Methodi Documentation

Projects

A project contains everything for one research question: papers, datasets, notes, notebooks, and timeline. Each project is self-contained and can be shared with team members.

The AI Pipeline

Methodi uses AI at every step:

1. 1. Paper Parsing — AI reads your PDF with scientific precision, preserving equations and tables
2. 2. Research Assistant — Discuss methodology, ask questions, get citations from your papers
3. 3. Notebook Architect — Generate documented, runnable code based on your plan
4. 4. Execution — Run notebooks and track results with AI-generated summaries

Timeline

Like git for research. Every notebook you complete becomes a "commit" showing what data went in, what came out, and what decisions were made. This creates a traceable record of your methodology.

Workspace

Your project's file system. Notes, notebooks, and outputs all live here. You can organize files into folders and access them from anywhere.

What is project.md?

project.md is the master project document that captures:

• Research goal: What methodology you're replicating and why
• Sample definition: Date ranges, filters, and universe selection
• Key variables: How they're calculated and any adaptations made
• Decisions made: Methodology choices and their justifications
• Current status: What's been done and what's next

How It Works

• 1Automatic injection: project.md is automatically loaded into every Research Agent conversation as context
• 2Recent progress: Your last 3 notebook commits are also included, so the AI knows what you've accomplished
• 3AI updates: Ask the Research Agent to update project.md when major decisions are made

Updating project.md

In the Research Agent, simply ask:

The AI will add the decision to the appropriate section and preserve existing content.

What the Research Agent Can Do

• Search papers: Semantically search your uploaded PDFs for methodology, definitions, and citations
• Read your data: Access your workspace files and dataset schemas, including comments and tags you've added
• Web search: Find relevant papers, documentation, and resources online
• Save findings: Write markdown notes to preserve discussion outcomes and methodology decisions
• Control notebooks: Read, edit, and execute cells in your running Marimo session—debug errors and iterate on code without leaving the chat

Uploading Papers

1. 1. Go to the Research tab in your project
2. 2. Click "Upload PDF" in the right sidebar
3. 3. Select one or more PDF files
4. 4. Wait for processing (status: pending → processing → ready)

Example Questions

• "How do they define the MAX variable in Bali et al.?"
• "What columns does my returns dataset have?"
• "Search for recent papers on momentum crash risk"
• "Save a summary of the methodology we discussed to notes"

The AI cites specific sections from your papers, making it easy to verify information.

Example Conversation

Running Notebooks in the Cloud

1. 1. Click Launch in the Execute tab to start Marimo
2. 2. Open your notebook from the /workspace/notebooks/ folder
3. 3. Run cells with Shift+Enter or click the play button
4. 4. Save your notebook (Cmd/Ctrl+S) — files save directly to your workspace
5. 5. Commit to Timeline to track your progress

What is a Commit?

A commit represents a completed notebook and records:

• Inputs: What datasets were used
• Outputs: What datasets were created
• Message: What you did (your description or AI-generated)
• Notebook: Link to the executed notebook

Creating a Commit

1. 1. Go to Timeline in your project
2. 2. Click "New Commit"
3. 3. Select the notebook you completed
4. 4. Confirm or edit the detected inputs/outputs
5. 5. Add a commit message (or let AI generate one)
6. 6. Click "Create Commit"

Why This Matters

• Onboarding: New team members can understand months of work in minutes
• Reproducibility: Trace any result back to its source
• Documentation: AI summarizes every step automatically

Uploading Data

Drag files into the Data Library or click "Upload".

What happens when you upload:

1. 1. File is uploaded to secure cloud storage
2. 2. Converted to Parquet format for efficient processing
3. 3. Columns and types are detected automatically
4. 4. Sample values are extracted for preview

Data Profiling

Click "Generate Insights" on any dataset to get:

• Column statistics (min, max, mean, percentiles)
• Panel structure analysis (entities, time periods, balance)
• Data quality indicators (missing values, outliers)
• Automatic ID detection (PERMNO, CUSIP, GVKEY, etc.)

