QuantDinger v5 AI Trading Research and Backtesting

QuantDinger v5.8.4 is a self-hosted AI trading research platform with market analysis, Python strategies, backtesting, paper and live trading and an MCP server. This template runs upstream's full stack as 15 services, with paper trading only by default. The web UI has a non-commercial licence.

What you get

  • QuantDinger backend v5.8.4 as API plus workers: trading worker, strategy dispatcher and evaluator, scheduler and Celery workers
  • Web UI behind nginx and a token-protected MCP endpoint, the only public services
  • PostgreSQL 18, two Redis instances (cache and durable jobs) and single-node Kafka
  • First administrator created from generated ADMIN_USER and ADMIN_PASSWORD; public sign-up is off
  • Live trading off by default (LIVE_TRADING_ENABLED=false), and agent tokens paper-only

Good for

  • Researching markets with AI analysis using your own OpenAI-compatible model key
  • Writing Python strategies and backtesting them
  • Paper trading against live market data before deciding about real money
  • Letting AI agents research, backtest and paper-trade through the MCP server

Services in this template

  • Kafka · image apache/kafka:4.3.1@sha256:77e3df9054047a88b520d0cc46e16696d3b22022e1d580aeccd2632df6532837
  • Redis Jobs · image redis:8.2
  • Celery Backtest Worker · from baranberkay96/quantdinger-railway
  • Redis · image redis:8.2
  • MCP Server · from baranberkay96/quantdinger-railway
  • Scheduler · from baranberkay96/quantdinger-railway
  • Postgres · image ghcr.io/railwayapp-templates/postgres-ssl:18
  • Kafka Audit Worker · from baranberkay96/quantdinger-railway
  • Strategy Evaluator · from baranberkay96/quantdinger-railway
  • Celery Beat · from baranberkay96/quantdinger-railway
  • Web · from baranberkay96/quantdinger-railway
  • Backend · from baranberkay96/quantdinger-railway
  • Strategy Dispatcher · from baranberkay96/quantdinger-railway
  • Trading Worker · from baranberkay96/quantdinger-railway
  • Celery Worker · from baranberkay96/quantdinger-railway

Questions people ask

Can I use it commercially?

Check the licence first. The web UI is under the QuantDinger Frontend Source-Available License: free only for non-commercial use, revocable, and requires the Powered by QuantDinger branding to stay. Commercial use needs a commercial licence from Open Byte Inc. The backend, workers and MCP server are Apache-2.0. This is not legal advice.

How do I log in?

Wait until Backend is healthy; the first deploy builds images and migrates the database. Copy ADMIN_USER (default admin) and ADMIN_PASSWORD from Backend > Variables, sign in on the Web domain and change the password under your profile.

Is live trading safe by default?

Live trading is off, but manual Quick Trade orders are not covered by that switch: they go to the exchange once you save exchange API keys. Use trade-only keys without withdrawal rights, restrict them by IP where possible and start small. Backtests and paper results do not predict real returns.

Do I need AI keys?

No, the stack runs without any. For AI analysis set LLM_PROVIDER and a key on Backend, which all workers reference. Model usage is billed by your provider.

How do I enable MCP?

The MCP endpoint answers 503 until connected. Issue an Agent Token in QuantDinger (paper-only is recommended), paste it into MCP Server as QUANTDINGER_AGENT_TOKEN and redeploy that service.

Full template README

Deploy and Host QuantDinger with Railway

QuantDinger is a self-hosted AI trading research platform: AI market analysis, Python strategies, factor research, backtesting, paper and live trading, monitoring and an MCP server for AI agents. This community template runs release v5.8.4 as upstream's full production stack, with paper trading only by default.

Licence notice. The QuantDinger web UI (and the H5 mobile client, not included here) is published by Open Byte Inc. under the QuantDinger Frontend Source-Available License. That licence is free only for non-commercial use, is revocable, and requires all "Powered by QuantDinger" branding to stay in place. If you run QuantDinger commercially, you need a commercial licence from Open Byte Inc. (support@quantdinger.com). The backend, workers and MCP server are Apache-2.0. This is a community template, not affiliated with or endorsed by Open Byte Inc. This is not legal advice; read the licences yourself.

Not investment advice. This template hosts software. Backtests, AI output and paper results do not predict real returns. Live trading can lose money, and that risk is entirely yours.

About Hosting QuantDinger

QuantDinger is not a single container. The API serves the web UI. Strategies run in a separate trading worker. Bar-driven strategies are dispatched and evaluated through Kafka. AI jobs and backtests go through Celery, with a durable Redis that never evicts jobs. A scheduler handles market data and portfolio monitoring. This template deploys that stack as upstream's docker-compose describes it: 15 Railway services, private networking, volumes for the database, the job queue and Kafka, and the database migration before the API starts. Only the web UI and the token-protected MCP endpoint are public. The first administrator is created from generated variables, so there is no default password.

