tick-stock-panel (TSP) is a self-hosted, zero-ops quantitative workbench for China A-shares that unifies stock screening, monitoring and backtesting in a single Docker container, built for individual quant researchers and developers who want their market data and strategies to stay on their own machine.

What it is

TSP is an open-source quant workbench for the China A-share market, written in Python and licensed under MIT. It pairs a FastAPI backend and a React frontend with Polars as the computation engine and DuckDB for local analytics, and it packages the whole stack as a Docker-deployable application. The project describes itself as 个人开源 (personal open source), developed and maintained primarily by an individual, and it is distributed with a plugin-based data source layer so that third-party data feeds can be attached freely. It sits in the AI and machine learning tooling space, with topics covering quant, quantitative-finance, backtesting, screener, llm, ai-agent and self-hosted.

The concrete problem it addresses is the fragmentation of a homemade quant workflow. Anyone building this with scripts typically ends up with a data puller, a screener, a backtester and a monitor that are separate programs with mismatched conventions, and with data access hard-coded to a single vendor. TSP replaces that assembled toolchain with one application that routes six dataset classes independently across sources, so a source can be swapped without rewriting indicator or backtest logic, and it applies one unified enriched data caliber across screening, backtesting, monitoring and review. The README is explicit that it is not investment software and not a market-watching terminal, and that it deliberately does not position itself against 同花顺 or 通达信 and does not ship AI stock recommendations or limit-up prediction.

Key capabilities

• Capability routing matrix that routes six dataset classes independently by source capability, so switching data vendors does not change indicator or backtest conventions; custom sources are documented in docs/custom-data-source.md.
• Screener running 25 built-in strategies plus custom signals and AI-generated signals, scanning the full A-share market at millisecond speed with Polars, with daily and minute strategies held in one pool and auto-routed by their declared cycle.
• Indicator pipeline producing 68 columns of indicators and signals, written to enriched Parquet in a single scan.
• Backtest module with four research views: factor backtest (IC/IR, layered returns, long-short portfolios over a 62+ factor catalog), strategy backtest (equity curve, drawdown, Sharpe, win rate, profit/loss ratio, Monte Carlo drawdown, with T+1, commissions, slippage and stop-loss, SSE streaming progress and factor attribution), minute strategy backtest replaying day by day with minute-close entry, and validation covering parameter sensitivity and rolling out-of-sample runs.
• Factor platform and mining: custom DSL factors built from 25 operators with bilingual fields, IC, layered and Newey-West tests, version and lifecycle management, plus nested out-of-sample factor and strategy search with correlation deduplication where candidates enter the library and publish only after explicit confirmation.
• Monitor center with four rule types (strategy, stock signal, price, abnormal move), AND/OR composition, watchlist group scope, real-time popup alerts, voice broadcast and Feishu push, with trigger records persisted; abnormal move monitoring splits into auction, intraday and offset tabs, the last using exchange deviation thresholds such as 3-day ±20% for the main board, ±30% for ChiNext and STAR, ±40% for the Beijing exchange, and 10-day +100%/−50% and 30-day +200%/−70%.
• Market regime and review tooling: a six-stage sentiment cycle (冰点, 启动, 主升, 高潮, 退潮, 修复) driven by the limit-up ladder with EMA smoothing and two-day confirmation, a Limit Up Ladder page with sealing-order monitoring, and a Beta Review page that generates a post-market AI recap with dragon-tiger list capital flows, scheduled execution, Feishu push and Markdown download.
• Data and extension layer: a Data page profiling local dimension tables, daily K, rights adjustment, enriched, index, ETF, minute K and financial data, with plugin data sources whose extension fields can be promoted into first-level menus and backfilled historically by day.

Who uses it and how

• Individual quant researchers and retail developers running the stack on a local machine or homelab, where a single Docker container keeps data in local Parquet files and requires no managed infrastructure.
• Users running a post-market pipeline: syncing dimension tables, daily K, rights adjustment, enriched data, index, ETF, minute K and financials, then reviewing coverage and sync status on the Data page.
• Intraday monitoring users who keep watchlist groups refreshed by tier and receive popup, voice and Feishu alerts for limit-up, failed limit-up, rebound, limit-down, new high, new low and volume-spike signals aggregated without additional collection.
• Strategy researchers working the research loop of backtesting, exporting CSV results (summary, equity curve, trade detail, per-symbol statistics), saving a candidate and reloading it for a repeat test in one click.
• Users with third-party or proprietary data, who register a plugin source, expose its extra fields as an analysis page and backfill history by day; AI-assisted users who generate strategies, signals and stock analysis and need an LLM endpoint configured in Settings.

Getting started

TSP deploys through Docker: the repository ships a Dockerfile and a Docker CI workflow, and the 快速开始 (Quick Start) section of the README gives the build and run steps. Data sources, the AI interface, real-time monitoring, extension pages and menus are configured from the in-app Settings page, and local data is written to Parquet.

How it compares

The project is MIT-licensed and self-hosted, with all market data landing in local Parquet, which contrasts with paid terminal subscriptions and cloud quant platforms on both cost model and data ownership — the README's own framing notes that paid terminals are expensive and that cloud platforms keep data off the local machine. It is not a drop-in substitute for those products in capability terms, and the README states it explicitly does not benchmark against 同花顺 or 通达信. The relationship is closer to consumption than replacement: tdx appears in the topic list and 同花顺-sourced dimensions feed its concept, industry and pre-market analysis pages.

