Skip to content

Telegram Signals Bot Guide

telegramsignalsbankbot is a trading bot that reads AI-summarized Telegram messages from the database, classifies them as BUY/SELL entry signals, and executes trades automatically.

It is the downstream consumer of the telegram-monitor CronJob — the monitor fills the telegram_messages table, the signals bot acts on it.

How It Works

  1. Queries telegram_messages where channel = '-1001998690333', symbol IS NOT NULL, acted_on = FALSE, published_at within last 3 days
  2. For each message, sets acted_on = TRUE immediately (crash-safe — won't re-process even if bot dies mid-run)
  3. Calls AI (run_ai_simple) to classify: is this a new entry signal or noise (TP hit / analysis / recap)?
  4. AI also returns the correct Yahoo Finance ticker — no hardcoded symbol map needed
  5. Executes buy(yf_ticker, quantityUSD=cash * 0.20) or sell(yf_ticker) via the Bot class

Position Sizing

POSITION_SIZE_PCT = 0.2 — spends 20% of available cash per signal.

With $10,000 starting capital and multiple signals: - Signal 1 BUY BTC-USD → $2,000 - Signal 2 BUY ^GSPC → $1,600 (20% of remaining $8,000) - Up to ~5 positions before cash runs low

Sells use all of the holding (sell(symbol) with no quantityUSD).

Signal Classification

The AI classifier receives the raw message text + the pre-computed summary from telegram_monitor. It returns:

{"is_signal": true, "direction": "BUY", "yf_ticker": "BTC-USD"}

Classified as signals: messages with an explicit entry price and BUY/SELL direction.

Not classified as signals (correctly skipped): - Take-profit hit notifications ("TP1 HIT ✅") - Performance recaps ("Performance 09.03.26 🏆") - Market analysis without a concrete entry - General commentary

The AI resolves broker-specific CFD names (US500, DJ30.c, XAUUSD) to Yahoo Finance tickers. yf.Search() was evaluated but doesn't understand these names — AI resolution is more reliable.

Deduplication

acted_on = TRUE is written per message in its own DB transaction, before the AI call and before any trade. This means:

  • If the bot crashes mid-run, already-evaluated messages stay acted_on = TRUE
  • No message is ever processed twice, regardless of crash timing
  • No in-memory ID lists or portfolio-dict tracking needed

Configuration

# telegramsignalsbankbot.py
CHANNEL_ID = "-1001998690333"   # The Signals Bank - FREE
LOOKBACK_DAYS = 3               # Only look at messages from last N days
POSITION_SIZE_PCT = 0.2         # 20% of cash per signal

To monitor a different channel, change CHANNEL_ID to the numeric Telegram channel ID (find with onetime_getTelegramChats.py).

Helm Deployment

In helm/tradingbots/values.yaml:

- name: telegramsignalsbankbot
  schedule: "*/5 * * * *"  # Every 5 minutes — matches telegram-monitor cadence

The bot runs every 5 minutes. If telegram-monitor just fetched new messages, this bot will pick them up within 5 minutes.

No special env vars beyond the standard set — it reads from the same postgres database as all other bots.

Key Files

File Purpose
tradingbot/telegramsignalsbankbot.py Bot implementation
tradingbot/utils/db.py TelegramMessage model (incl. acted_on)
helm/tradingbots/values.yaml Schedule configuration

See Also