YoungCan-Wang/WyckoffTradingAgent

WyckoffAgent – An AI‑powered trading assistant for Chinese‑, Hong‑Kong‑ and US‑stock markets

What it does

• Acts as a conversational “Wyckoff master”. You type natural‑language questions and the agent replies with market‑level analysis based on the Wyckoff price‑volume methodology.
• Pulls real‑time daily quotes from TickFlow, stores user‑specific data (portfolio, signal feedback, back‑test results, etc.) in Supabase, and runs a chain of tools that:
  1. Identify Wyckoff structures in A‑share, HK‑share and US‑share daily bars.
  2. Combine those structures with AI‑generated research reports, risk‑control checks and funnel‑screening logic.
  3. Push the results to a web UI, a CLI dashboard, an Electron desktop app, or external MCP tools.

Key components

Component	Role
TickFlow	Real‑time market data provider (no Supabase cache).
Supabase	Stores user config, portfolio, back‑test logs, signal feedback, and task results.
LLM back‑ends	Gemini, Claude, OpenAI, DeepSeek V4 – used for natural‑language dialogue and AI‑driven research reports.
React web app	Interactive chat, funnel‑screen, portfolio view, news overlay, etc.
CLI / TUI	wyckoff command launches a terminal UI, a dashboard (wyckoff dashboard), or one‑off tasks (wyckoff run …).
Electron desktop	Bundles the same UI and tools for Windows/macOS, with in‑dialog write‑operations.
MCP (Model‑Context‑Protocol) server	Exposes 19 public tools that can be added from GitHub, file‑system or custom data sources; all actions go through an approval queue.
Cloudflare Workers + Vercel sandbox	Executes heavy research calculations in a secure, short‑lived environment.

Main workflow

1. Data ingestion – /v1/quotes from TickFlow returns a full‑market snapshot; the agent filters locally.
2. Wyckoff funnel – A‑share market (~5 000 symbols) is scanned for concept‑level “main‑line” structures, eight‑channel strength, candidate lanes and buy‑point confirmation.
3. AI research – LLMs generate three‑tier reports (logic‑breakdown, reserve‑camp, launch‑pad) and judge them.
4. Signal feedback loop – After market close, observations are fed back to improve the funnel and allow shadow‑strategy validation.
5. Execution guard – A “daily funnel × next‑day open” rule only permits a trade if the opening price falls inside a model‑generated interval; otherwise the signal is logged as a warning.
6. User interaction – Through chat, you can request portfolio diagnostics, back‑test grids, or run custom /screen, /checkup, /report, /backtest skills. All write actions are queued for manual approval.

Installation & usage

# one‑click install
curl -fsSL https://raw.githubusercontent.com/YoungCan-Wang/WyckoffTradingAgent/main/install.sh | bash
# or via Homebrew / pip
brew tap YoungCan-Wang/wyckoff && brew install wyckoff
uv pip install youngcan-wyckoff-analysis

wyckoff                # start the conversational agent
wyckoff dashboard      # launch the local visual panel

During the first run you select an LLM model, provide its API key, and the agent is ready.

Cost & licensing

• The code is AGPL‑3.0 and fully open‑source; you can self‑host it.
• Running the hosted service incurs cloud costs (data provider, Supabase, LLM usage, Cloudflare workers). A detailed cost model is documented in docs/COST_MODEL.md.
• A optional “knowledge‑planet” membership (CNY 518 / year) subsidises shared infrastructure; it does not constitute investment advice.

Who it’s for

• Traders who follow the Wyckoff method and want AI‑assisted pattern detection and research.
• Developers interested in a full‑stack example that combines LLM chat, real‑time market data, a database backend, and multi‑platform front‑ends.
• Anyone looking for a reference implementation of a regulated, approval‑based trading automation pipeline.

Where to learn more

• Architecture: docs/ARCHITECTURE.md
• Funnel flow: docs/A_SHARE_FUNNEL_FLOW.md
• Signal feedback loop: docs/SIGNAL_FEEDBACK_LOOP.md
• Public demo videos and screenshots are embedded in the README.