go-stock

Implements differential update polling that respects market trading hours across A-shares (SH/SZ), Hong Kong (HK), and US stocks, aggregating data from Sina, Tencent, Eastmoney, and Tushare APIs. Uses market-hour awareness to adjust polling frequency during trading vs non-trading periods, reducing unnecessary API calls while maintaining real-time accuracy. Data flows through a GORM+SQLite persistence layer with FreeCache for high-speed in-memory access, enabling sub-second UI updates without repeated database queries.

Aggregates news from 15+ providers (Telegraph/财联社, Reuters, TradingView, etc.) and applies GSE (Generic Segmentation Engine) for Chinese text tokenization with frequency-weighted sentiment scoring. The pipeline extracts entities (stocks, funds, sectors) from news content, segments text into meaningful chunks, and scores sentiment polarity using frequency analysis of positive/negative keywords. Results are stored in SQLite with timestamps, enabling historical sentiment trend analysis and market-wide vs individual-stock sentiment comparison.

Computes dynamic market rankings (gainers, losers, most active by volume) and sector-level analysis (sector returns, sector sentiment, sector fund flows) by aggregating individual stock data from SQLite. Rankings are computed on-demand or cached with configurable TTL (time-to-live) to balance freshness vs performance. Sector analysis groups stocks by industry classification (from data provider APIs) and computes aggregate metrics (weighted returns, average P/E, sector sentiment). Results are displayed in sortable tables with drill-down to individual stocks. Supports custom ranking criteria (e.g., 'highest dividend yield') via configurable sort expressions.

Implements a task scheduler that executes background jobs (price polling, news fetching, sentiment analysis, AI analysis) on configurable schedules with market-hour awareness. Tasks are defined in SQLite with cron expressions or simple interval schedules (e.g., 'every 5 minutes during market hours'). The scheduler respects market trading hours across different exchanges (A-shares, HK, US) and skips execution during non-trading periods. Task execution is asynchronous and non-blocking; results are stored in SQLite with execution logs. Supports task dependencies (e.g., 'run sentiment analysis only after news fetching completes') and error handling with retry logic.

Builds a cross-platform desktop application using Wails v2 framework, which bridges Vue.js frontend with Go backend via IPC (inter-process communication). The application compiles to native executables for Windows (WebView2), macOS (Universal/Intel/ARM builds), and Linux. Wails handles window management, file dialogs, system tray integration, and native notifications. The frontend uses NaiveUI component library for consistent UI across platforms. Application state is persisted to SQLite, enabling data retention across sessions. Supports auto-update mechanism for distributing new versions to users.

Implements a provider abstraction layer that supports 8+ LLM providers (OpenAI, DeepSeek, Ollama, LMStudio, AnythingLLM, 硅基流动, 火山方舟, 阿里云百炼) with unified interface for model selection and API key management. Configuration is stored in SQLite with encrypted API keys (using Go's crypto/aes package). Users can configure multiple providers simultaneously and switch between them via UI without code changes. The abstraction handles provider-specific API differences (request/response format, function-calling syntax, error handling) transparently. Supports local LLM providers (Ollama, LMStudio) for offline analysis without cloud dependencies.

Provides data export/import functionality for backing up and restoring user data (stocks, groups, alerts, settings, analysis history) in JSON or CSV format. Export creates a snapshot of SQLite data at a point in time, enabling disaster recovery and data portability. Import validates data schema before insertion, preventing corruption from malformed files. Supports selective export (e.g., export only specific stock groups) and merge import (append imported data to existing database without overwriting). Export files can be encrypted with user-provided password for secure backup.

Implements an AI agent interface that routes user queries to configurable LLM providers (DeepSeek, OpenAI, Ollama, LMStudio, AnythingLLM, 硅基流动, 火山方舟, 阿里云百炼) with a function-calling registry of 14+ tools for stock analysis, fund monitoring, sentiment analysis, and market rankings. The agent uses chain-of-thought reasoning to decompose user queries into tool calls, executes tools against local data (SQLite) and external APIs, and synthesizes results into natural language responses. All data remains local; only the LLM provider receives query context (configurable via system prompts).

