API & MCP

The earliest session with bar data, YYYY-MM-DD.

The most recent session loaded. Scans, stats and alerts answer as of this day unless asked otherwise. To judge freshness, read data_stale rather than comparing this with today — it already knows the market's calendar.

The session that should have published has not arrived. NSE closes at 15:30 IST and the daily data lands during the evening, so this turns true after 17:00 IST on a trading day whose session is still missing, and false again when it loads. Weekends never raise it, and neither do the market holidays we have been told about.

OHLCV rows in the database, across all symbols and history.

Rows of derived per-session metrics — returns, RSI, moving averages and the rest of the column vocabulary.

Instruments in the database.

How many years back the derived metrics reach. Bars go back further — to first_date — but scans need metrics, so this is the scannable window.

Ticker or company name, whole or partial.

Maximum hits to return.

Matches, best first: an exact ticker beats a ticker prefix, which beats a name prefix, which beats a substring anywhere.

The canonical NSE ticker — the identifier every other endpoint takes in its path.

The ticker as displayed. Usually identical to symbol.

The full company or instrument name.

The instrument class — equity, index, etf.

Close of the most recent session.

The smallest liquidity universe the symbol belongs to: 100 means it is among the 100 most liquid equities by 20-day turnover, 500 within the top 500.

Every ticker in the database, sorted. Each is valid in the :symbol routes.

First letter of the display symbol, a–z, or 0-9 for the tickers that start with a digit. The 27 buckets partition the whole table.

Exact NSE sector, e.g. Capital Goods. Rows with no sector are unreachable this way — use letter for exhaustive coverage.

Exact NSE industry, e.g. Automobiles.

Rows to return, 1–5000. Default 1000.

Rows to skip, for paging through a large bucket.

Instruments matching the selector, before limit — what a page needs to say how many it is showing.

The canonical NSE ticker — what the :symbol routes take.

The ticker as NSE writes it, which is what to show a reader.

Company or index name.

equity, sme, etf or index.

NSE's sector classification. Null for every SME and index, and for a few hundred equities NSE never classified.

NSE's industry, one level below sector. Null on the same rows.

The smallest liquidity universe the symbol belongs to: 100 means it is among the 100 most liquid equities by 20-day turnover, 500 within the top 500.

Rank by 20-day turnover, 1 being the most traded.

Close of the symbol's last session — which may not be the market's.

Percent change on the last session, null when the symbol did not trade in it.

Average traded value over the last 20 sessions, in rupees.

Sessions of history held for the symbol.

The symbol's last session, YYYY-MM-DD.

The NSE ticker.

Peers to return, 1–50. Default 12.

What made these peers: industry, sector, or kind when the symbol carries no classification.

The industry, sector or kind the basis names — say it in the heading, so the list explains itself.

The canonical NSE ticker — what the :symbol routes take.

The ticker as NSE writes it, which is what to show a reader.

Company or index name.

equity, sme, etf or index.

NSE's sector classification. Null for every SME and index, and for a few hundred equities NSE never classified.

NSE's industry, one level below sector. Null on the same rows.

The smallest liquidity universe the symbol belongs to: 100 means it is among the 100 most liquid equities by 20-day turnover, 500 within the top 500.

Rank by 20-day turnover, 1 being the most traded.

Close of the symbol's last session — which may not be the market's.

Percent change on the last session, null when the symbol did not trade in it.

Average traded value over the last 20 sessions, in rupees.

Sessions of history held for the symbol.

The symbol's last session, YYYY-MM-DD.

The NSE ticker.

The canonical NSE ticker.

The ticker as displayed. Usually identical to symbol.

The full company or instrument name.

The instrument's ISIN.

The instrument class — equity, index, etf.

The NSE trading series code — EQ, BE — not a data series.

The first session with a bar on record, YYYY-MM-DD.

The most recent session with a bar.

Daily bars on record — the depth of history behind the metrics.

Close of the most recent session.

