BotTrader

Autonomous intraday trading bot (Elixir/OTP, always-on). Runs every 5 minutes 24/7 on BR stocks, US stocks, and crypto: 15m candles, deepseek-v4-flash signals via an internal Hermes MCP link (localhost only), market-wide news each run plus deeper news on volatility triggers, paper trades with fees/slippage and hard risk limits (incl. 15-min min-hold), SQLite history, Telegram announcements and slash commands. Daily 21:30 UTC deep run uses deepseek-v4-pro and sends the digest. After 30 days an evaluation gate emits a go/no-go verdict (return ≥ 2%, max drawdown ≤ 5%, ≥ 10 trades); on PASS the bot stays paper until a broker adapter is attached.

Commands

mix bot.backtest   # ~90-day indicator-only backtest on the first watchlist symbol
mix test           # test suite
mix run --no-halt  # production entrypoint (scheduler + telegram poller)

Telegram slash commands

• /status — equity, positions, last run age
• /hour — equity change over the last hour
• /day — today's trades, P&L and open positions
• /week — last 7 days summary
• /month — last 30 days summary
• /force — run the pipeline now (queued if a run is executing)
• /positions [N] — open positions + paginated trade history

Environment Variables

Variable	Default	Description
DEEPSEEK_API_KEY	— (required)	API key for the OpenAI-compatible endpoint
DEEPSEEK_BASE_URL	https://api.deepseek.com/v1	Base URL (may point to the opencode-go endpoint)
DEEPSEEK_MODEL	deepseek-v4-pro	Model name (direct backend)
LLM_BACKEND	hermes	hermes (MCP) or direct (HTTP)
HERMES_MCP_BIN	hermes	Hermes binary for the MCP child
HERMES_MCP_ARGS	mcp	Comma-separated args for the MCP child
LLM_MODEL_FLASH	deepseek-v4-flash	Model for standard/forced runs
LLM_MODEL_PRO	deepseek-v4-pro	Model for the daily deep run
RUN_INTERVAL_MS	300000	Scheduler tick interval
ANALYSIS_INTERVAL	15m	Candle interval
VOLATILITY_THRESHOLD	0.02	15m move triggering per-symbol news
DAILY_CALL_BUDGET	2400	LLM calls/day before a Telegram alert
MIN_HOLD_MINUTES	15	Minimum holding time per position
DEEP_RUN_HOUR_UTC	21	Deep run hour (UTC)
DEEP_RUN_MINUTE_UTC	30	Deep run minute (UTC)
TELEGRAM_BOT_TOKEN	— (required)	Telegram bot token
TELEGRAM_CHAT_ID	— (required)	Chat id receiving announcements
BOT_STATE_DIR	./data	State directory (Railway: /data)
WATCHLIST_PATH	config/watchlist.json	Seed watchlist file (first boot only; watchlist persists in SQLite)
UNIVERSE_PATH	config/market_universe.json	Full US+BR market universe for scanning
UNIVERSE_SCAN_ENABLED	true	Adds one candidate to the watchlist each run
MOVER_COUNT	5	Top movers deep-dived per run (market hours)
MAX_MOVER_POSITION_PCT	0.10	Max capital per mover position
MARKET_HOURS_ENABLED	true	Skip mover screen when markets closed
UNIVERSE_QUOTE_CHUNK	50	Symbols per batch quote request
UNIVERSE_VOLUME_FLOOR	1000000	Volume threshold for the candidate score bonus
HERMES_MEMORY_PATH	/data/hermes_memory.md	Trade memory file absorbed by Hermes
TRADINGVIEW_ENABLED	true	Enable the internal TradingView scraper
TRADINGVIEW_BATCH_SIZE	25	Symbols scraped per round-robin tick
TRADINGVIEW_MAX_CONCURRENCY	2	Concurrent Playwright pages
CRYPTO_FEE	0.001	Crypto fee rate (0.1%)
US_FEE_USD	1.0	US stock flat fee (USD)
B3_FEE_BRL	5.0	B3 stock flat fee (BRL)
SLIPPAGE	0.0005	Slippage (0.05%)
USD_BRL_RATE	5.5	USD→BRL conversion rate
MAX_POSITION_PCT	0.25	Max allocation per position
MAX_POSITIONS	6	Max concurrent positions
STOP_LOSS_PCT	0.05	Stop-loss threshold
DAILY_LOSS_PCT	0.03	Daily loss limit (blocks new buys)
LLM_CONFIDENCE_THRESHOLD	0.6	Min confidence for BUY/SELL
START_CAPITAL_BRL	1000.0	Paper capital
GATE_MIN_RETURN	2.0	Gate: min return %
GATE_MAX_DRAWDOWN	5.0	Gate: max drawdown %
GATE_MIN_TRADES	10	Gate: min trade count
GATE_DAYS	30	Gate: evaluation window in days
CANDIDATE_CAP	3	Max new LLM candidates per day
YAHOO_BASE_URL	https://query1.finance.yahoo.com	Yahoo chart API base
COINGECKO_BASE_URL	https://api.coingecko.com	CoinGecko API base

Railway Deployment

1. Create a Railway project and deploy this repo (Dockerfile builds Elixir + Hermes; railway.toml runs mix run --no-halt with restartPolicy ALWAYS).
2. Attach a volume at /data (service → Settings → Volumes, or railway volume add -m /data).
3. Set the env vars above as service variables (OPENCODE_GO_API_KEY, TELEGRAM_BOT_TOKEN, TELEGRAM_CHAT_ID at minimum; BOT_STATE_DIR=/data). The entrypoint seeds ~/.hermes/.env from OPENCODE_GO_API_KEY.
4. Verify with MIX_ENV=prod mix compile locally.
5. Rollback: archived JSON (*.json.archived) remains in /data; redeploy the previous image.