Loading persona data...
Your prompts never leave your machine. On-device feature extraction and behavioral analysis turn them into a profile that captures how you prompt.
Read local histories from Claude Code, Codex, Copilot Chat, Cursor, and LM Studio.
Keep human-authored prompts. Remove system messages, tool output, and assistant text.
Extract 38 features per prompt — length, structure, tone markers, code patterns — and score with calibrated logistic regression. No model download needed.
Two axes — Detail Level and Communication Style — place you on a 2×2 grid validated on 21k prompts.
Style scores use on-device feature extraction with learned weights. Persona axes use behavioral patterns validated on 21k prompts from open research datasets (WildChat, OpenAssistant).
Two axes form a 2×2 grid. Your quadrant determines which persona fits your prompting pattern.
1 command
Private, local-first analytics in one command. No hosting, no setup, no extra steps.
npx @eeshans/howiprompt 3 steps
Publish your own persona and stats as a shareable static site in three short steps.
git clone https://github.com/eeshansrivastava89/howiprompt.git cd howiprompt && npm install && npm run dev:cli cd howiprompt/frontend && DEMO_DEPLOY=true npm run build Then commit and push `docs/`, and enable GitHub Pages from `main` / `docs`. Setup guide
Let's find your AI coding assistant data. Everything stays on your machine.
Toggle which backends to include in your analysis.
Claude Code projects in these directories will be skipped.
Analyzing your prompting patterns...
Claude Code projects in these directories will be skipped.
This removes the current metrics file and reopens the setup wizard. Cached raw logs and database history stay on disk and will be reused on the next analysis run.