Mean daily traded value over the last 20 sessions, in rupees — the liquidity measure behind liquidity_rank and universe_tier.

Position when all equities are ordered by avg_turnover_20d — 1 is the most traded name on the exchange.

The smallest liquidity universe the symbol belongs to: 100 means it is among the 100 most liquid equities by 20-day turnover, 500 within the top 500.

The NSE ticker.

Sessions to average delivery_pct over.

The latest session's derived numbers. Any of them can be null when the symbol's history is too short to compute it.

The session the numbers describe, YYYY-MM-DD.

The session's closing price.

Percent change in close over the last trading day.

Percent change in close over the trailing week.

Percent change in close over the trailing month.

Percent change in close over the trailing three months.

Percent change in close over the trailing six months.

Percent change in close over the trailing year.

Highest high of the last 52 weeks.

Lowest low of the last 52 weeks.

Distance from the 52-week high, percent — 0 means at the high, negative below it.

Distance above the 52-week low, percent.

14-session relative strength index, 0–100.

14-session average true range, in price units.

20-session simple moving average of close.

50-session simple moving average of close.

200-session simple moving average of close.

Percent of the session's traded shares actually taken into delivery rather than squared off intraday — from NSE delivery data.

The session's volume as a multiple of its recent daily average — 2.0 is twice normal.

Traded value for the session, in rupees.

Shares traded in the session.

Mean delivery_pct over the delivery_sessions window requested — a steadier read of the stock's delivery character than any single session.

The NSE ticker.

Bar size. daily, weekly and monthly are accepted as synonyms; anything else is a 400.

One candle per session, oldest first. Prices are split-adjusted, so a long history reads as one continuous series.

The session date, YYYY-MM-DD. For weekly and monthly bars, the period's last trading day.

Opening price of the period.

Highest trade of the period.

Lowest trade of the period.

Closing price of the period.

Shares traded over the period.

The NSE ticker.

A Sift expression naming a value — ema(21), bb(20,2).upper, (high + low) / 2, or a candlestick pattern by name. Repeat the parameter once per series; a comma-separated list would break on bb(20,2). Conditions are not values: count(...), a bare number and a bar offset such as close[-1] are all rejected. At most 24, each at most 200 characters.

Daily only in v1 — indicators are precomputed against the daily metrics, and no weekly or monthly table exists to read them from. daily is accepted as a synonym; 1w and 1mo are a 400 here even though /bars takes them.

YYYY-MM-DD. The series stops at this session. Clamped into the database's range, so an earlier date than the history returns the first session rather than nothing.

How many sessions to return, counting back from as_of. The ceiling is the same 10,000 /bars uses, which is above the longest history in the database — so the default is the whole of it. Indicators are always computed over the full history whatever this says: a shorter window would give a 50-day average a different value from the one the screener filtered on.

The ticker, upper-cased.

Always 1d in v1.

The last session the series covers, after clamping — the date the values were actually computed to, not the one requested.

The session dates, oldest first. With the default sessions, the same dates /bars returns for the symbol, one for one, so a series overlays candles without realigning.

One entry per f, in the order they were sent. An unknown symbol returns no dates and no values rather than a 404, matching /bars.

Which f this answers, counting from 0 in the order they were sent.

The expression as sent, trimmed.

A display name — the catalog's label where the expression names a catalogued field (ema(21) becomes Exponential moving average 21), otherwise the expression itself.

What the values measure — price, percent, ratio, shares — for labelling an axis. Empty for a composite expression, which no catalog entry describes.

flag for a candlestick pattern, whose values are 1 on a session the shape appeared and 0 on every other, and which a chart draws as markers rather than a line. value for everything else.

Exactly as long as d, one entry per session in the same order. null where the value has not warmed up — a 50-bar average is null until 50 bars exist — rather than a zero a chart would draw.

The same coverage caveats a scan returns: a field that only has values for F&O stocks, or an as_of before the dividend record begins.

