Chameleon

Promptfree, for everyone, and the only version there is

Copy the prompt for Claude CodeCopy the prompt for CodexCopy the prompt for Cursor

Copy

Build the in-product guidance I actually need instead of Chameleon.

One idea in this product is worth copying above all the others: **the launcher**. A menu the user opens when they want help beats a tour that arrives when they do not, and it is far less work.

STACK
- Node 20+ with Fastify serving a small script
- SQLite through better-sqlite3, WAL mode
- Vanilla JS on the page, styled from CSS custom properties your app already defines
- Caddy in front

THE DATA MODEL
- items: id, kind, title, body_html, anchor, audience_json, frequency_json, is_live, version — kinds are tour, tip, survey, announcement, launcher entry
- exposures: id, item_id, person_hash, action, at
- launcher: an ordered list of items marked as reachable on demand

THE LAUNCHER, BUILT FIRST
- A small button in the corner opening a searchable list: the tours, the help articles, what shipped recently, and a way to leave feedback
- **Pull rather than push.** People open it at the moment they are stuck, which is the only moment guidance works
- It costs a fraction of a targeting engine and is used more

PUSHED GUIDANCE, WITH RESTRAINT
- Tooltip, modal, banner. Anchored by an explicit data attribute; skip silently and log if the anchor is gone
- One item per session, never two at once, dismissal permanent, snooze available
- An audience rule evaluated on the server and returned with the payload

STYLING WITHOUT A THEME EDITOR
- Read your application's own custom properties for colour, radius and font. Nothing to configure and it can never drift out of step with a redesign
- Everything inside a shadow root, so your CSS cannot break it and it cannot break yours

PUBLISHING SAFELY, WHICH IS THE CLAIM YOU ARE COPYING
- Items are versioned and have a live flag. Turning one off is a database write, not a deployment
- **A preview mode keyed by a query parameter**, so a change is seen on the real product before anybody else gets it
- A percentage rollout on each item, so a bad tooltip reaches five percent rather than everybody
- An audit of who changed what and when

MEASURING
- Per-step completion and dismissal rates; the step people leave at is the one to cut
- A goal event per item and a holdback slice, or you are measuring nothing

WHAT MATTERS MOST
The launcher and the percentage rollout. One gives people help when they want it; the other means a mistake in the interface layer is small.