Stop building on vibes.

Capture what users actually experience. Understand it with analytics. Then ask what to build and get an answer with the receipts attached — and hand it all to your agents.

Get early access See the whole loop

Feedback · analytics · beam · reality-check — one corpus, one source of truth

lumen · evidence desk

Drag the receipts into the decision

The premise

Opinions are cheap. Evidence compounds.

Every feature you almost built, every redesign you debated at 2am — your users already weighed in. Lumen turns their screenshots and voice notes into a searchable corpus, so the next “should we?” takes thirty seconds instead of a leap of faith.

~30s

from a user’s voice note to a filed, classified report

100%

of reports transcribed, embedded, and deduped

answers without a citation from your own corpus

The package

More than feedback. The whole build loop.

Capture what users experience, understand it with analytics, decide with Reality-Check, and hand the receipts to your agents. One project, one corpus, one source of truth.

Feedback

Screenshot, voice and text from inside your product. AI transcribes, classifies and files every report.

Analytics

Funnels, retention, journeys, replays, revenue and experiments — the whole product story in one place.

Beam

Phone screenshots straight into your terminal and clipboard. No cloud, no account, no detour.

Reality-Check

Ask “should I build this?” and get an answer with the receipts behind it.

Agent toolkit

CLI, MCP and probes so Claude Code, Cursor and your own scripts read the corpus themselves.

SDKs

React, Expo and a framework-agnostic core — one tag, or one provider, and you’re live.

Capture

Users talk in their own words.

A screenshot with annotations, a 10-second voice note, or plain text — straight from inside your product. URL, device, and console context ride along automatically. Nobody fills out a form, so people actually send things.

Voice transcribed in under a second
Screenshots with tap-to-annotate
Screen recording & file upload on Pro

Send feedback

Screenshot attached · export page

0:07

“The export button just… isn’t there on my phone?”

Auto-file

AI files every report. You read verdicts.

Within ~30 seconds each report is transcribed, embedded into your corpus, classified as bug, feature or UX with a severity — and linked to its lookalikes. Your inbox sorts itself while you ship.

Bug / feature / UX / praise, with severity
Duplicates cluster automatically
No AI calls on the capture path — users never wait

Inbox

4 new

“export button missing on mobile”
voice · 2m ago
Bug · high
“would pay extra for recurring invoices”
text · 1h ago
Feature
“this modal traps me — no way back”
screenshot · 3h ago
UX
“love the new editor btw”
text · 5h ago
Praise

Reality-Check

Ask “should I build this?” Get receipts.

Before you spend a week on a feature, ask the agent. It searches everything your users ever said and answers with a verdict — plus the cited reports behind it. It argues only from your corpus, never from generic startup advice.

Verdicts backed by counted, cited reports
Catches what nobody asked for, too
Thirty seconds instead of a leap of faith

Do people actually want dark mode?

Verdict: hold off

Only 2 of 214 reports mention dark mode — both from the same user. Meanwhile 11 reports hit friction on the export flow this month.

report #182report #147

Analytics

Understand what actually moves.

Feedback is the why. Analytics is the what — funnels, retention, journeys, session replays, revenue, experiments. One corpus, one story, no duct-taped stack of three tools.

Funnels, retention & journeys
Masked session replays & heatmaps
Revenue, impact & experiments

Weekly retention

+18%

64%

D7 return

3.2

sessions / user

funnel steps

Beam

Phone screenshots, straight to your terminal.

Screenshot on your iPhone → share → beam → the image lands on your machine as a file and in your clipboard. No cloud, no account, no WhatsApp detour.

One-tap iCloud shortcut
mDNS discovery — zero config
Also in the CLI as `lumen receive`

beam

$ lumen receive
→ IMG_0213.png saved (2.1 MB)
→ copied to clipboard
# pasted from your phone · no cloud

For you and your agents

Your agents read the receipts too.

Everything Lumen does is pipeable. Let Claude Code, Cursor or your own scripts ask the corpus directly.