The Sift query. At most 8,000 characters.

The most liquid N equities by 20-day turnover. 9999 means all equities.

Run as of this session, YYYY-MM-DD. Clamped to the database's range; defaults to the latest session. Scanning a past day costs the same as today.

Maximum rows returned.

Result column to sort by, e.g. turnover or rsi_14. The query's own sort by clause wins if both are given.

Sort direction.

Instrument kinds to include, e.g. ["equity"].

true when the query compiled and ran. false means the compiler refused it — the response then carries error, hint and span instead of rows; see the note below.

The matches, one per symbol, in the requested order. Every row carries the symbol's identity, the session's OHLCV and the standard derived columns below — the same columns whatever the query asked, so a client can parse rows once.

The NSE ticker — feed it to the symbol endpoints for detail.

The company or instrument name.

The instrument class — equity, index, etf.

The session the row describes — an RFC 3339 timestamp at midnight UTC, unlike the plain dates elsewhere. Matches the envelope's asOf.

The session's opening price.

The session's high.

The session's low.

The session's closing price.

Shares traded in the session.

Traded value for the session, in rupees.

Percent of the session's traded shares actually taken into delivery rather than squared off intraday — from NSE delivery data.

The session's volume as a multiple of its recent daily average — 2.0 is twice normal.

14-session relative strength index, 0–100.

50-session simple moving average of close.

200-session simple moving average of close.

14-session average true range, in price units.

14-session average directional index — trend strength, 0–100.

Highest high of the last 52 weeks.

Lowest low of the last 52 weeks.

Distance from the 52-week high, percent — 0 means at the high, negative below it.

Percent change in close over the last trading day.

Percent change in close over the trailing week.

Percent change in close over the trailing month.

Percent change in close over the trailing three months.

Percent change in close over the trailing year.

The smallest liquidity universe the symbol belongs to: 100 means it is among the 100 most liquid equities by 20-day turnover, 500 within the top 500.

How many rows came back — rows.length, for convenience.

true when more symbols matched than limit allowed. Raise the limit, or sort with orderBy so the rows you care about make the cut.

The session the scan actually ran against, YYYY-MM-DD — the requested asOf clamped to a real trading day.

How long the scan took to execute, milliseconds.

The columns the query referenced — useful for showing which data a saved scan depends on.

Non-fatal notes the compiler attached. The query still ran; empty in the common case.

The query as the compiler understood it, canonically spelled — confirm shorthand like 2x parsed the way you meant.

The Sift query to replay. At most 8,000 characters.

Liquidity universe.

How many past sessions to replay.

Forward windows in trading days, each clamped 1..60, at most 4 distinct values. Deduplicated and sorted ascending.

Where the position is priced: close (the signal bar's own close) or nextOpen (the next session's open — the earliest fill a reader of an EOD scan could take). Both exit at the close of bar signal+h.

Index for excess-return measurement: NIFTY 50, NIFTY 100, NIFTY 200, NIFTY 500, NIFTY BANK, NIFTY MIDCAP 150, NIFTY SMALLCAP 250, or none to skip it.

Which horizon the monthly, histogram and equityCurve detail describes. Must be one of horizons.

Simulate a stop-loss this many percent below entry (clamped 0.1..95). Setting this or targetPct switches the replay into a trade simulation: each signal walks its forward bars and exits at the stop, the target, or the horizon close — whichever comes first. When one bar's range covers both levels, the stop wins; that is the conservative reading, stated rather than hidden.

Simulate a profit target this many percent above entry (clamped 0.1..95). Fills at the target price, no slippage.

true when the query compiled and the replay ran; false is a compiler refusal, answered the same way as /v1/scan.

How many past sessions were replayed.

How many of those sessions produced at least one match — a scan that fires on 8 days out of 250 behaves very differently from one that fires every day.

The first replayed session, YYYY-MM-DD.

The last replayed session.

Matches summed over every replayed session — each symbol-day the query fired counts once.

totalSignals / sessions — the scan's average selectivity.

