Trading Bots

Intraday Bot

9-strategy multi-signal engine for day trading.

stockwise bot 25000 2       # ₹25K capital, 2% daily target
stockwise -m us bot 5000 1  # US market, $5K, 1% target

Strategies

Scoring

• Each strategy generates a strength score (0-100)
• Consensus bonus: +8 per additional agreeing strategy
• Volume multiplier: 1.15x if volume >1.5x average
• Regime multiplier: 1.2x if trade aligns with strong trend, 0.6x if counter-trend
• Confidence: HIGH (score >=75, 3+ strategies), MEDIUM (>=60, 2+), LOW (others)

Position Sizing (Kelly Criterion)

• Half-Kelly fraction for safety (max 25% of capital)
• 1% max risk per trade
• Never more than 30% of capital in one position
• At most 5 plans, and together they never need more than your capital: walking down the ranking, a plan that doesn’t fit in what’s left is shrunk to what does (or dropped)
• Exit level for each paper position: uses support levels when available, else ATR-based

Output

• Sector heat map (which sectors are moving)
• Market regime detection (Strong Uptrend → Strong Downtrend)
• Top 15 signals ranked by score
• The strategies behind the five strongest signals, with each one’s strength and reason (intraday shows no position sizes or price levels; the paper-trading daemon and sim still size simulated trades and show them as long or short with level 1, level 2 and an exit level)

Long-Term Scores

6-pillar scoring of the market’s stock universe with a Monte Carlo illustration. It shows data and how each score was built: no buy, sell or hold calls, no entries, stops or targets, and no allocation advice.

stockwise longterm 15000         # ₹15K/month SIP budget
stockwise -m us longterm 1000    # US market, $1K/month

6 Scoring Pillars

Moat Analysis

Stocks get Wide/Narrow/No Moat signals from these rules:

• Profit margins >25% (pricing power)
• ROE >25% (capital efficiency)
• Revenue growth >25% (market dominance)
• Market cap >$1T (scale advantage)

Monte Carlo Simulation

• 1000 simulated 5-year paths using historical volatility
• Reports: P10 (worst case), Median, P90 (best case) returns
• Uses deterministic PRNG for reproducibility

Output

• Top 12 names ranked by composite score, with each pillar’s score (a dash where the pillar had no data; n/a when under half the weight had data)
• Detailed top 3: score band, data coverage, risk level (Lower / Moderate / Higher), PEG, earnings yield, and the facts that added or took away points
• 5Y and 10Y growth at an assumed annual return set from the score (not a forecast), with dividends reinvested
• SIP arithmetic (the monthly amount that would reach ₹10L in 10 years at that rate, and what your budget compounds to)
• A simulated 5-year range: 10th percentile, median and 90th percentile

Paper Trading Simulator

Test the intraday bot without risking real money.

stockwise sim start 25000 2    # Morning: lock in trades
stockwise sim status           # During day: live P&L
stockwise sim settle           # After the close: replay the day, record results
stockwise sim history          # Performance over time

sim start only runs on a trading day before the close (the clock is IST, or ET with -m us, and knows exchange holidays), and records when each trade was entered.

How sim settle decides

Each open session is settled from its own day’s 5-minute bars, not from whatever the price is when you run the command:

1. Bars before the entry time are ignored.
2. The bars are walked with the same position manager as the daemon: book half at level 1 and move the exit level to the opening price, close the rest at level 2, trail the exit level after level 1, tighten it in the wind-down. (Levels are fixed when the paper position opens; they are simulation mechanics, not advice.)
3. Whichever of the exit level or level 2 the price reached first decides the trade. Inside a single bar the order of the high and low is unknown, so a bar that touched both counts as the exit level (conservative). Levels fill at their price; a bar that opens beyond one fills at the open.
4. Otherwise the trade exits at the open of the 15:15 IST bar (15:45 ET).

Yahoo keeps 5-minute bars for about 60 days. Older sessions fall back to hourly bars, then to the daily bar (same rules, exit at the close). Today’s session is refused until the market closes; sim settle --now settles it at the latest bar. A session whose bars can’t be fetched stays open and the command exits non-zero if nothing was settled.