Provides a hierarchical stock grouping system (user-defined portfolios, watchlists, sector groups) backed by GORM ORM with SQLite persistence. Users can organize stocks into groups, track group-level metrics (total value, weighted returns, sector allocation), and apply group-level operations (bulk alerts, group sentiment analysis). The data model supports many-to-many relationships between stocks and groups, enabling stocks to appear in multiple portfolios simultaneously. Group composition changes are persisted to SQLite with timestamps for historical tracking.

Renders candlestick (K-line) charts for stocks with support for multiple timeframes (1m, 5m, 15m, 30m, 1h, 1d, 1w, 1M) using data aggregated from Sina, Tencent, and Eastmoney APIs. The frontend uses NaiveUI charting components to display OHLCV (Open, High, Low, Close, Volume) data with interactive zoom, pan, and crosshair tools. Backend aggregates raw tick data into OHLCV bars for each timeframe, caches results in FreeCache, and persists historical data to SQLite for offline viewing. Supports overlay of technical indicators (moving averages, RSI, MACD, Bollinger Bands) computed server-side.

Monitors mutual funds and ETFs through web scraping of fund data sources, tracking net asset value (NAV), estimated daily value, and fund composition. The system periodically scrapes fund provider websites to extract NAV and estimated value, stores results in SQLite with timestamps, and computes daily/weekly/monthly returns. Supports partial functionality for fund composition viewing (holdings, sector allocation) where data is available. Fund data is aggregated into watchlists and groups similar to stock monitoring, enabling portfolio-level fund analysis.

Implements a rule-based alert engine that monitors stock prices against user-defined thresholds (absolute price levels, percentage changes, volume spikes) and triggers notifications via multiple channels (in-app toast, system notifications, optional push to mobile). Alerts are defined per stock or group, stored in SQLite with enable/disable toggles, and evaluated continuously during market hours. The system supports alert templates (e.g., 'alert if price drops 5%') for quick setup, and logs all triggered alerts with timestamps for historical review. Notification delivery is asynchronous to avoid blocking price polling.

Provides AI-assisted stock selection through customizable prompt templates that guide the LLM to analyze stocks based on user-defined criteria (e.g., 'find undervalued tech stocks with positive sentiment'). Templates are stored in SQLite and can be edited via the UI; when executed, they combine real-time market data (prices, sentiment, financial metrics) with the template prompt and send to the configured LLM provider. The LLM returns ranked stock recommendations with reasoning. Results are cached in SQLite with execution timestamp, enabling comparison of recommendations over time. Supports batch analysis of stock groups or entire market sectors.

Aggregates sentiment scores from news analysis across all stocks and the overall market, providing both market-wide sentiment (bullish/bearish consensus) and per-stock sentiment. The system computes sentiment by news source (Telegraph, Reuters, TradingView, etc.) to show which sources are most bullish/bearish, and tracks sentiment evolution over time. Results are stored in SQLite with timestamps, enabling sentiment trend analysis and correlation with price movements. Market-wide sentiment is computed as a weighted average across all stocks, with weighting options (equal-weight, volume-weight, market-cap-weight).

Analyzes fund flows (inflow/outflow by volume) and financial metrics (P/E ratio, dividend yield, ROE, debt-to-equity, etc.) for stocks, aggregating data from multiple sources (Sina, Tencent, Eastmoney, Tushare). Money flow analysis breaks down volume into institutional vs retail flows and tracks accumulation/distribution patterns. Financial metrics are fetched from provider APIs and cached in SQLite with update timestamps. The system computes derived metrics (e.g., PEG ratio from P/E and growth rate) server-side and visualizes money flow as stacked bar charts (inflow/outflow) in the frontend. Supports historical money flow analysis to identify accumulation/distribution trends.