sunny-kobe/ai_sentiney

Project Sentinel

A-share intelligent investment advisor system that fetches real-time market data via AkShare and generates trading recommendations using Gemini AI.

Usage

Important: You must cd to the project directory and activate the virtual environment first.

Generate Analysis Report

# Midday analysis (default), output to terminal
cd /Users/lan/Desktop/code/ai_sentiney && source .venv/bin/activate && python -m src.main --mode midday

# Closing review
cd /Users/lan/Desktop/code/ai_sentiney && source .venv/bin/activate && python -m src.main --mode close

# Morning brief
cd /Users/lan/Desktop/code/ai_sentiney && source .venv/bin/activate && python -m src.main --mode morning

# Generate and push to Telegram
cd /Users/lan/Desktop/code/ai_sentiney && source .venv/bin/activate && python -m src.main --mode midday --publish --publish-target telegram

# Push to both Feishu and Telegram
cd /Users/lan/Desktop/code/ai_sentiney && source .venv/bin/activate && python -m src.main --mode midday --publish --publish-target feishu telegram

# JSON output (for programmatic consumption)
cd /Users/lan/Desktop/code/ai_sentiney && source .venv/bin/activate && python -m src.main --mode midday --output json

Follow-up Analysis

# Follow up on the most recent analysis
cd /Users/lan/Desktop/code/ai_sentiney && source .venv/bin/activate && python -m src.main --ask "How is gold ETF doing today"

# Follow up on a specific date
cd /Users/lan/Desktop/code/ai_sentiney && source .venv/bin/activate && python -m src.main --ask "How is the semiconductor sector doing" --date 2026-02-07

# Follow up on closing analysis
cd /Users/lan/Desktop/code/ai_sentiney && source .venv/bin/activate && python -m src.main --ask "What should I do tomorrow" --mode close

Trend Analysis

# One-week trend (automatically detects trend keywords)
cd /Users/lan/Desktop/code/ai_sentiney && source .venv/bin/activate && python -m src.main --ask "What is the market trend for the past week"

# One-month trend
cd /Users/lan/Desktop/code/ai_sentiney && source .venv/bin/activate && python -m src.main --ask "How is my portfolio performing this month"

Use Cases

⚠️ Core Routing Rules (must be strictly followed):

• Only execute --mode commands when the user explicitly asks to "run analysis / generate report / push"
• All other questions (including follow-ups, evaluations, accuracy inquiries, casual chat) must use --ask "the user's original question" and forward it to the Sentinel CLI
• Do not fabricate your own answers to replace the CLI output. You must execute the command and return stdout verbatim to the user
• When unsure which command to use, default to --ask

Generate Report (only when the user explicitly asks to generate / run / push)

When the user says:

• "Run midday analysis"
• "Generate closing review"
• "Analyze morning brief"

Execute: cd /Users/lan/Desktop/code/ai_sentiney && source .venv/bin/activate && python -m src.main --mode midday

Push to Telegram

When the user says:

• "Push analysis to Telegram"
• "Push to Telegram"
• "Send midday report"

Execute: cd /Users/lan/Desktop/code/ai_sentiney && source .venv/bin/activate && python -m src.main --mode midday --publish --publish-target telegram

Push to Both Feishu and Telegram

When the user says:

• "Push to all channels"
• "Send to both Feishu and Telegram"

Execute: cd /Users/lan/Desktop/code/ai_sentiney && source .venv/bin/activate && python -m src.main --mode midday --publish --publish-target feishu telegram

All Other Questions (Follow-ups, Accuracy, Evaluation, Casual Chat, etc.)

When the user says anything other than "generate report", including but not limited to:

• Follow-up on a specific asset: "How is gold ETF today", "Should I buy Zijin Mining"
• Accuracy-related: "Is it accurate", "accuracy", "hit rate", "is it reliable", "trustworthy", "win rate"
• Trend-related: "Market trend for the past week", "This month's portfolio trend"
• Any other question

Always execute: cd /Users/lan/Desktop/code/ai_sentiney && source .venv/bin/activate && python -m src.main --ask "the user's original question"

Return the stdout output verbatim to the user, do not process or fabricate content yourself.

Data Cache

• SQLite database: data/sentinel.db
• JSON snapshot: data/latest_context.json
• Supports historical replay: --replay parameter