When to use it — and when not to

Running TSP means operating a Docker container, keeping Parquet storage, scheduling intraday and post-market data syncs, configuring at least one third-party data source, and supplying an LLM API endpoint and a Feishu channel if the AI and push features are wanted; the project ships no market data of its own. It is not suitable for anyone seeking investment advice, AI stock picking, limit-up prediction or a general market-watching terminal, since those are explicitly out of scope. Honest limitations: it is an individually maintained open-source project rather than a vendor product, the Stock Analysis and Review pages are marked Beta, and the full-market auction scan is listed as a pending collection task, so parts of the feature set are still incomplete.

flowchart TB
    subgraph DATA["数据源层 · 插件化"]
        direction LR
        D1["TickFlow SDK"] ~~~ D2["fuyao<br/>同花顺 REST"] ~~~ D3["stock-sdk"] ~~~ D4["YAML 自定义源"] ~~~ D5["+ 更多插件…"]
    end

    subgraph ROUTE["能力路由层"]
        direction LR
        R(["多数据集 · 按能力独立路由 · 档位探测"])
    end

    subgraph STORE["存储层"]
        direction LR
        ST1[("Parquet 分区表")] ~~~ ST2[("DuckDB")] ~~~ ST3[("JSON 按日缓存")]
    end

    subgraph CALC["计算层 · Polars"]
        direction LR
        C1["指标流水线"] ~~~ C2["复权与信号"]
    end

    subgraph RES["研究层"]
        direction LR
        G1["因子引擎"] ~~~ G2["回测引擎"] ~~~ G3["因子挖掘"]
    end

    subgraph SVC["应用层 · FastAPI"]
        direction LR
        S1["REST · SSE"] ~~~ S2["监控引擎"] ~~~ S3["任务调度"] ~~~ S4["交易日探针"]
    end

    subgraph FE["呈现层 · React 18"]
        direction LR
        F1["功能页面"] ~~~ F2["图表可视化"] ~~~ F3["实时推送"]
    end

    subgraph EXT["二次开发与扩展 · 贯穿各层的插槽"]
        direction LR
        X1["自定义策略"] ~~~ X2["自定义信号"] ~~~ X3["扩展分析页面"] ~~~ X4["AI 接口"]
    end

    DATA --- ROUTE
    ROUTE --- STORE
    STORE --- CALC
    CALC --- RES
    RES --- SVC
    SVC --- FE
    FE ~~~ EXT

    classDef fe fill:#eef2ff,stroke:#6366f1,color:#312e81
    classDef svc fill:#ecfeff,stroke:#06b6d4,color:#164e63
    classDef res fill:#fff7ed,stroke:#f97316,color:#7c2d12
    classDef calc fill:#f0f9ff,stroke:#0ea5e9,color:#0c4a6e
    classDef store fill:#ecfdf5,stroke:#10b981,color:#064e3b
    classDef data fill:#fdf2f8,stroke:#ec4899,color:#831843
    classDef pluginSlot fill:#fdf2f8,stroke:#ec4899,color:#831843,stroke-dasharray:5 4
    classDef route fill:#faf5ff,stroke:#8b5cf6,color:#6b21a8,stroke-width:2px
    classDef ext fill:#f8fafc,stroke:#94a3b8,color:#334155,stroke-dasharray:5 4

    class F1,F2,F3 fe
    class S1,S2,S3,S4 svc
    class G1,G2,G3 res
    class C1,C2 calc
    class ST1,ST2,ST3 store
    class D1,D2,D3,D4 data
    class D5 pluginSlot
    class R route
    class X1,X2,X3,X4 ext

    style FE fill:#f5f3ff,stroke:#c7d2fe,color:#3730a3
    style SVC fill:#ecfeff,stroke:#a5f3fc,color:#155e75
    style RES fill:#fff7ed,stroke:#fed7aa,color:#9a3412
    style CALC fill:#f0f9ff,stroke:#bae6fd,color:#075985
    style STORE fill:#ecfdf5,stroke:#a7f3d0,color:#065f46
    style ROUTE fill:#faf5ff,stroke:#ddd6fe,color:#6b21a8
    style DATA fill:#fdf2f8,stroke:#fbcfe8,color:#9d174d
    style EXT fill:#f8fafc,stroke:#94a3b8,color:#334155,stroke-dasharray:7 5

docker run -d --name tsp -p 3018:3018 -v ${PWD}/data:/app/data ghcr.io/shy3130/tick-stock-panel:latest

cp .env.example .env
docker compose up --build
# 打开 http://localhost:3018

CODEX_CLI_VERSION=0.144.3 docker compose up --build

帮我部署开源项目 https://github.com/shy3130/tick-stock-panel 到本机:
克隆到当前文件夹;有 Docker 优先拉 ghcr.io/shy3130/tick-stock-panel:latest 现成镜像,没有就走 Dev 模式;
缺少的依赖(Docker / Python / Node)帮我一起装好;
最后告诉我浏览器打开哪个地址、需要填哪些 Key。

cp .env.example .env       # 按需填 TICKFLOW_API_KEY(留空 = None 模式)
./dev.sh                   # Windows: .\dev.ps1

TICKFLOW_API_KEY=              # 留空 = None 模式(历史日K免费);填 Key 解锁更多
AI_API_KEY=                    # 留空 = 关闭 AI;填 Key 启用策略生成
PORT=3018                      # 服务端口