QuantDinger: A Self-Hosted AI Trading OS for Python Strategy Authors

What QuantDinger Is and Who Uses It

QuantDinger is an open-source AI trading OS that covers the full strategy lifecycle: AI market research, Python strategy development with Strategy API V2, server-side backtesting, paper trading, live execution, and monitoring. The README describes it as targeting independent traders, Python strategy authors, and small teams who want a local-first, self-hosted design that keeps market data, strategy code, broker credentials, and deployment under the operator's control.

The project is a product of Open Byte Inc., supported by Atlas Cloud and Amazon Web Services. A hosted live app runs at ai.quantdinger.com, but the repository is designed to be run independently. The platform explicitly is not a black-box signal service: the operator controls strategy code, risk settings, credentials, and deployment.

The README includes a risk warning: QuantDinger can submit real orders when live trading is enabled, and it recommends starting with paper trading. The project does not provide investment advice.

The v5 Architecture: Separate Processes for Each Concern

Version 5 reorganized the backend around explicit runtime and operational boundaries. The HTTP API no longer owns long-running trading or scheduler loops. Trading, scheduling, Celery jobs, and database migrations each run as separate processes.

The architecture as shown in the README's Mermaid diagram has the following container layout: web, mobile, and API clients connect through an Nginx frontend to a Flask and Gunicorn API. The API writes to PostgreSQL, a Redis cache instance, and a Redis jobs instance. Three workers run separately: the trading worker owns strategy runtimes, pending orders, broker sessions, and reconciliation; the scheduler worker handles portfolio, deployment, and monitoring schedules; the Celery worker handles finite, retryable async jobs. A Prometheus, Grafana, and Alertmanager observability stack is available as an optional overlay.

Cache Redis and jobs Redis use separate instances with separate eviction policies, preventing job queue entries from being evicted under memory pressure in the same way cache entries can be. The backend image is reused across the API, trading worker, scheduler worker, and Celery worker containers, with different startup commands determining the role.

Deploying QuantDinger with Docker Compose

The docker-compose.yml comments give the deployment sequence:

cp backend_api_python/env.example backend_api_python/.env
SECRET_KEY=$(python3 -c "import secrets; print(secrets.token_hex(32))")
docker-compose up -d --build

The web interface opens at localhost:8888 and the mobile H5 interface at localhost:8889. The compose file will not start the container if `SECRET_KEY` is using the default placeholder value, enforcing the secret key requirement at startup rather than leaving insecure defaults in place.

The .env.example at the project root lists optional port overrides and production credentials. For production, the following should be set before deploying:

POSTGRES_PASSWORD=
REDIS_PASSWORD=
CELERY_REDIS_PASSWORD=
GRAFANA_ADMIN_USER=admin
GRAFANA_ADMIN_PASSWORD=

Separate Redis memory limits are configured via `REDIS_CACHE_MAXMEMORY` (defaults to 128mb) and `REDIS_JOBS_MAXMEMORY` (defaults to 512mb). The production compose overlay runs backend processes as a non-root user with a read-only root filesystem, dropped capabilities, and resource limits.

Strategy Evolution and Backtesting Workflow

QuantDinger's backtesting system is called strategy evolution. It searches tunable parameters using random, grid, or TPE optimization with bar-count walk-forward validation and a final blind holdout. Jobs run asynchronously with per-strategy history, automatic pruning, composite scoring, Probability of Backtest Overfitting (PBO), Deflated Sharpe, block-bootstrap Monte Carlo, and transaction-cost stress tests. The README explicitly states that results compare parameter robustness and do not forecast future returns.

Signal-only virtual accounts are a separate mechanism. They turn notification-mode signals into internal virtual orders, fills, positions, trade records, profit and loss calculations, and an equity curve, without connecting to a broker or submitting live orders. Each fill uses a fixed 0.05% commission on executed notional and 0.05% adverse slippage. The commission calculation does not charge leverage a second time.

This separation between parameter optimization and paper-trading execution lets strategy authors test whether a strategy is overfitting before committing it to even paper capital.

AI Market Research and MCP Integration

The platform includes multi-provider AI market research as one of its core components. The README describes a data flow from AI research to strategy code to execution, and the docs/agent/README.md file covers AI agents and MCP integration. The repository contains an mcp_server/ directory, indicating a Model Context Protocol server implementation that connects the platform to AI tools and agents.

The platform supports web, mobile H5, a human API, an Agent Gateway, and MCP access. This range of access methods reflects the platform's intent to serve both human traders through a browser or mobile interface and automated agents that call the API programmatically.

The README also mentions a public roadmap on GitHub Projects and a list of contributor-ready tasks, suggesting the AI agent integration features are an active development area.

Where QuantDinger Is Not the Right Tool

QuantDinger requires PostgreSQL and Redis as persistent services. Teams running on a single small server where those services do not already exist will need to provision and maintain them alongside the trading platform. The compose file sets a 32-connection PostgreSQL pool for the trading worker by default, which may be more than a small deployment needs.

The platform is built around Python strategies. Traders who work in a different language or who want to run compiled strategy code cannot use the Strategy API V2 without a Python wrapper layer. The README does not document supported languages beyond Python.

The live trading capability requires explicitly enabling it, and the README warns that regulatory requirements vary by jurisdiction. Teams outside the jurisdictions where crypto and stock trading via API is permitted must verify local requirements before enabling live execution.

For users who only need financial data analysis and research without execution, the platform's full stack adds complexity that may not be justified.

QuantDinger vs OpenBB

OpenBB is a well-known open-source financial data and analysis platform. It focuses on data access, normalization, and research workflows, providing a terminal and SDK for querying financial data across providers. It does not ship a live trading execution engine or a backtesting framework as integrated components.

QuantDinger combines data, strategy development, backtesting, and execution in a single platform. The trade-off is scope: QuantDinger's full stack is more complex to deploy and maintain, but provides a complete workflow from research to live trading in one system. OpenBB is the better tool for analysts who need financial data and research capabilities without managing a trading execution environment.

The related searches for QuantDinger include VeighNa and TradingAgents, which are other open-source frameworks in the same space, suggesting the market for self-hosted AI trading platforms is active and the options are worth comparing before committing to any one stack.

Editorial conclusion

Independent traders and small teams who want a complete strategy lifecycle platform, from AI research to live execution, on infrastructure they control should evaluate QuantDinger. The README warns clearly that the platform can submit real orders and recommends starting with paper trading, using restricted API keys, and reviewing local regulatory requirements. The secret key must be set before the container will start, which is enforced in the compose file as a security measure. The latest release is v5.4.5 as of 2026-09-26, and the last push was on 2026-09-25, indicating the project is actively changing. Teams that only need market data or financial analysis without execution should consider whether a dedicated research tool better fits their use case.

Frequently asked questions

Official sources

1. Official documentation
2. Official README
3. Project repository
4. Release notes