daily_stock_analysis

Fetches OHLCV data, real-time quotes, and chip distribution across A-shares, HK, and US markets from a 7-tier provider hierarchy (EFinance → AkShare → Tushare → Pytdx → Baostock → YFinance → Longbridge) with automatic circuit-breaker failover and data validation. Each provider is prioritized by reliability and latency; if one fails or times out, the system transparently falls back to the next tier without interrupting the analysis pipeline.

Routes stock data through a unified LiteLLM interface to multiple LLM backends (Gemini, Claude, DeepSeek, OpenAI, Ollama) with embedded trading philosophy rules and 11 built-in strategies (Bull Trend, Golden Cross, Wave Theory, etc.). Each strategy is implemented as a 'skill' that guides the LLM's reasoning via system prompts and structured output templates, ensuring analysis adheres to quantitative trading principles rather than generating arbitrary commentary.

Integrates with messaging platform bots (Telegram Bot API, Discord Webhooks, WeChat Work Bot API) to enable interactive analysis queries and report delivery. Users can send commands to the bot (e.g., '/analyze AAPL' or '/portfolio') and receive analysis results directly in the chat. The bot supports slash commands, inline buttons for quick actions (buy/sell/hold), and rich message formatting (embeds, cards, rich text). Bots run as separate processes and poll for messages or listen to webhooks.

Enables deployment of the analysis system to GitHub Actions, a free CI/CD platform that runs workflows on a schedule (cron) or on-demand. The system is packaged as a Docker container or Python script that runs in the GitHub Actions environment, fetches stock data, runs analysis, and sends notifications. No server hosting is required; GitHub Actions provides free compute for public repositories (2000 min/month) and paid plans for private repositories. Workflows are defined in YAML and version-controlled alongside the code.

Packages the entire analysis system (backend, frontend, database, notification services) as a Docker Compose stack that can be deployed locally or to cloud platforms (AWS, Google Cloud, DigitalOcean). The Compose file defines services for the FastAPI backend, React frontend, PostgreSQL database, and optional Redis cache. Deployment is as simple as 'docker-compose up', with all dependencies and configuration managed by the Compose file. Supports environment-based configuration (dev, staging, prod) via .env files.

Enables deployment of the analysis system as a systemd service (Linux) or cron job that runs on a local machine or VPS. The system runs continuously as a background service, polling for scheduled analysis times and executing them. Systemd provides service management (start, stop, restart, status) and automatic restart on failure. Cron provides simple time-based scheduling without a persistent service. Both approaches require minimal infrastructure (just a Linux machine) and zero cloud hosting costs.

Aggregates news, risk alerts, earnings data, and capital flow from 4+ specialized search APIs (Anspire, Tavily, Bocha, SerpAPI) and enriches the LLM analysis context with up-to-date fundamental information. The search service queries for stock-specific news, regulatory filings, insider trading, and market sentiment, then embeds results into the LLM prompt as structured context to ground recommendations in real-world events rather than historical price patterns alone.

Implements a multi-agent system that decomposes complex investment questions into sub-tasks, each handled by specialized agents (technical analyst, fundamental analyst, risk manager, sentiment analyzer). Agents communicate via a shared context store and iteratively refine recommendations through multi-turn reasoning. The orchestrator routes user queries to appropriate agents, aggregates their outputs, and synthesizes a final recommendation with consensus scoring and dissent tracking.

Distributes analysis reports to 10+ notification channels (Telegram, Discord, WeChat Work, Feishu, Email, Webhook, etc.) with platform-specific formatting and routing logic. Each channel has a dedicated formatter that adapts the structured analysis output (recommendation, reasoning, metrics) to the platform's native format (Telegram markdown, Discord embeds, WeChat rich text, Feishu cards). Supports group routing rules (e.g., send high-confidence recommendations to all channels, low-confidence only to email) and scheduling (immediate, daily digest, weekly summary).

Evaluates AI recommendation accuracy by comparing predicted buy/sell signals against actual 1-day forward returns. For each stock analyzed, the system records the recommendation (buy/hold/sell) and confidence score, then checks the next trading day's price movement to calculate hit rate, precision, recall, and Sharpe ratio. Results are aggregated per strategy and per LLM provider, enabling performance comparison and model selection. The backtesting engine runs continuously as new analysis is generated, building a historical performance database.

Tracks open positions, entry prices, and current P&L for a user's stock portfolio. The system monitors each position against the AI's current recommendation (buy/hold/sell) and alerts when recommendations change (e.g., a held stock receives a sell signal). Calculates portfolio-level metrics (total P&L, concentration risk, sector exposure) and suggests rebalancing actions based on AI recommendations and risk thresholds. Supports multiple portfolio snapshots (e.g., 'aggressive', 'conservative') with different risk parameters.

Centralizes all system configuration (API keys, LLM model selection, strategy parameters, notification channels, data source priorities) in a registry that reads from environment variables and config files. The registry supports environment-based overrides (dev, staging, prod) and enables dynamic reconfiguration without code changes. Configuration is validated at startup to catch missing API keys or invalid parameters early. Supports config inheritance (base config + environment-specific overrides) and secret management (API keys stored in .env, not in code).

Provides a CLI for running stock analysis on-demand or on a schedule (daily, weekly, monthly). Users can specify stock symbols, strategies, and notification channels via command-line arguments or interactive prompts. The CLI supports batch analysis (analyze 100 stocks in one run) and dry-run mode (show what would be analyzed without actually running). Integrates with system schedulers (cron on Linux, Task Scheduler on Windows) to enable zero-cost scheduled execution via GitHub Actions or local systemd.

Provides a web-based dashboard for viewing stock analysis results, managing portfolios, and configuring analysis parameters. The backend is built with FastAPI and exposes REST APIs for fetching analysis history, portfolio data, and performance metrics. The frontend is a React SPA that displays charts (price, technical indicators, performance trends), recommendation cards, and portfolio summary. Users can trigger ad-hoc analysis, adjust strategy parameters, and view detailed reasoning for each recommendation without touching code.