Organization

• Folders: Organize datasets into folders (e.g., "raw", "processed")
• Tags: Classify datasets with custom tags for filtering
• Comments: Add notes about data sources or transformations

Default Folders

Every project starts with three key folders:

• notes/Markdown files for research notes, methodology discussions, and documentation
• notebooks/Marimo notebooks (.py files) generated by the Research Agent and your executed analyses
• data/Intermediate datasets, outputs, and processed files from your notebooks

You can create additional folders to organize your work however you prefer.

What You Can Do

• Browse files: Navigate folders in the left panel
• Preview: Click a file to see its contents in the right panel
• Upload: Drag files or click the upload button
• Download: Click the download icon on any file
• Create: Make new markdown notes or folders

Key Differences

Reading Python Output

DataFrames (like SAS datasets)

      permno       date       ret
0      10001 2010-01-04  0.023451
1      10001 2010-01-05 -0.012345
2      10001 2010-01-06  0.008901

This is like PROC PRINT output.

Summary Statistics

count    524382.000000
mean          0.031245
std           0.042156
min          -0.089234
50%           0.024567
max           0.234567

This is like PROC MEANS output.

Common Translations

This table helps you read AI-generated code. You can also right-click any code cell and select "Explain" for a detailed breakdown.

Getting Help with Code

Use the Research Agent in the Research tab for AI assistance:

• "Fix cell 3" — The agent reads the error and fixes the code directly
• "Explain this code" — Get a plain-language explanation of what the code does
• "Improve this approach" — Suggest optimizations based on your methodology

The Research Agent understands your project context—papers, datasets, and decisions—so it can give better suggestions.

The Most Important Skill: Shift+Enter

This runs the current cell. Marimo will automatically re-run any cells that depend on it.

The Basic Workflow

1. 1. Read the markdown cell (understand what this step does)
2. 2. Run the code cell (Shift+Enter or click the play button)
3. 3. Check the output (Marimo shows results inline)
4. 4. Continue to the next step—dependent cells update automatically

Getting Help

Use the Research Agent in the Research tab for AI assistance:

• Fix errors: "I got an error in cell 3, can you fix it?"
• Explain code: "What does this pandas operation do?"
• Edit cells: "Change cell 5 to use a log transformation instead"

The Research Agent has full project context—your papers, datasets, and methodology decisions. It can edit cells directly in your notebook.

If a Cell Gets Stuck

1. 1. Click the Stop button in the toolbar (⏹️)
2. 2. If needed, use the Interrupt Kernel option from the menu
3. 3. Edit your code and try again—Marimo will only re-run affected cells

Free Credits

• $10.00 starting balance when you sign up
• $10.00 monthly refresh every 30 days
• No credit card required to start

Project-Based Billing

All usage within a project is billed to the project owner. This enables seamless team collaboration:

• Team members work without needing their own balance
• Centralized billing for easy expense tracking
• Project owners can monitor usage in Settings → Billing

What Counts as Usage?

• AI conversations: Research Agent chats and notebook generation
• Paper parsing: Converting PDFs to searchable content
• Notebook execution: Running JupyterLab in the cloud

Adding Funds

When your balance runs low, you can top up from Settings → Billing:

• Minimum top-up: $10.00
• Presets: $10, $25, $50, $100
• Auto-recharge: Automatically add funds when balance drops below $2.00

Do I need to know Python?

Not required. AI generates the code, so you can focus on your research. That said, Python experience helps you customize and extend what AI creates.

What if the AI-generated code has an error?

Ask the Research Agent to fix it: "I got an error in cell 5, can you fix it?" The agent can read your notebook, see the error, edit the cell, and re-execute it—all from the Research tab. No API key required.

Can I edit the generated code?

Yes! The code is fully editable. Modify it however you like. If you break something, regenerate the notebook.

What's the largest file I can upload?

5GB per file. For very large datasets, use Parquet format—it's much faster to process than CSV.

Can I use R instead of Python?

Not currently. Methodi generates Python code using pandas and statsmodels.

How do I share results with someone outside my team?

Export notebooks as PDF or HTML from Marimo's export options, or share the .py file directly since Marimo notebooks are pure Python.

Marimo Documentation

• Marimo Documentation (Official)
• Marimo GitHub Repository
• Marimo Blog (Tips and tutorials)

Python for Finance (Optional)

If you want to understand the code better:

• QuantEcon: Python Programming (Free online lectures)
• pandas Documentation (Data manipulation library)