CLI

Headless Lumen. List feedback, run a reality-check, search the corpus — everything the dashboard does, scriptable and pipeable into agents.

npx @lumen-stack/cli

$ lumen feedback list -t bug
$ lumen reality-check "Add native iOS share sheet"

MCP server

Plug your corpus into any MCP-aware assistant — Claude Code, Cursor, Claude Desktop — and let it search feedback and run reality-checks itself.

npx @lumen/mcp install --client claude-code

$ lumen_feedback_search "users complaining about onboarding"
$ lumen_reality_check "should we build dark mode?"

Probe

Non-breaking runtime probes. Capture the exact sanitised variable state when something misbehaves, and hand your AI real evidence instead of logs.

import { probe } from "@lumen-stack/probe"

$ probe("checkout.ts:214", { cart, total, userId }, {
$   when: (s) => s.total < 0, level: "error"
$ })

How it works

Three moves. No ceremony.

Drop it in

One script tag — or one React provider, if that’s your thing. The feedback button appears in your product. No backend, no webhooks, no config.

Works with Webflow, Framer, and plain HTML

Users talk

Screenshot with annotations, a 10-second voice note, or plain text. The SDK grabs URL, device, and console context automatically — users never fill out a form.

Voice to transcript in under a second

You ask

“Should I build this?” The Reality-Check agent searches everything your users ever said and answers with a verdict — and the receipts to back it.

Every answer cites its sources

Who it’s for

For people who’d rather ship than survey.

Solo founders

One brain, ten hats. You can’t also be the QA team and the research department after midnight.

“let the corpus argue for you”

Vibe coders

Cursor, Lovable, v0 — you generate features faster than you can validate them. Lumen tells you which ones earned their keep.

“ship now, reality-check at coffee”

No-code builders

Webflow, Framer, Bubble. Paste one tag and you have the same feedback machinery as a funded product team.

“zero code, full signal”

Small product teams

Five people, fifty ideas in flight. Stop settling roadmap debates with the loudest voice in the room.

“PMs read verdicts, not transcripts”

Install

No code? No problem.

One paste and your product collects evidence. Developers get first-class SDKs — but you don’t need one to start.

<script src="https://cdn.lumen.dev/sdk.js" data-key="lk_pub_XXXX"></script>

Paste once before </body>. Works with Webflow, Framer, Squarespace, Shopify — anything that renders HTML.

Pricing

Cheaper than one wrong feature.

No per-seat traps, no surprise overages. The free tier is genuinely usable.

Free

€0forever

For solo builders kicking the tires.

1 project
100 feedbacks / month
Voice transcription + AI filing
Web SDK & no-code embed
Community support

Start free

Pro

€29/ month

For founders who decide with receipts.

Unlimited projects
10k feedbacks / month
Unlimited Reality-Checks
Video & upload capture
EU data residency + PII redaction
Priority support

Start 14-day trial

Enterprise

Customlet’s talk

For organizations with serious requirements.

SSO + SAML
Self-host option
SLA + uptime credits
DPA on request
Dedicated success engineer

Talk to sales

Questions

Fair questions.

Before you commit to a feature, you ask the agent — “should I rebuild export?”, “do people care about dark mode?” — and it searches every piece of feedback your users ever sent, then answers with a verdict and the cited reports behind it. It only argues from your corpus, never from generic startup advice.
No. Paste one script tag into Webflow, Framer, Squarespace, Shopify or any HTML page and the feedback button appears. If you do write code, there’s a React provider (@lumen-stack/react on npm) and a React Native/Expo SDK in beta.
An annotated screenshot, a screen recording, an uploaded clip, a voice note, or plain text — straight from inside your product. URL, device, and console context attach automatically, so nobody fills out a form.
Within about 30 seconds it’s transcribed (if voice), embedded into your corpus, classified (bug / feature / UX with severity), and linked to similar reports. Nothing blocks the user — there are no AI calls on the capture path.
PII redaction is on by default for transcripts and screenshots. EU data residency is included on Pro, a DPA is available on request, and you (or your users) can delete any report at any time.
Surveys catch the 2% who answer; the Slack channel becomes an unread museum. Lumen captures feedback at the moment of friction and keeps it queryable — six months of context that answers questions instead of accumulating guilt.

Early access · letting builders in weekly

Your next feature is already in the feedback.

Drop your email — we’ll ship you a project key, and your product starts collecting receipts the same afternoon.

Only early-access emails. Unsubscribe anytime.