惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

月光博客
月光博客
C
Check Point Blog
J
Java Code Geeks
腾讯CDC
Apple Machine Learning Research
Apple Machine Learning Research
宝玉的分享
宝玉的分享
Microsoft Azure Blog
Microsoft Azure Blog
WordPress大学
WordPress大学
量子位
Google DeepMind News
Google DeepMind News
I
InfoQ
The GitHub Blog
The GitHub Blog
aimingoo的专栏
aimingoo的专栏
N
Netflix TechBlog - Medium
Hugging Face - Blog
Hugging Face - Blog
博客园 - Franky
V
V2EX
Blog — PlanetScale
Blog — PlanetScale
T
The Blog of Author Tim Ferriss
小众软件
小众软件
博客园_首页
人人都是产品经理
人人都是产品经理
博客园 - 聂微东
IT之家
IT之家

DEV Community

Authentication Security Deep Dive: From Brute Force to Salted Hashing (With Java Examples) Why AI Systems Don’t Fail — They Drift Spilling beans for how i learn for exam😁"Reinforcement Learning Cheat Sheet" I Replaced Chrome with Safari for AI Browser Automation. Here's What Broke (and What Finally Worked) How Python Borrows Other People's Work The $40 Architecture: Processing 1 Billion API Requests with 99.99% Uptime Vibe Coding: A Workflow Guide (From Zero to SaaS) Most webhook security guides protect the wrong side. The scary part is delivery. Headless CMS for TanStack Start: Build a Blog with Cosmic EU Age Verification App "Hacked in 2 Minutes" — What Actually Happened Comfy Cloud’s delete function does not actually remove files Running AI Models on GPU Cloud Servers: A Beginner Guide Event-driven media intelligence with AWS Step Functions and Bedrock I scored 500 AI prompts across 8 quality dimensions — here's what broke How to Call Google Gemini API from Next.js (Free Tier, No Backend Needed) The Portal Protocol: Reclaiming Human Connection in the Age of AI How to Fix Your Team's Scattered Knowledge Problem With a Self-Hosted Forum Intro to tc Cloud Functors: A Graph-First Mental Model for the Modern Cloud Designing Multi-Tenant Backends With Both Ownership and Team Access I Built a Neumorphic CSS Library with 77+ Components — Here's What I Learned PostgreSQL Performance Optimization: Why Connection Pooling Is Critical at Scale Cómo construí un SaaS multi-rubro para gestionar expensas en Argentina con FastAPI + Vue 3 🚀 I Built an Ethical Hacking Scanner Tool – Open Source Project I Replaced /usage and /context in Claude Code With a Single Statusline A Pythonic Way to Handle Emails (IMAP/SMTP) with Auto-Discovery and AI-Ready Design I Collected 8.9 Million Polymarket Price Points — Here's What I Found About How Markets Really Move EcoTrack AI — Carbon Footprint Tracker & Dashboard Everyone's Using AI. No One Agrees How. 5 self-hosted ebook managers worth trying in 2026 Building Your First AI Agent with LangChain: From Chatbot to Autonomous Assistant
Building an AI-Powered Quantitative Trading System with H...
Michael Zhan · 2026-05-04 · via DEV Community

Michael Zhang

Building an AI-Powered Quantitative Trading System with Hermes Agent and IBKR

How I set up a multi-signal ETF trading bot that runs on autopilot — and the 7 things that broke along the way


I wanted a system that watches the market 24/7, analyzes technical indicators across multiple ETFs, and executes trades automatically. No manual chart-checking. No emotional decisions. Just cold, calculated signals.

Here's what I built and everything I learned.

The Stack

Twelve Data (market data) → Python strategy engine → IB Gateway → IBKR
                                    ↑
                            Hermes Agent (cron scheduling)

Enter fullscreen mode Exit fullscreen mode

  • Market Data: Twelve Data API (free tier, 800 calls/day)
  • Execution: ib_insync Python library → IB Gateway → Interactive Brokers
  • Scheduling: Hermes Agent cron jobs, every 3 minutes
  • Dashboard: Real-time HTML dashboard with signal indicators and market news
  • Strategy: 6-condition buy signal, 3-condition sell trigger

The Strategy: Why 6/6, Not 2/3

Most retail strategies use 2-3 indicators. I chose 6 for a reason: every additional condition exponentially reduces false signals.

Here are the six gates a stock must pass before I'll buy:

# Indicator Condition What It Means
1 RSI < 30 Oversold bounce The stock has been beaten down
2 Price ≤ Lower Bollinger At support level It's at a statistical floor
3 Price > MA20 Uptrend The direction is right
4 ADX > 20 Trending, not ranging There's actual momentum
5 MACD histogram turning positive Momentum shift Bears are becoming bulls
6 Volume confirmation Real interest It's not a fake-out

Sell triggers are simpler — any one of three fires: RSI overbought, price hitting upper Bollinger, or MACD death cross. Getting out fast matters more than getting in perfectly.

I track three ETFs: SPY, QQQ, IWM. One position at a time. $1,000 paper account.

Setting Up IB Gateway: The 3-Hour Rabbit Hole

Installing IB Gateway for headless automated trading should take 15 minutes. It took me three hours. Here's what nobody tells you.

