catalog is the source of truth. If it does not advertise a field, strategy or SQL permission, the assistant should not use it.

Step 2: Turn your idea into a definition

You describe the scan in plain language, for example: "NIFTY 50 stocks on daily where close is above SMA20 and RSI14 is under 70." The assistant maps it to advertised fields:

{"mode":"custom","timeframe":"1D","match":"all",
 "conditions":[{"field":"close","op":"gt","value_field":"sma20"},
               {"field":"rsi14","op":"lt","value":70}]}

Other mappings:

• Gainers: "stocks up at least 4% from the previous close" uses the gainers preset with min_change_pct:4 and change_basis:"previous_close". It reads the latest stored bar per ticker, not live movers.
• Original strategies: these use mode:"classic", with up to five parameter instances.
• Custom SQL: only available when your account's catalog grants it.

The universe can be the active list, a named watchlist, a flag, a category or explicit tickers.

Step 3: Configure, check, then run

screening.configure only fills in the draft. Configuring never runs a scan. The assistant reads the state again, confirms the definition and list count, and then calls screening.run with the new revision. Running consumes allowance, so a good assistant does not quietly split a large list into extra chargeable scans.

Step 4: Read the results honestly

The assistant polls screening.getResults and pages through it, 10 rows by default and at most 25 per command. It should report:

• status, warnings and coverage counts;
• truncation (capped results are only the returned stocks, not every market match);
• data warnings, such as possible split or bonus defects.

For the filtered, sorted grid you see on screen, it uses screening.getAnalysis and screening.setAnalysis. Up to 20 typed filters are supported.

Step 5: Refine, save or revisit

• Refine: screening.refine takes the grid's filtered or selected stocks across all pages as the input universe for the next scan. Then a fresh screening.run starts the follow-up scan.
• Save the rule: screening.save stores a reusable screener. Saved screeners are rules; saved scans are executed snapshots.
• Revisit: screening.listHistory and screening.openHistory reopen a past snapshot without rerunning it.

Step 6: Hand off the matches

• Watchlist: the assistant collects canonical tickers across all pages, removes duplicates and calls watchlists.create or watchlists.addSymbols from tl-markets. Each command takes up to 500 symbols. These are TradingLead cloud watchlists.
• Chart Grid: grid.open with the run ID charts the results as cards. You can limit it to the filtered or selected rows.
• Single chart: chart.open opens one verified result.

Scan with your own indicator on broker data

On Desktop, providerScreener.run evaluates a TLScript-Pine indicator on 1 to 50 explicit tickers using a supported connected provider's candles:

• Bars are bounded to 2–1,000, and the newest candle is excluded.
• Broker requests use provider quota.
• Pine strategies and secondary-data requests are not supported on this surface.
• Local runs expire after 30 minutes.

See Pine Script and build an indicator with an AI assistant.

What this is not

• Not chart-pattern detection. For one ticker's behaviour on certain weekdays or IST time windows, the separate Historical Patterns sidebar (tl-patterns) is the right tool.
• Not live advice. Scans are research filters over stored data.

Example prompts

• "Scan my 'Swing' watchlist on 1D for RSI14 above 60 and close above SMA20, then list the top 10 by % change."
• "Refine those matches to stocks with RSI14 above 65 and save the rule as 'Momentum v2'."
• "Open all matches in Chart Grid on daily."

For manual screening, see scan stocks with the screener and build a breakout screener. More prompts are in the AI prompt cookbook.