AI-first Trading Framework for Cursor

This project is a semi-manual framework for managing a personal Brokerage Account (Interactive Brokers in example).

It is intentionally simple: you still think and learn at every step, while the assistant helps with the heavy lifting.

---

🗂️ Folder Map

• data/ – raw data landing zone. Drop files manually or collect them automatically with the tools.
• knowledge/ – AI-ready distillates (token-optimised) produced by prompt workflows.
• memory_bank/ – hand-picked knowledge you want the agent to remember. See its README for structure.
• tools/ – scripts the chat agent can call to fetch context, analyse an idea or even place an order in TWS.
• prompts/ – reusable prompt templates that standardise recurring processes.
• learning/ – notes produced via the learning prompt for the human side of the workflow.

Sample JSON, CSV and Markdown files are provided purely for learning and experimentation. Replace them with your own data when ready.

🔄 Typical Usage Flows

1. Collect → Distil → Ask
   Gather fresh data (or let the tools do it), convert it into knowledge distillates, learn yourself and feed the agent. Then ask portfolio or idea questions.
2. Decide → Execute → Log
   When you like an idea, ask the agent to execute via tools/TWS, then append the order to the trade journal in memory_bank.
3. Reflect
   Review winning/losing ideas and update the knowledge base accordingly.
4. Code & Extend
   Add new tools, automate the memory bank or create an ideas folder – the framework is yours to evolve.
5. Reflect & Automate
   Design your own reflection loop: review outcomes, update distillates, and gradually automate any repetitive steps that are currently triggered manually in chat.

❓ Why Semi-Manual?

1. Learning by doing – you touch the raw data, prompts and code.
2. Full visibility – every stage of the investment workflow is explicit; nothing happens behind the curtain.

🎛️ Context Switches

The assistant can operate in four high-level contexts:

Context	Purpose
analysis	Explore data, run deep research prompts, interpret indicators.
execution	Call tools (IBRK, Bybit, SEC, indicators) and perform actions that change state.
reflection	Write trade-journal entries, update mindsets, review decisions.
coding	Extend or modify the utilities inside tools/ (new collectors, CLI wrappers, etc.).

Switching context is implicit: the assistant infers the mode from your request. You can force it, e.g. “switch to coding context”.

🧠 Memory Model

Active knowledge is loaded automatically from memory_bank/active_memory/ on every assistant run.

Folder	What to copy (manually or via ChatGPT)
indicators/	Latest JSON dump from tools/indicators
portfolio/	Fresh portfolio CSV exported from IBKR
mindset/	Current mindset / frameworks markdowns
trade_journal/	When you decide to buy/sell, ask ChatGPT to append a record here – this will power the reflection module in v2 (bring popcorn).

Files are NOT moved automatically – copy them yourself or instruct the chat to do so when appropriate.

Sample data included: Several JSON, CSV and Markdown files in the repo are example snapshots left intentionally for learning and demonstration purposes (e.g. indicator dumps, portfolio exports, mindset distillates). Feel free to explore or replace them with your own real data.

---

---

🚀 Quick-Start Cheat-Sheet

# 0.  Activate virtualenv + load secrets
source .venv/bin/activate      # VS Code should auto-activate thanks to .vscode/settings.json
cp env.example .env               # first run only; then edit .env and add BYBIT_API_KEY, FRED_API_KEY, etc.


# 1.  Collect fresh data (Collect)
# Put reference PDFs into data/books/ for later summarisation
python tools/indicators/collect_indicators.py --out data/indicators/          # macro/liquidity
python tools/IBRK/export_portfolio.py --host localhost --port 7497 \          # portfolio snapshot
       --out data/portfolio/positions_$(date +%F_%H-%M)_full.csv
python tools/sec/secctl.py fetch --ticker AMD --section RiskFactors \         # SEC risk factors
       --out data/sec_data/

# 2.  Distil into knowledge (Distil)
#     – ask the chat agent to run the appropriate prompt on new data using prompts

# 3.  Explore / decide (Ask & Decide)
streamlit run tools/dashboard/dashboard.py                                    # visual overview
python tools/backtest/backtestctl.py run config/reit_strategy.yaml            # sandbox back-test

# 4.  Execute & log (Execute → Log)
#     – via chat: place_order / log to memory_bank/trade_journal

For detailed documentation of each utility see tools/README.md.

⚠️ Model choice matters

Use the o3 model in the Cursor chat. At the time of building this framework it scored best in financial reasoning tests.