QuantDinger

QuantDinger The open-source AI infrastructure layer for quant trading Turn trading ideas into Python strategies, backtests, paper trading, and live execution - all in one self-hosted stack. QuantDinger is a product of Open Byte Inc. AI research -> Strategy code -> Backtest -> Paper/Live execution -> Monitoring

English / Chinese

SaaS  ·  API Docs  ·  Video Demo  ·  Website  ·  AWS Marketplace

 

 

 

---

Contents

[Try in 2 minutes](#try-in-2-minutes) · [Why QuantDinger](#why-quantdinger) · [Safety model](#safety-model) · [Technical highlights](#technical-highlights) · [Repositories](#related-repositories) · [AI agents & MCP](#use-it-from-an-ai-agent-cursor--claude-code--codex--mcp) · [Overview](#product-overview) · [Features](#features-at-a-glance) · [Visual tour](#visual-tour) · [Architecture](#architecture) · [Install](#installation--first-time-setup-docker-compose) · [Docs](#documentation) · [FAQ](#faq) · [License](#license-and-commercial-terms)

---

Try in 2 minutes

> Fastest path: one command. No git clone, no npm, no Vue source tree. The installer asks for the admin account, writes secure secrets, pulls GHCR images, and starts Docker Compose.

Prerequisites: Docker with Compose v2 (Docker Desktop on Windows/macOS). Node.js is not required.

curl -fsSL https://raw.githubusercontent.com/brokermr810/QuantDinger/main/install.sh | bash

Windows PowerShell:

irm https://raw.githubusercontent.com/brokermr810/QuantDinger/main/install.ps1 | iex

Installs to ~/quantdinger on Linux/macOS and $HOME\quantdinger on Windows by default. Override with ... | bash -s -- /opt/quantdinger or $env:QUANTDINGER_INSTALL_DIR="C:\QuantDinger" before the PowerShell one-liner.

Then open http://localhost:8888 and sign in with the admin username/password you entered during installation. The same stack also serves the mobile H5 client at http://localhost:8889.

Windows, manual clone, or mirror troubleshooting

Windows (PowerShell) - manual clone path:

git clone https://github.com/brokermr810/QuantDinger.git
Set-Location QuantDinger
Copy-Item backend_api_python\env.example -Destination backend_api_python\.env
$key = & python -c "import secrets; print(secrets.token_hex(32))" 2>$null
if (-not $key) { $key = & py -c "import secrets; print(secrets.token_hex(32))" 2>$null }
(Get-Content backend_api_python\.env) -replace '^SECRET_KEY=.*$', "SECRET_KEY=$key" | Set-Content backend_api_python\.env -Encoding utf8
# Edit backend_api_python\.env before first start:
#   ADMIN_USER=your_admin_user
#   ADMIN_PASSWORD=your_secure_password
docker compose pull
docker compose up -d

Standard clone (macOS / Linux):

git clone https://github.com/brokermr810/QuantDinger.git
cd QuantDinger
cp backend_api_python/env.example backend_api_python/.env
./scripts/generate-secret-key.sh
# Edit backend_api_python/.env before first start:
#   ADMIN_USER=your_admin_user
#   ADMIN_PASSWORD=your_secure_password
docker compose pull
docker compose up -d

Slow docker pull (China / VPN): add IMAGE_PREFIX=docker.m.daocloud.io/library/ to a repository-root .env, or configure Docker Desktop → Proxies.

For step-by-step detail and troubleshooting, see [Installation & first-time setup](#installation--first-time-setup-docker-compose).

---

Why QuantDinger

| Traditional workflow | QuantDinger | |----------------------|-------------| | ChatGPT only generates code | Runs, backtests, and executes strategies in one stack | | TradingView + Jupyter + exchange bots are fragmented | One self-hosted stack from research to execution | | SaaS platforms hold API keys | User-owned deployment — your infra, your keys | | AI agents without scopes or audit | Scoped Agent Gateway, paper-only default, audit logs |

QuantDinger is a self-hosted, local-first quantitative infrastructure layer — not a chatbot with a buy button. It unifies multi-LLM research, Python-native strategy engines, server-side backtesting, and multi-broker live execution (10+ crypto venues, IBKR, Alpaca) in one production-grade stack you fully control.

Safety model

• Agent tokens are paper-only by default — live trading requires explicit server-side unlock.
• Live execution requires explicit permission — token scope + AGENT_LIVE_TRADING_ENABLED on self-hosted stacks.
• Exchange keys stay inside the user's own deployment — not held by QuantDinger SaaS operators on self-hosted installs.
• Every agent call is audit-logged — append-only audit trail for automation and compliance review.
• QuantDinger does not provide investment advice — software for lawful research and execution only; you are responsible for compliance and risk.

API documentation

| Resource | Link | |----------|------| | Human Web API (OpenAPI) | [docs/api/openapi.yaml](docs/api/openapi.yaml) | | ReDoc viewer (serve over HTTP) | [docs/api/index.html](docs/api/index.html) — run python -m http.server from docs/api/ | | Conventions (auth, envelopes) | [docs/API_CONVENTIONS.md](docs/API_CONVENTIONS.md) | | Agent Gateway | [docs/agent/agent-openapi.json](docs/agent/agent-openapi.json) |

---

From zero to running stack — charting, AI research, and strategy workflow in minutes.