One bucket per forward horizon — 1, 5 and 20 trading days — each measuring what all the signals did over that many sessions after firing.

Trading days ahead the bucket measures.

Signals with enough forward history to measure. Smaller at longer horizons — the most recent signals have not lived 20 sessions yet.

Mean close-to-close return over the horizon, percent.

Median return, percent — compare with the mean to spot a few big winners carrying an otherwise flat scan.

Percent of signals whose return over the horizon was positive.

The single best signal's return, percent.

The single worst signal's return, percent.

Population standard deviation and the interquartile band of the per-signal returns — the shape around the mean.

Mean of the positive returns, mean of the rest (kept signed), and their ratio avgWin / |avgLoss|. Read win rate and payoff as a pair: a 45% win rate is fine at a 2.5 payoff and fatal at 0.8. There is no expectancy field — with these definitions it equals meanPct.

Signals excluded from the horizon because the dataset ends inside their forward window.

Present unless the request said none: the index's own mean return over the same windows (meanPct), the mean and median per-signal excess (meanExcessPct, medianExcessPct), and what share of signals beat it (beatBenchmarkPct).

How long the replay took, milliseconds.

The liquidity universe the replay used, echoed back.

The resolved choices, echoed back — what the numbers were actually computed with, after defaults and clamps.

Present only under a simulated stop/target: how the horizon's trades ended — stop, target and time counts, summing to n.

Every detail-horizon trade, oldest first: d (signal session), symbol, entryPrice, exitD and exitPrice (the session it closed on and the stop/target level or that session's close), retPct, exitBar (sessions held) and exitReason (stop, target or time). Capped at 5,000 rows; tradesTruncated says when the cap bit. The audit trail behind the summary.

The detail horizon by calendar month: month, signals, winRatePct, meanPct and, when benchmarked, meanExcessPct. The regime view — a scan whose whole edge sits in one hot quarter shows it here.

Distribution of per-signal returns at the detail horizon. edges (one more than counts) span the middle 96% of the data; outliers land in underflow/overflow rather than stretching the axis.

An index starting at 100 of equal-weighting every signal at the detail horizon, one point per session with signals, with maxDrawdownPct. Overlap-adjusted: each session's mean h-bar return is spread geometrically over h bars, so overlapping windows do not compound multiple times. method names the approximation — it is an approximation, not a track record. bench points build the same construction from the index's returns on the same sessions.

The caveats that bound these numbers: survivorship lean, thin samples, truncated horizons, missing benchmark days.

Every scan on the account, one object each.

The scan's id — what the :id routes, alerts and the delivery log key on. Stable for the scan's lifetime.

The name it was saved under, unique per account.

The Sift query, exactly as saved.

The liquidity universe the scan runs against.

When the scan was first saved, RFC 3339 UTC.

When it last changed — saving over the name or renaming moves it.

What to call it. At most 120 characters.

The Sift query. At most 8,000 characters.

Liquidity universe the scan runs against.

The scan's id — what the :id routes, alerts and the delivery log key on. Stable for the scan's lifetime.

The name it was saved under, unique per account.

The Sift query, exactly as saved.

The liquidity universe the scan runs against.

When the scan was first saved, RFC 3339 UTC.

When it last changed — saving over the name or renaming moves it.

The scan's id, from the list.

The new name. Non-empty, at most 120 characters.

The scan's id — what the :id routes, alerts and the delivery log key on. Stable for the scan's lifetime.

The name it was saved under, unique per account.

The Sift query, exactly as saved.

The liquidity universe the scan runs against.

When the scan was first saved, RFC 3339 UTC.

When it last changed — saving over the name or renaming moves it.

The scan's id.

Always true — the scan and its alert are gone. A failure is an HTTP error, never ok: false.

Every alert on the account, one object each.

The saved scan this alert watches. Alerts have no id of their own — one scan carries at most one alert, addressed by the scan's id.