What It Tracks

• Per trade: opening and closing price, whether level 2 or the exit level was reached, P&L, strategies used
• Per day: total P&L, win/loss count, how often level 2 was reached
• Cumulative: equity curve, Sharpe ratio, win streaks, day/trade win rates

Confidence Rating (after 5+ days)

Practice mode

To practise your own decisions rather than the bot’s, replay a past market day by day with the future hidden — the COVID crash, 2008, the dot-com bust — take on goal-based challenges, and get a risk-adjusted score, badges and a shareable card: stockwise sim replay start covid-crash-2020. See simulator.md.

Daemon Mode

Guide: Set up alerts and the daemon.

Persistent background workers that run the bots continuously.

Intraday Daemon

stockwise daemon intraday 25000         # NSE, ₹25K paper capital
stockwise -m us daemon intraday 5000    # NYSE hours (ET), $5K
stockwise daemon intraday 25000 --once  # one session, then exit

The daemon keeps running: after it settles a session it waits for the next trading day (skipping weekends and exchange holidays) and runs that one, so one process covers every day until you stop it with Ctrl+C or SIGTERM. Start it any time: before or during a session it picks that session up (a late start enters no new positions); after the close or on a weekend it waits for the next pre-market. Add --once to run a single session and exit with status 0 (for example from cron or a systemd timer).

Lifecycle (NSE; NYSE uses the same offsets from 09:30–16:00 ET):

9:00 AM   Pre-market scan (sectors, gaps)
9:15 AM   Enter up to 5 positions (9 strategies, Kelly sizing, ≤ capital in total)
9:30–3:15 Monitor every 60 seconds:
          - Level 1 reached → book 50%, move the exit level to the opening price
          - Level 2 reached → close fully
          - Exit level reached → close, record the loss
          - Trailing exit level updates (1.5% trail)
          - Wind-down (2 PM+): tighten the exit level to 0.5%
3:15 PM   Square off all remaining positions
3:20 PM   Settle → save to sim_history.json → AI EOD report → wait for the
          next session (with --once: exit)

The clock is the exchange’s, not your machine’s: IST (+05:30) for NSE, US Eastern with daylight saving for NYSE. Weekends and exchange holidays are closed all day; the daemon prints the reason and the next session, then sleeps until 15 minutes before it.

Risk controls:

• 1% max risk per trade, 30% max per position, total allocation ≤ capital
• 3% daily loss kill switch (closes everything; counts every realised exit, including level 2 exits)
• Max 5 concurrent positions
• Entries only in the first 15 minutes; no new entries after that
• Forced square-off 15 minutes before the close

Resilience:

• One Yahoo session is reused across ticks. A failed tick is logged and retried after 15 s, 30 s, 60 s … (capped at 5 minutes); after three failures in a row the session is rebuilt. Network blips never stop it.
• The day’s state (positions, risk manager, entries) is saved to daemon_state.json (daemon_state_us.json with -m us) in the data directory after every tick. Restarting the daemon the same day resumes it — the capital from the saved state wins.
• Ctrl+C (or SIGTERM, e.g. from systemd) squares off open positions at their last price, records the day in the simulator history, saves, and exits 0. A second Ctrl+C exits immediately; the saved state is resumed next time.
• If the daemon missed the square-off (crash, laptop asleep), it hands that day’s entries to the simulator as an open session instead of closing them at stale prices: run stockwise sim settle to settle them from the day’s 5-minute bars.
• One NSE and one NYSE daemon can run side by side; don’t start two for the same market — they would share one state file.

The daemon is paper-only: it never calls a broker order API. Results are saved to simulation history — check with stockwise sim history. With notifications configured, it also sends entries, exits, the kill switch and the end-of-day summary.

Market calendar

Holidays come from a table in src/clock.rs: the NSE equity-segment list (2026 official; 2027 provisional — only fixed-date holidays until NSE publishes its circular in December) and the NYSE calendar for 2026–2028, including 1:00 pm early closes. When the current year’s list is missing or provisional, the daemon prints a warning. Muhurat trading sessions are not traded.