Common Use Cases

  • Research markets with AI analysis using your own OpenAI-compatible model key
  • Write Python strategies and indicators, then backtest them on crypto, US, HK and other markets
  • Paper-trade strategies against live market data before deciding about real money
  • Let AI agents (Cursor, Claude, Codex) research, backtest and paper-trade through the MCP server
  • Run a private quant workspace for yourself or a small team

Dependencies for QuantDinger Hosting

  • QuantDinger backend v5.8.4 (ghcr.io/openbyteinc/quantdinger-backend, Apache-2.0): API, eight workers and migrations
  • QuantDinger web UI 5.8.4 (ghcr.io/openbyteinc/quantdinger-frontend, source-available licence)
  • QuantDinger MCP server (built from the v5.8.4 source, Apache-2.0)
  • PostgreSQL 18
  • Redis 8.2, two instances: a cache and durable jobs
  • Apache Kafka 4.3.1 (single-node KRaft)

Deployment Dependencies

Implementation Details

Service Role Public
Web Web UI (nginx), proxies /api to Backend yes
Backend Flask API (gunicorn); pre-deploy runs the migration no
Trading Worker Strategy runtimes, order execution, position sync no
Strategy Dispatcher / Strategy Evaluator Kafka-driven bar-close strategy evaluation no
Kafka Audit Worker Runtime event audit trail no
Scheduler Market data, portfolio monitor, periodic jobs no
Celery Worker / Celery Backtest Worker / Celery Beat AI jobs, maintenance, backtests (1.5 GB cap each), schedules no
MCP Server MCP endpoint /mcp, bearer token required yes
Postgres, Redis, Redis Jobs, Kafka Database, cache (LRU), durable job queue (AOF, no eviction), event backbone no

First login

  1. Wait until Backend is healthy. The first deploy builds the images, migrates the database and seeds the market catalog, which takes a few minutes.
  2. Open Backend → Variables and copy ADMIN_USER (default admin) and ADMIN_PASSWORD (generated).
  3. Open the Web domain, sign in and change the password under your profile. Public sign-up is off.

Safety defaults

  • Live trading is off. LIVE_TRADING_ENABLED=false blocks live-mode strategies, and AGENT_LIVE_TRADING_ENABLED=false keeps agent tokens paper-only.
  • Billing, payments, marketplace and referrals are off.
  • Manual Quick Trade orders are not covered by these switches. They go to the exchange as soon as you save exchange API keys.
  • Create trade-only API keys without withdrawal rights, restrict them by IP where the exchange allows, and start small.
  • Change the switches in Railway → Backend → Variables. Railway variables override the Settings page at every restart.

AI and data keys (optional)

The stack runs without any key. To use AI analysis, set LLM_PROVIDER and one key on Backend (all workers reference it): OPENROUTER_API_KEY (default provider), OPENAI_API_KEY, DEEPSEEK_API_KEY, GOOGLE_API_KEY, GROK_API_KEY, MINIMAX_API_KEY, or CUSTOM_API_URL + CUSTOM_API_KEY + CUSTOM_MODEL for any OpenAI-compatible endpoint. Optional data keys are TAVILY_API_KEYS, FINNHUB_API_KEY and TWELVE_DATA_API_KEY. Then redeploy. Keys entered only on the Settings page reach the API process but not the workers, so prefer Railway variables. Model usage is billed by your provider.

MCP

The MCP endpoint answers 503 until you connect it:

  1. In QuantDinger, open your profile and issue an Agent Token. Paper-only is recommended.
  2. Paste it into MCP Server → QUANTDINGER_AGENT_TOKEN and redeploy MCP Server.
  3. Point your client at https://{mcp-domain}/mcp with the header Authorization: Bearer {QUANTDINGER_MCP_AUTH_TOKEN}.

Scaling and versions

  • Give memory to the Celery Backtest Worker for parallel backtests (CELERY_BACKTEST_CONCURRENCY).
  • Raise CELERY_CONCURRENCY for more AI jobs.
  • Keep the sum of the DB_POOL_MAX values below Postgres max_connections=200.
  • Images are pinned to v5.8.4 by digest. To upgrade, change QUANTDINGER_BACKEND_IMAGE on Backend and QUANTDINGER_FRONTEND_IMAGE on Web to the same new release, then redeploy all services. Read upstream's release notes first.
  • Email (SMTP) needs Railway's Pro plan for outbound mail.

Cost (estimate): about $50/month idle and $90/month in light use (≈5–7 GB RAM across 15 services). Hobby plan or higher.

Why Deploy QuantDinger on Railway?

Railway runs QuantDinger's full worker topology without managing servers. Private networking keeps Postgres, Redis, Kafka and the API off the internet. Volumes keep strategies, jobs and events across redeploys. Workers and memory scale from the dashboard, so you can start in paper mode and grow the stack only when you need it.

View QuantDinger v5 AI Trading Research and Backtesting on railway.com

Similar templates

TemplateCategoryDeploymentsDeploys
9Router
Self-hosted AI router for Claude, Gemini, Cursor & OpenAI tools.
AI/ML9,734
flowise-railway
Flowise - low-code LLM apps builder
AI/ML8,512
LibreChat
Open-source ChatGPT clone: multi-model, agents, search, multi-user auth
AI/ML6,762
Hermes Agent by Nous Research
Template for Hermes Agent by Nous Research
AI/ML6,413
OpenClaw
[Homebrew support] The AI that actually does things
AI/ML5,167
Hermes Agent
Self Host Hermes Agent. Autonomous AI agent w/ self-improving, tools & more
AI/ML4,092