The watched scan's name, for display.

Whether the alert runs in the nightly evaluation.

new_matches fires only when a symbol enters the result set; any_match fires whenever anything matches.

Whether matches are sent by email.

The URL matches are POSTed to. Absent when no webhook is configured.

Keys the HMAC-SHA256 signature on every delivery — see the webhooks section. Travels only alongside a configured webhook_url, and never changes once generated.

The alert's health: ok runs normally, broken means the scan no longer compiles, webhook_disabled means the URL failed too many days running and was switched off.

Why the alert is broken, in the compiler's words.

When the nightly evaluation last considered this alert, RFC 3339 UTC. Absent until it has run once.

When the alert's configuration last changed, RFC 3339 UTC.

The account's plan.

How many scans the plan lets you alert on — compare with alerts.length before creating another.

Whether the plan allows webhook delivery.

Only this scan's events. Omit for all scans.

Maximum events to return.

One event per alert per evaluated session, newest first — quiet nights included, so absence of matches is recorded, not silent.

The event's id, unique and ascending.

The saved scan the event belongs to.

The scan's name at evaluation time.

The trading session that was evaluated, YYYY-MM-DD — the day the data describes, not the moment the evaluation ran.

Every symbol the scan matched that session.

The subset of matched that was not in the previous session's results — what a new_matches alert fires on.

Whether notifications went out. false on a quiet night, or when trigger_mode is new_matches and everything that matched had matched before.

When the evaluation ran, RFC 3339 UTC.

The saved scan's id the alert belongs to — there is no separate alert id.

Maximum rows to return.

Attempts, newest first — one row per HTTP request made.

The attempt's id.

The saved scan whose alert made the delivery.

1 for the first try. A failed delivery is retried on a backoff schedule, so one alert firing can log several rows with rising attempt numbers.

The HTTP status your receiver answered. null when no response arrived at all — a timeout or an unreachable host.

true when the receiver answered 2xx.

null on success; otherwise a short classification of what went wrong — the note below lists the vocabulary.

How long the request took, milliseconds.

When the attempt was made, RFC 3339 UTC.

The saved scan's id.

Whether the alert runs at all.

new_matches fires only when a symbol enters the result set; any_match fires whenever the scan matches anything.

Send matches by email.

HTTPS URL to POST matches to. Max plan only. Validated against private-address targets when saved.

The saved scan this alert watches. Alerts have no id of their own — one scan carries at most one alert, addressed by the scan's id.

The watched scan's name, for display.

Whether the alert runs in the nightly evaluation.

new_matches fires only when a symbol enters the result set; any_match fires whenever anything matches.

Whether matches are sent by email.

The URL matches are POSTed to. Absent when no webhook is configured.

Keys the HMAC-SHA256 signature on every delivery — see the webhooks section. Travels only alongside a configured webhook_url, and never changes once generated.

The alert's health: ok runs normally, broken means the scan no longer compiles, webhook_disabled means the URL failed too many days running and was switched off.

Why the alert is broken, in the compiler's words.

When the nightly evaluation last considered this alert, RFC 3339 UTC. Absent until it has run once.

When the alert's configuration last changed, RFC 3339 UTC.

The saved scan's id.

Always true — the alert is gone. A failure is an HTTP error, never ok: false.

The account's id.

The address the account was registered with.

The display name, if one was given at signup.

When the account was created, RFC 3339 UTC.

The plan the key's requests are metered against.

The subscription's standing — active when the plan is in force. Anything else means payment is still settling, billing is paused, or the plan is winding down. Read limits rather than this to decide what the account may do.

How many scans the plan can keep.

How many scans can alert.

Whether alerts may deliver by webhook.

Requests per minute across the API.

Requests per minute for /v1/scan and /v1/hitrate, which cost more than the rest.

How many active API keys the account may hold.

Scans currently saved — against limits.max_saved_scans.

Scans currently alerting — against limits.max_alert_scans.