1. Use the OFFLINE installer, not the auto-updating one

IBC (Interactive Brokers Controller — the tool that types your password into the Gateway's login screen) does not work with the auto-updating version. You MUST download the "Stable Standalone" build:

https://download2.interactivebrokers.com/installers/ibgateway/stable-standalone/ibgateway-stable-standalone-linux-x64.sh

Enter fullscreen mode Exit fullscreen mode

2. JavaFX: The silent killer

IB Gateway 10.45 uses JavaFX for its GUI. System OpenJDK does NOT include JavaFX. If you start Gateway via IBC with system Java, it crashes in 0.5 seconds with a cryptic NullPointerException.

The fix: IB Gateway bundles its own Azul Zulu JRE (which includes JavaFX). Leave JAVA_PATH= empty in IBC's config, and it auto-discovers the bundled JRE from .install4j/inst_jre.cfg. Do not set JAVA_PATH=/usr/bin.

3. IBC strips -D VM options. Patch it.

IBC's ibcstart.sh explicitly filters out ALL -D prefixed options from ibgateway.vmoptions (line 322: ! "${line:0:2}" = "-D"). Gateway 10.45 NEEDS -DinstallDir and -DvmOptionsPath to initialize. Without them, it silently exits after "LauncherFontUpdater."

You must add these directly to the java_vm_options variable in ibcstart.sh:

java_vm_options="$java_vm_options -DinstallDir=${program_path}"
java_vm_options="$java_vm_options -DvmOptionsPath=${program_path}/ibgateway.vmoptions"

Enter fullscreen mode Exit fullscreen mode

4. Paper trading credentials are separate from Live

I spent 30 minutes debugging "invalid username or password" because IBKR's paper trading account has its OWN username and password, different from your live login. Find them in: Client Portal → Settings → Paper Trading Account.

5. Read-Only API mode

When you first log in manually, the Gateway's API is in Read-Only mode. ib_insync connects but every order fails with Error 321: The API interface is currently in Read-Only mode. Fix: Configuration → API → Settings → uncheck "Read Only API."

6. Live accounts require 2FA

Live trading accounts need IBKR Mobile authentication. IBC handles the 2FA flow but you MUST acknowledge the push notification on your phone within 180 seconds. For automated paper trading, no 2FA needed — use a dedicated paper trading username.

7. WSLg focus events flood the log

Running in WSL2 with WSLg, the Gateway window generates hundreds of focus/lost-focus events. IBC logs them all. It doesn't break anything but makes reading the log... noisy.

The Code

The full strategy is a Python script that:

  • Fetches 5-minute candles from Twelve Data
  • Calculates all 6 indicators locally with NumPy
  • Connects to IB Gateway via ib_insync
  • Checks buy/sell conditions
  • Executes market orders on signal

Core signal check:

def check_buy(rsi, price, sma, upper, middle, lower, adx, ml, sl, hist, ph, vol):
    return {
        "RSI oversold":    rsi < 30,
        "At lower band":   price <= lower * 1.02,
        "Uptrend":         price > sma,
        "Trending":        adx > 20,
        "MACD turning":    hist > 0 and ph < hist,
        "Volume OK":       vol is not None,
    }

Enter fullscreen mode Exit fullscreen mode

All six must be True. If any one fails, no trade. This filters out ~95% of false signals compared to a simple RSI-only strategy.

The Dashboard

Dashboard showing three ETF cards with signal indicators

I built a real-time HTML dashboard that shows:

  • Each ETF's 6 signal lights (green ✓ / red ✗) with hover explanations
  • Progress bar toward 6/6
  • RSI, ADX, MA values
  • Market status (open/closed/weekend)
  • Latest Yahoo Finance news
  • Auto-refresh every 3 seconds

The strategy script writes a data.json file each run, and the dashboard reads it. No server-side rendering needed — just static HTML + JavaScript.

Hermes Agent Cron Scheduling

Hermes Agent handles scheduling. One command:

hermes cron create "*/3 * * * *" --prompt "Run the strategy script"

Enter fullscreen mode Exit fullscreen mode

The agent executes the Python script, connects to IB Gateway, gets market data, evaluates signals, and places orders — all autonomously. If a trade happens, I get notified.

Results (Paper Trading)

After one weekend of running:

  • 0 trades executed (markets closed)
  • 780 API calls saved (rate limit not wasted on weekends)
  • Dashboard functional with live market status detection

Monday will be the real test.

Key Takeaways

  1. Offline installer, not auto-updating. This alone saved what could have been hours of debugging.
  2. Let IBC find its own Java. Don't override JAVA_PATH. The bundled JRE has JavaFX; your system JRE probably doesn't.
  3. Patch ibcstart.sh. IBC intentionally strips -D options. Gateway 10.45 needs them. Add them manually.
  4. 6/6 is strict but worth it. Each additional condition exponentially reduces noise.
  5. Paper trading credentials are separate from Live. This one cost me 30 minutes of confusion.
  6. Build a dashboard. You can't debug what you can't see. Visual feedback for every indicator state is invaluable.

This setup runs on a $1,000 paper trading account. Full code and configuration available at maiweb.co.nz.

Built with Hermes Agent, ib_insync, Twelve Data, and too much coffee.