# QuantDinger reference

Canonical website: https://quantdinger.com/
English documentation: https://quantdinger.com/doc/index-en.html
Chinese documentation: https://quantdinger.com/doc/index.html
Source: https://github.com/OpenByteInc/QuantDinger

## Product definition

QuantDinger is an open-source, self-hosted AI quant trading platform. It combines market research, AI-assisted analysis, Python strategy authoring, chart indicators, historical backtesting, signal notifications, deployment management, and controlled live execution. The website is prepared for the upcoming platform release 5.5.1. The current public `quantdinger-mcp` package is 0.6.2.

QuantDinger is designed as a production-grade operating system. Its production path includes durable workers, auditable order intents, catalog-validated instruments, explicit risk controls, observability, recovery, and operator-controlled live execution.

The platform is designed for individual researchers, engineering teams, internal research desks, strategy-service operators, finance education, and organizations that need control of infrastructure, credentials, data, and execution policy.

## Company

QuantDinger is developed by OpenByte Inc., a technology company focused on open-source innovation and intelligent software. OpenByte builds reliable, transparent, and self-hostable digital infrastructure, turning artificial intelligence, data engineering, and automation into practical products for developers, teams, and businesses. The company believes software should remain open and extensible while giving users control over their data, systems, and choices.

OpenByte welcomes builders across engineering, product, design, quantitative research, and open-source communities. Careers and resumes: support@quantdinger.com.

## Core workflow

1. Deploy QuantDinger and configure administrator, AI model, data-provider, and notification settings.
2. Use AI Research to investigate a market or instrument and verify source freshness and market identity.
3. Implement a strategy with Strategy API V2 and validate its compiled manifest.
4. Backtest with an explicit range, warmup, capital, fees, slippage, benchmark, and execution assumptions.
5. Review drawdown, execution ledger, missing data, and unmodeled risks rather than return alone.
6. Create a stopped deployment, observe it in signal mode, and verify notifications and state restoration.
7. Complete account, instrument, position-ownership, risk-limit, and incident-response checks before enabling live execution.

## Distributed runtime architecture

The current runtime transports versioned closed-bar and strategy-shard events over Kafka. A bar dispatcher assigns stable strategy shards, evaluator consumer groups process those shards, and a Kafka audit role validates contracts and records delivery evidence. PostgreSQL event inboxes, checkpoints, commands, renewable leases, heartbeats and fencing tokens provide durable ownership and replay safety around distributed work.

Exchange side effects remain behind the trading-worker ownership boundary. Pending orders use fenced claims, while persistent grid Actor mailboxes serialize strategy-local state and commands. Cache Redis remains evictable and separate from the durable Celery job Redis configured with AOF and no-eviction behavior.

The supported turnkey deployment currently runs the complete topology on one Docker Compose host. API, trading, dispatcher, evaluator, audit, scheduler and job roles are isolated so operators can scale them independently and evolve toward multi-host or multi-cluster deployments without changing the Strategy API V2 contract. This is an architectural scaling boundary, not a claim of a particular benchmark capacity.

Architecture overview: https://quantdinger.com/doc/architecture/README.html
架构总览: https://quantdinger.com/doc/architecture/README_CN.html
Distributed runtime scaling: https://quantdinger.com/doc/deployment/DISTRIBUTED_RUNTIME_SCALING.html
分布式运行时扩展: https://quantdinger.com/doc/deployment/DISTRIBUTED_RUNTIME_SCALING_CN.html

## Strategy API V2

Strategy API V2 is the canonical Python strategy contract. A strategy declares its universe and subscriptions during initialization and reacts to data or scheduled events through supported callbacks. Validation produces a manifest that drives backtest and deployment behavior. Strategy source runs in a restricted environment and must use canonical market identifiers.

Documentation: https://quantdinger.com/doc/trading/STRATEGY_DEV_GUIDE.html

## AI agents and MCP

QuantDinger exposes a separate Agent Gateway with scoped Agent Tokens. R is read, W changes configuration, B runs backtests, N manages notifications, T controls trading and runtime actions, and C is administrative. Scopes do not bypass the platform live-trading switch, market allowlists, notional limits, expirations, rate limits, or audit logging. Side-effecting calls should use a unique `Idempotency-Key`.

The public `quantdinger-mcp` 0.6.2 package exposes 58 tenant-scoped tools through local stdio and supported network transports. It can be connected to Codex, Cursor, Claude Code, and other MCP clients. Hosted and self-hosted endpoints use the same MCP wiring; the base URL and operator policy differ.

MCP setup: https://quantdinger.com/doc/agent/MCP_SETUP.html
Agent contract: https://quantdinger.com/doc/agent/agent-openapi.json

## Market and execution boundaries

Crypto spot and swap workflows cover Binance, Bitget, Bybit, OKX, Gate, and HTX according to venue and account capability. Gate provides catalog-validated direct-equity routing. OKX, Bybit, and Bitget Reality may expose catalog-validated tokenized-equity products. Binance, OKX, Bitget, Bybit, and Gate may expose catalog-validated equity perpetuals. Binance bStocks use spot semantics without leverage. Native USStock broker workflows support Alpaca and IBKR under the current long-only policy.

Parsed symbols or historical data do not imply live support. Availability depends on current venue catalogs, region, permissions, settlement currency, API family, and platform policy. Mixed-market live deployments are not supported. Exact native product identity must be retained.

Live safety guide: https://quantdinger.com/doc/trading/LIVE_TRADING_SAFETY.html

## Safety and limitations

AI output may be incomplete or wrong. Backtests are historical simulations and do not guarantee future performance. Strategy API V2 backtests currently do not model Crypto funding payments. Live execution must begin with minimum permissions, withdrawal disabled, strict notional and exposure limits, and a tested stop path. A strategy can manage only its allocated position and must not absorb manual or other-strategy inventory.

QuantDinger does not provide investment advice. Operators remain responsible for credentials, infrastructure, market access, legal requirements, execution decisions, and risk.

Security documentation: https://quantdinger.com/doc/security/README.html