QuantDinger is a self-hosted AI trading platform for independent traders, Python strategy authors, and small teams. It takes a trading idea through AI-assisted research, Python strategy code, backtesting, paper trading, live execution, and monitoring in one stack, and keeps market data, strategy code, broker credentials, and deployment under the operator's control. It is not a signal service and does not give investment advice.
The backend is a Flask API backed by PostgreSQL, two Redis instances (cache and durable jobs), and a Kafka event stream. One backend image runs as several separate processes: the HTTP API, trading workers that own exchange sessions and order reconciliation, strategy dispatcher and evaluator workers that process closed-bar events, a Kafka audit worker, a scheduler, and Celery workers for finite AI, backtest, experiment, and report jobs. An optional overlay adds Prometheus metrics, Grafana dashboards, and Alertmanager alerts.
| Area | Supported surface |
|---|---|
| Indicators | Python chart overlays, markers, bands, and signals |
| Strategies | Strategy API V2 intents, sizing, risk, backtests, and live runtime |
| Crypto exchanges | Binance, OKX, Bitget, Bybit, Gate, HTX, and adapter extensions |
| Traditional brokers | IBKR and Alpaca |
| AI providers | OpenRouter, OpenAI-compatible APIs, Google, DeepSeek, Grok, MiniMax, and custom endpoints |
| Automation | Human API, Agent Gateway, MCP server, Celery jobs, schedules, and notifications |
An optional AI decision filter sits in front of live entry orders. Before an entry reaches the exchange, QuantDinger sends the order, strategy context, exposure, positions, and budget to TypeSafe Jev, which returns a typed choice with probabilities and confidence, and falls back to the configured LLM when Jev is not set up. Rejected entries never reach the exchange, exits and stop-losses bypass the filter, and a provider failure fails open and is logged so an outage cannot trap an open position.
The Agent Gateway and the included MCP server let clients such as Cursor, Claude Code, and Codex call approved tools without receiving broker credentials or administrator tokens. Agent tokens are hashed, scoped, rate-limited, and audit-logged, and agent trading is paper-only unless both the token and the server allow live orders.
Features
- Strategy development: Python indicators and Strategy API V2 for intents, position sizing, and risk settings
- Backtesting and experiments: server-side backtests and experiment jobs run asynchronously on Celery workers
- Strategy evolution: random, grid, or TPE parameter search with walk-forward validation, a blind holdout, PBO, Deflated Sharpe, Monte Carlo, and cost stress tests
- Signal-only virtual accounts: notification-mode signals become virtual orders, fills, positions, PnL, and an equity curve without touching a broker
- Paper and live execution: crypto exchanges (Binance, OKX, Bitget, Bybit, Gate, HTX) and traditional brokers (IBKR, Alpaca)
- AI market research: multi-provider market analysis through OpenRouter, OpenAI-compatible APIs, Google, DeepSeek, Grok, and MiniMax
- AI decision filter: typed pre-trade checks on live entries through TypeSafe Jev or an LLM, recorded in an auditable decision timeline
- Agent Gateway and MCP: scoped, rate-limited agent tokens, paper-only by default, for clients such as Cursor, Claude Code, and Codex
- Access surfaces: web app, mobile H5 app, human API, Agent Gateway, and MCP server
- Durable runtime: strategy ownership through leases, heartbeats, and fencing tokens, with PostgreSQL state and audit logs
- Credential protection: broker credentials and MFA secrets encrypted at rest, with production containers running without root privileges
- Observability: optional Prometheus metrics, Grafana dashboards, and Alertmanager alerts
