Userpilot

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 onboarding and survey layer I actually need instead of Userpilot.

The honest split: the rendering is a day, the targeting is a week, and the visual builder is the product. Decide whether a developer adds each flow before you start.

STACK
- Node 20+ with Fastify serving a small script
- SQLite through better-sqlite3, WAL mode
- Vanilla JS on the page, no framework
- Caddy in front

THE DATA MODEL
- flows: id, name, kind, steps_json, audience_json, frequency_json, goal_event, is_live, version
- surveys: id, name, questions_json, audience_json, trigger_json, is_live
- responses: id, survey_id, person_hash, answers_json, at
- exposures: id, flow_id, person_hash, step_index, action, at
- events: id, person_hash, name, props_json, at — the same table the targeting reads

ONE EVENT STREAM FOR BOTH JOBS
- The reason these products bundle analytics with flows is that targeting needs behaviour. Write both to the same table and the audience rule is a query
- Audiences as saved predicates over events and traits: signed up in the last week, has never created a project, is on the paid plan
- **Evaluate the audience on the server, once per session**, and send the qualifying flows in the same response. Evaluating in the browser means shipping your customer segments to the browser

RENDERING
- Tooltip, modal, slideout, checklist, banner, and a survey card. Six shapes cover everything these products sell
- Anchoring by a data attribute your own code adds, never a generated class. If the anchor is missing after a short wait, skip and record it
- Keyboard reachable, escape closes, focus returned, prefers-reduced-motion honoured

SURVEYS DONE PROPERLY
- NPS, one question with a comment, and a short multiple choice. That is the whole vocabulary anybody uses
- **Never ask the same person twice within a stated window**, and never during a task they are in the middle of. A survey that interrupts a checkout is a survey that lowers your revenue
- Store the raw answer and the page it was asked on; NPS without context is a number nobody can act on

PROVING IT WORKED
- Every flow declares a goal event. Compare people exposed with people who qualified and were not, in the following seven days
- **Hold back a random slice from every flow**, permanently. Without a control group, 'onboarding improved activation' is a story rather than a finding, and this is the single feature that makes the whole system worth having

WHAT MATTERS MOST
The shared event stream and the permanent holdback. The first makes targeting easy; the second is the only way you will ever know if any of it works.