AI research -> Strategy code -> Backtest -> Paper/Live execution -> Monitoring

Technical highlights

| | What makes QuantDinger different | |---|----------------------------------| | Full-stack quant OS | Charting, indicator IDE, AI research, backtests, live bots, quick trade, and broker account management — one product, one Postgres state store. | | Agent-native | First-class Agent Gateway (/api/agent/v1) + quantdinger-mcp on PyPI — Cursor, Claude Code, and Codex can read markets, run backtests, and trade (paper by default) with full audit logs. | | Dual strategy runtimes | IndicatorStrategy (vectorized dataframe signals + chart overlays) and ScriptStrategy (event-driven on_bar, explicit orders) — research and production in the same codebase. | | Multi-venue execution | CCXT crypto (Binance, OKX, Bybit, ...), IBKR stocks, Alpaca US equities/ETFs/crypto — unified Broker Accounts page with isolated multi-tenant sessions. | | Production-grade infra | PostgreSQL 16 + Redis 7, connection pooling, background workers (orders, portfolio monitor, reflection), idempotent schema bootstrap, GHCR multi-arch images (amd64/arm64). | | Security by default | Refuses default SECRET_KEY, agent tokens hashed at rest, paper-only trading unless explicitly unlocked server-side, every agent call audit-logged. | | Operator-ready | OAuth, multi-user roles, credits/membership/USDT billing toggles, AWS Marketplace AMI, an 11-language web UI, and multilingual docs — build a commercial quant product on top, not just a hobby bot. |

More install paths (GHCR-only, build notes)

Lightest — two files only (no git clone):

curl -O https://raw.githubusercontent.com/brokermr810/QuantDinger/main/docker-compose.ghcr.yml
curl -o backend.env https://raw.githubusercontent.com/brokermr810/QuantDinger/main/backend_api_python/env.example
# Edit backend.env before first start:
#   ADMIN_USER=your_admin_user
#   ADMIN_PASSWORD=your_secure_password
docker compose -f docker-compose.ghcr.yml pull
docker compose -f docker-compose.ghcr.yml up -d

Do not use docker compose up --build for a normal install — the main compose file only declares image: for the frontend; --build affects the backend only. Rebuild backend after code changes: docker compose up -d --build backend. For Vue source builds, use docker-compose.build.yml (see [Installation](#installation--first-time-setup-docker-compose)).

Related repositories

This repo ships the backend, Docker Compose stack, and documentation. The web UI and mobile H5 images are published independently to GHCR by sibling repos. Use the sibling repos when you need source-level UI changes or native mobile builds:

| Repository | What it is | |------------|------------| | QuantDinger (this repo) | Backend (Flask/Python), Compose stack, docs | | QuantDinger-Vue | Web frontend source (Vue) — tagging v* publishes ghcr.io/brokermr810/quantdinger-frontend automatically | | QuantDinger-Mobile | Mobile + H5 client — Compose serves the H5 image at http://localhost:8889; the repo also builds native shells |

Note: Node.js is only required if you build the web or mobile UI from source; the default Docker quick start pulls the published images and does not need it. For source builds, use Node 22 LTS as the shared local version: the mobile H5 repo requires Node 20.19+ or 22.12+ (Vite 7), and the desktop web repo also works on Node 22.

Use it from an AI agent (Cursor / Claude Code / Codex / MCP)

QuantDinger ships an Agent Gateway at /api/agent/v1 plus a small MCP server (quantdinger-mcp on PyPI) that wraps it as Model Context Protocol tools. Issue a token once and your AI client can read markets, manage strategies, run backtests, and (paper-only by default) place trades — without ever seeing your exchange keys or your admin JWT.

> Every agent call is audit-logged, and trading-class tokens are paper-only by default. Live execution requires both paper_only=false on the token AND AGENT_LIVE_TRADING_ENABLED=true on the server.

Two backends, same client config — only QUANTDINGER_BASE_URL differs:

• Hosted (30 s try-out) — sign up at ai.quantdinger.com → Profile → My Agent Token → Issue Token. T (Trading) scope is available; paper-only by default. Live execution still requires paper_only=false on the token, explicit risk acknowledgment at issuance, and AGENT_LIVE_TRADING_ENABLED=true on the server. On multi-tenant SaaS, opening T scope increases shared infrastructure load and platform operational risk — see the in-app risk disclosure.
• Self-hosted (this repo) — after the [Try in 2 minutes](#try-in-2-minutes) Docker bring-up, open Profile → My Agent Token (or the admin-only /agent-tokens page for cross-tenant audit). You control scopes, allowlists, rate limits, and the live-trading flag.

