M1 Pie DCA Allocator

Structured portfolio modeling for M1 Finance investors. This tool helps users allocate funds across nested "pie-of-pies" structures, visualize adjustments, and persist data using browser cookies. Powered by GPT-4o Vision to extract portfolio snapshots from screenshots.

---

🔧 Features

• 🧠 Vision OCR: Parse M1 screenshots using GPT-4o
• 🧮 Smart DCA: Adjust allocations using a weighted strategy
• 🧱 Recursive structure: Supports pies within pies, and tickers
• 🧁 Classification: Differentiates between tickers and pie folders
• 📸 Image cache and re-parse control
• 🍪 Client-side persistence: Uses browser cookies for data storage
• 📊 Side-by-side charts and review tables
• 📉 Weight-based portfolio normalization
• ⚙️ Toggleable logging level from UI

---

🖼️ Upload & Parse Screenshots

Upload .png, .jpg, or .jpeg screenshots from M1 Finance. The app extracts portfolio composition using OpenAI's GPT-4o Vision model. Each image is hashed and cached to avoid redundant parsing.

{
  "FRB23Q1": { "type": "pie", "value": 1833.72 },
  "RB21Q4": { "type": "pie", "value": 874.34 },
  "FB25-4": { "type": "pie", "value": 301.24 }
}

Re-parsing is supported with a checkbox toggle. Parsed slices replace the current portfolio state.

---

🗂️ Example Portfolio (Expanded)

{
  "name": "main",
  "type": "pie",
  "value": 3009.3,
  "children": {
    "FRB23Q1": {
      "type": "pie",
      "value": 1833.72,
      "children": {
        "TSLA": { "type": "ticker", "value": 950.0 },
        "NVDA": { "type": "ticker", "value": 883.72 }
      }
    },
    "RB21Q4": {
      "type": "pie",
      "value": 874.34,
      "children": {
        "AAPL": { "type": "ticker", "value": 874.34 }
      }
    },
    "FB25-4": {
      "type": "pie",
      "value": 301.24,
      "children": {
        "APP": { "type": "ticker", "value": 94.06 },
        "MELI": { "type": "ticker", "value": 72.16 },
        "PAYC": { "type": "ticker", "value": 70.71 },
        "BAM": { "type": "ticker", "value": 68.27 }
      }
    }
  }
}

---

💸 Dollar-Cost Averaging (DCA)

Input:

• New capital amount
• % allocation to new vs existing tickers
• Target number of new tickers

Output:

• Adjusted portfolio with new positions and updated weights
• Pie charts comparing original vs adjusted allocation
• Table of capital distribution, formatted with currency

---

💾 Persistence & Cookies

All portfolio data is compressed and stored in real browser cookies via extra-streamlit-components. No server-side storage is used.

• Auto-saves on image parse or adjustment
• Reloads last session portfolio automatically
• Warns if data exceeds cookie storage limits (~4KB)

---

🔑 API Usage

Create a ~/.streamlit/secrets.toml file to store your OpenAI API key:

[openai]
api_key = "sk-..."

---

📦 Requirements

Python 3.11+, with minimal environment specified in requirements.txt.

Install via pip:

pip install -r requirements.txt

---

📍 Roadmap

• Replace portfolio on upload instead of merging (in progress)
• Responsive layout optimization
• CSV export of DCA transactions
• Optional integration with cloud persistence (e.g. Storj, S3)

---

🧠 Author & License

Maintained by @stonematt
Licensed under the MIT License