How I Prompt

Local-first analytics for your AI coding conversations — prompts, personas, trends, and a personal wrapped view.

Live Dashboard • Live Wrapped • Source

npx @eeshans/howiprompt

Requirements: Node.js 18+. Works with Claude Code, Codex, Copilot Chat, Cursor, LM Studio, Pi, and OpenCode logs. The local package ships with analytics disabled.

Highlights

| | | |---|---| | Local-first pipeline | Sync, parsing, feature extraction, scoring, and metrics run on your machine | | Multi-source support | Claude Code, Codex, Copilot Chat, Cursor, LM Studio, Pi, and OpenCode | | Two ways to explore | Standard dashboard plus a scroll-through wrapped experience | | Metrics that feel personal | Vibe Coder Index, Politeness, activity trends, heatmaps, and personas | | Private by default | Raw logs stay local and the npm package ships with analytics disabled | | Fast repeat refreshes | Incremental rebuilds reuse the local DB, caches, and exclusions |

What it does

How I Prompt syncs local AI conversation logs, computes prompting metrics on-device, and opens a browser dashboard with both an overview page and a scroll-through wrapped experience.

That starts a local server and opens the dashboard. On first run, a setup wizard detects supported backends, lets you confirm sources, and then runs the pipeline.

Options

npx @eeshans/howiprompt --no-open     # don't auto-open browser
npx @eeshans/howiprompt --port 4000   # custom port
npx @eeshans/howiprompt --help        # usage info

What Happens

1. Detects supported local backends and writes setup to ~/.howiprompt/config.json
2. Copies raw conversation data into ~/.howiprompt/raw/
3. Parses and stores messages in a local SQLite database at ~/.howiprompt/data.db
4. Extracts features and scores prompts for dashboard metrics (no model downloads needed)
5. Writes ~/.howiprompt/metrics.json and serves the dashboard at localhost

Subsequent refreshes are incremental and reuse the local database, caches, and configured exclusions.

Self-Host On GitHub Pages

If you want to publish your own stats as a static site:

git clone https://github.com/eeshansrivastava89/howiprompt.git
cd howiprompt && npm install && npm run dev:cli
cd frontend && DEMO_DEPLOY=true npm run build

Then commit and push docs/, and enable GitHub Pages from main / docs in your repo settings.

What You Get

| Dashboard | Full Experience | |-----------|-----------------| | One-page overview of your stats | Scroll-through "Wrapped" presentation | | howiprompt.eeshans.com | howiprompt.eeshans.com/wrapped |

Metrics include: total prompts, conversation depth, activity heatmap, model usage, Vibe Coder Index (with explainability), Politeness (with explainability), persona classification (2×2: Detail Level × Communication Style), and trends.

Data Sources

| Source | Location | |--------|----------| | Claude Code | ~/.claude/projects/*.jsonl | | Codex | ~/.codex/history.jsonl | | Copilot Chat | macOS: ~/Library/Application Support/Code/… · Windows: %APPDATA%\Code\… · Linux: ~/.config/Code/… | | Cursor | macOS: ~/Library/Application Support/Cursor/… · Windows: %APPDATA%\Cursor\… · Linux: ~/.config/Cursor/… | | LM Studio | ~/.lmstudio/conversations | | Pi | ~/.pi/agent/sessions | | OpenCode | macOS/Linux: ~/.local/share/opencode/storage · Windows: %LOCALAPPDATA%\opencode\storage |

All supported sources are auto-synced into ~/.howiprompt/raw/ and reused across refreshes.

Backend Status

• Supported today: Claude Code, Codex, Copilot Chat, Cursor, LM Studio, Pi, OpenCode

Privacy

• Local by default — Sync, parsing, scoring, and metrics run on your machine. No model downloads, no network access required
• Persistent storage — Raw copies, local DB, config, and metrics live under ~/.howiprompt/
• No prompt text leaves your machine — The app does not upload raw logs or prompt content
• No analytics in the local app — The npx @eeshans/howiprompt package ships with analytics disabled. PostHog is only enabled on my demo website
• Fully offline — Unlike previous versions, no ML model download is required. All scoring uses lightweight pattern matching with learned weights
• Ancillary network requests — The dashboard loads ApexCharts from a CDN. The CLI checks npm for version updates. These do not transmit prompt data

The 4 Personas

Your persona is derived from two independent axes: Detail Level (brief → detailed) and Communication Style (directive → collaborative). These form a 2×2 grid validated on 21k prompts.

• The Commander: Brief + Directive — short, decisive instructions
• The Partner: Brief + Collaborative — quick exchanges, conversational flow
• The Architect: Detailed + Directive — specs, constraints, numbered requirements
• The Explorer: Detailed + Collaborative — context-rich, question-driven investigation

Development

# Install deps
npm install

# Build TypeScript
npm run build

# Run tests
npm test

# Rebuild package assets, then launch the CLI locally
npm run dev:cli

# Build frontend
cd frontend && npm run build

# Build the hosted demo from ~/.howiprompt/metrics.json
cd frontend && DEMO_DEPLOY=true npm run build

# Build for distribution
npm run build:cli

# Privacy gate (run before publish)
npm run check:privacy

Requirements

• Node.js 18+

Disclaimer: This is an independent personal project, not affiliated with, endorsed by, or connected to Anthropic or Claude in any way. This is a non-commercial, open-source tool created for personal use and educational purposes only.

License

MIT License — Not affiliated with Anthropic.