Then point Cursor / Claude Code / Codex at the MCP server (.cursor/mcp.json template: [docs/agent/cursor-mcp.example.json](docs/agent/cursor-mcp.example.json)):

{ "mcpServers": { "quantdinger": {
  "command": "uvx", "args": ["quantdinger-mcp"],
  "env": { "QUANTDINGER_BASE_URL": "http://localhost:8888",
           "QUANTDINGER_AGENT_TOKEN": "qd_agent_xxxxxxxx" }
} } }

Full setup recipe — local stdio config, remote HTTP transport, Claude Code CLI helper, example agent prompts, audit-log walkthrough: [docs/agent/MCP_SETUP.md](docs/agent/MCP_SETUP.md).

Deeper references: [AI Integration design](docs/agent/AIINTEGRATIONDESIGN.md) · [Quickstart with curl](docs/agent/AGENTQUICKSTART.md) · [OpenAPI 3.0 spec](docs/agent/agent-openapi.json) · [MCP server README](mcpserver/README.md)

Product overview

Audience: independent quants, Python strategy authors, prop/small teams, and operators building white-label quant products on private infrastructure — without handing API keys to a black-box SaaS.

Visual Tour

Watch Product Demo on YouTube

Click the preview card above to open the full video walkthrough.

Indicator IDE, charting, backtest, and quick trade AI asset analysis and opportunity radar

Trading bot workspace and automation templates Strategy live operations, performance, and monitoring

Features at a glance

• Research & AI — Multi-LLM ensemble analysis, watchlists, opportunity radar, NL→Indicator/strategy, post-backtest AI hints; optional confidence calibration. [Agent Gateway + MCP](#use-it-from-an-ai-agent-cursor--claude-code--codex--mcp) for Cursor / Claude Code / Codex with scoped tokens and SSE job streaming.
• Build — Professional KLine chart UI; IndicatorStrategy (dataframe buy/sell signals) and ScriptStrategy (on_bar, ctx.buy() / ctx.sell()); AI code generation as a starting point, Python as source of truth.
• Validate — Server-side backtests with equity curves, drawdown metrics, trade logs, and strategy snapshots — no client-side-only backtest theater.
• Operate — Live strategy bots, quick trade, 10+ crypto exchanges via CCXT, IBKR / Alpaca (US stocks, ETFs, crypto); unified Broker Accounts page; notifications (Telegram, email, SMS, Discord, webhooks).
• Platform — Docker Compose + GHCR images, PostgreSQL 16, Redis 7, OAuth, multi-user RBAC, credits / membership / USDT billing toggles, AWS Marketplace AMI, an 11-language web UI, and multilingual documentation.

Architecture

Design principle: separate market data ingestion, strategy/backtest compute, and order execution so research never shares a code path with live capital unless you explicitly promote a strategy.

Stack: Nginx serves the prebuilt Vue SPA (ghcr.io/brokermr810/quantdinger-frontend); Flask + Gunicorn API hosts strategy, AI, billing, and agent services; PostgreSQL 16 is the system of record; Redis 7 backs cache and worker coordination. Exchanges, brokers, LLMs, and payment rails plug in through env-driven adapters — swap providers without forking core code.

Runtime flow: market feeds → indicator/signal layer → strategy engine → backtest or live runtime → venue-specific execution adapters; pending orders dispatched by background workers with health checks and retry semantics.

Deploy surfaces: one-line install.sh, zero-repo GHCR Compose, full-repo Compose (local backend build), AWS Marketplace AMI, and SaaS at ai.quantdinger.com for trials.

System diagram

flowchart LR
    U[Trader / Operator / Researcher]

    subgraph FE[Frontend Layer]
        WEB[Vue Web App]
        NG[Nginx Delivery]
    end

    subgraph BE[Application Layer]
        API[Flask API Gateway]
        AI[AI Analysis Services]
        STRAT[Strategy and Backtest Engine]
        EXEC[Execution and Quick Trade]
        BILL[Billing and Membership]
    end

    subgraph DATA[State Layer]
        PG[(PostgreSQL 16)]
        REDIS[(Redis 7)]
        FILES[Logs and Runtime Data]
    end

    subgraph EXT[External Integrations]
        LLM[LLM Providers]
        EXCH[Crypto Exchanges]
        BROKER[IBKR / Alpaca]
        MARKET[Market Data / News]
        PAY[TronGrid / USDT Payment]
        NOTIFY[Telegram / Email / SMS / Webhook]
    end

    U --> WEB
    WEB --> NG --> API
    API --> AI
    API --> STRAT
    API --> EXEC
    API --> BILL

    AI --> PG
    STRAT --> PG
    EXEC --> PG
    BILL --> PG
    API --> REDIS
    API --> FILES

    AI --> LLM
    AI --> MARKET
    EXEC --> EXCH
    EXEC --> BROKER
    BILL --> PAY
    API --> NOTIFY

Installation & first-time setup (Docker Compose)

> Already ran [Try in 2 minutes](#try-in-2-minutes)? Skip this section — it's the same outcome, just expanded into a step-by-step checklist for first-time deployers and operations folks who want to understand every knob.

This section mi

…

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: brokermr810
• Source: brokermr810/QuantDinger
• License: Apache-2.0
• Homepage: https://www.quantdinger.com

Install and usage instructions live in the source repository linked above.