Long-Term Daemon

stockwise daemon longterm

Runs 5 steps:

1. Score 30 stocks on 6 pillars + Monte Carlo and print the three highest scores
2. Check portfolio value, snapshot wealth tracker
3. Check all price alerts
4. List holdings priced below their average cost
5. Generate AI research notes via Ollama

Run daily after market close, or weekly for deep analysis.

Deep Dive

One command for everything about a single stock:

stockwise deep RELIANCE

Shows: company info, price action, valuation (P/E, PEG, P/B, EV/EBITDA), profitability, financial health, dividends, analyst ratings, 1Y chart, all technicals (RSI + gauge, SMA, MACD, Bollinger, ATR, VWAP), support/resistance, Fibonacci, risk (vol, Sharpe, drawdown, VaR), OBV trend, open gaps, insights, the 6-pillar score, and (with --ai) AI analysis.

Price and Technical Alerts

Guide: Set up alerts and the daemon walks through alerts, notifications and cron.

stockwise alert add TCS above 4200
stockwise alert add AAPL -m us below 180
stockwise alert add INFY rsi-below 30         # RSI(14) at or below 30
stockwise alert add RELIANCE cross-above-sma 200
stockwise alert add AAPL -m us breakout-high 52
stockwise alert add HDFCBANK change-below 3   # a fall of 3% or more today
stockwise alert add TCS volume-spike 2        # 2x the 20-day average volume
stockwise alert add TCS results-in 7          # results due within a week
stockwise alert watch                 # check every 60 s while markets are open
stockwise alert watch --interval 30   # every 30 s (minimum 10)
stockwise alert watch --once          # one pass over every alert, for cron

alert watch checks each alert only while its market is open — .NS/.BO symbols on the NSE clock, US tickers on the NYSE clock, crypto and FX (BTC-USD, USDINR=X) around the clock — and sleeps until the next open otherwise. Alerts are re-read every pass, so alert add in another terminal takes effect without a restart.

Technical conditions (RSI, moving-average crosses, breakouts, day change, volume spikes) are evaluated on daily bars, fetched once per symbol per check and brought up to the live price. A cross fires on the session it happens in, not on the days after. results-in reads the next results date from stockwise calendar (Yahoo, plus NSE board meetings for Indian stocks, cached for a day). Every condition is listed in commands.md.

Each trigger is announced once: an alert fires when its condition becomes true, re-arms after the condition turns false again, and fires at most once a trading day. Day-change and volume-spike alerts describe one session, so they re-arm with each new session, using the quote’s own session date. Until today’s session has traded they give no reading, so a weekend or holiday does not repeat Friday’s move; results-in re-arms once the results date has passed. That memory lives in alert_watch.json, so a restart doesn’t repeat notifications. If every configured channel fails, the trigger is not marked and the next check tries again. Network errors are retried with backoff; Ctrl+C stops it.

Notifications

alert watch and daemon intraday send to every channel you configure. All channels are opt-in, set with environment variables:

The same settings can live in the config file under [notify], named without the STOCKWISE_ prefix in lower case (telegram_chat_id = "123", smtp_host = "smtp.gmail.com"). An environment variable wins over the file. If you put a token or password there, keep the file private (chmod 600); config show and doctor warn otherwise.

stockwise notify          # which channels are configured (secrets are never shown)
stockwise notify test     # send a test message to each; exits 1 if any fails

Security: the bot token, webhook URLs and SMTP password are only sent to their own endpoint. Redirects are not followed, error messages are stripped of URLs, and notify shows only the chat id, the webhook’s host, or the SMTP server. Email uses TLS (rustls); STOCKWISE_SMTP_TLS=none with a password is refused unless the server is localhost. A failed notification is printed as a warning and never stops the daemon or the alert loop.

Telegram setup: message @BotFather → /newbot → copy the token; send your bot a message, then open https://api.telegram.org/bot<TOKEN>/getUpdates in a browser and copy chat.id.