Desktop AI reads candlesticks, diagnoses markets, then stops

What it does

PA Agent is a PyQt6 desktop application for subjective traders who practice Price Action analysis. It ingests structured candlestick data from sources like MT5, TradingView, yfinance, or AkShare, computes features, and pipes them through a two-stage LLM pipeline: first a market diagnosis, then a strategy-routed trade decision. The app renders an interactive, cyber-sci-fi decision tree, supports follow-up chat with streaming reasoning, and logs every prompt, JSON decision, and token count to disk, but it explicitly does not take screenshots or interface with broker APIs.

The interesting bit

Instead of asking an LLM to squint at chart screenshots, the tool feeds it structured OHLCV data and pre-computed indicators, then locks the reasoning into a rigid diagnosis-to-decision flow. The “experience library” retrieves historical cases by market cycle position to ground the model’s context, and the entire conversation history persists locally with encrypted API key storage.

Key highlights

• Multi-source ingestion: MT5 (Windows), TradingView (cross-platform), yfinance (futures/crypto), and AkShare/baostock (A-shares)
• Two-stage LLM pipeline: market diagnosis first, then strategy routing into specific order types—limit, breakout, market, or no order
• Incremental analysis that reuses previous conclusions as new bars close, with optional auto-trigger via keep_analysis
• Cyber-sci-fi interactive decision tree with animated gate-to-path playback
• Full audit trail: prompts, raw responses, diagnosis/decision JSON, token usage, and chat history written to disk
• Local encrypted storage for API keys and a configurable validation layer with JSON checks, semantic validation, and automatic retry

Caveats

• Windows 10/11 is the primary target; macOS 12+ is only supported when using the TradingView data source
• Requires Python 3.11+ and a reachable third-party AI API (e.g., DeepSeek, PackyAPI)
• The README carries a blunt disclaimer that this is for learning and research, not investment advice

Verdict

Worth a look if you are a discretionary trader who wants a local, auditable AI research assistant with structured data and zero execution risk. Skip it if you are hunting for an automated trading bot or a screenshot-based chart GPT.