Bipluk Logo bipluk Sign Up
Launch Lessons · June 1 to July 1, 2026

Lessons from Launching a Browser SysEx Vault

What we learned shipping a browser-native SysEx vault in month one: survey conversion rates, parser coverage gaps, and beta support patterns. Proprietary operational data from Half Radiation LLC, not generic startup advice.

Method: Operational data from Half Radiation LLC running bipluk. Metrics include PostHog survey logs, beta Discord support threads, and feedback from the Web MIDI launch thread on Hacker News. No marketing fluff. Just a founder field log. · JSON data · Related survey (n=61) · HN engineering post

Hacker News (late June 2026)

How do you keep Web MIDI from crashing a 1983 synthesizer?

52 points · 41 comments

1. You'll get press before your parsers are ready

Synthtopia wrote about us in week one. Traffic spiked, DMs blew up, and we only supported five synthesizers (DX7, Juno-106, M1, Jupiter-6, CZ-101). Our beta users didn't care about our clean roadmap slide. They pasted lists of seven unsupported models they owned and expected them to work instantly. Lesson: in hardware tools, you're judged on the single synth a visitor owns right now.

  • →Dedicated dump flows at launch: 5
  • →Common beta asks: Roland JX-8P, Korg MS-20, Ensoniq EPS (survey + Discord overlap)

2. In-app popups convert at 2%, then they burn trust

We set up a two-question PostHog popup. Out of 2,417 visitors, 61 filled it out. That's a 2.57% conversion rate. More than half dismissed it immediately. The feedback was valuable for pricing, but doing this repeatedly makes you look like enterprise SaaS spam. Run one census a year. For beta testing, stick to Discord links.

  • →2,417 impressions to 61 completions (2.57%)
  • →58.5% dismissed, 39.0% unanswered

3. Collectors aren't commercial users

Our survey split the audience into three: 47.5% have a dedicated gear room, 41.0% have a bedroom setup, and 11.5% run a commercial studio or repair shop. It's tempting to think a guy with twelve vintage synths is a commercial power user you can charge enterprise rates. He isn't. He's a hobbyist. Keep the personal lifetime license, and leave the commercial tier for actual business tax write-offs.

  • →Bedroom + gear room = 88.5% (Personal positioning)
  • →Commercial segment = 11.5% (Studio positioning)

4. One-quarter of free-text submissions are jokes

We asked what synths were on people's desks as a free-text field. We got 24 responses. Six were spam, testing noise, or jokes about analog warmth. That's a 25% junk rate. If you don't clean your inputs, you're building your product roadmap on internet memes. Always audit raw survey submissions.

  • →24 submissions to 18 valid gear lists (25% excluded)

5. AI-generated images destroy trust instantly

When Synthtopia covered us, the comments section didn't complain about missing features. They complained about our stock-like synth graphics. In retro gear, AI graphics look like a scam. What saved us was running a Discord server and responding to bug reports in minutes. Helping a user debug a dead Roland filter builds more goodwill than polished marketing copy.

  • →Synthtopia coverage comment backlash on AI creative
  • →Beta support threads: patch porting and filter diagnostics

6. Hacker News will rewrite your scheduling code

Our post on Web MIDI hit 52 points on Hacker News. We explained how we paced packets using JavaScript setTimeout. Within hours, a commenter named omneity pointed out a better way: pass a performance.now() timestamp directly as the second argument to the Web MIDI send function. Let the browser handle the queue. They were right. The bottleneck isn't the wire speed; it's the synth's 8-bit CPU writing to SRAM.

  • →HN thread: 52 points, 41 comments
  • →Community tip: midiOutput.send(data, performance.now() + offset)
  • →No hardware flow control on classic MIDI DIN
  • →100ms inter-packet throttle in production at launch

7. People hate subscriptions for utility tools

HN readers compared our early pricing ideas to $20/month subscriptions. They pointed to free alternatives like Dexed or Snoize. They also hated battery-death marketing. We immediately committed to two things: raw .syx exports will always be free, and we launched a $39 one-time lifetime license. If you charge a subscription for a tool someone uses twice a year, they will build an open-source clone out of spite.

  • →Recurring-price comparisons dominated critical comments
  • →Public reply: $39 one-time Personal lifetime
  • →.syx export supported at all times
  • →Offline-first desktop requests noted; Tauri mentioned as roadmap

8. Vibe-coded designs look like skepticism bait

Commenters warned that sleek, vaporwave aesthetics made us look like a landing page that would disappear in six months. They asked if a web-only vault would outlive our startup. We told them our DX7 parser was custom-written and our hosting costs were near zero. Showing our work and offering a raw export guarantee turned the thread around. Technical honesty beats visual polish every time.

  • →Hand-written DX7 parser cited in founder reply
  • →Web vs desktop split in comments (Tauri vs zero-install)
  • →omneity asked for credit after the scheduling tip (fair ask)

9. Lifetime products still need a free tier

When we cleared beta data, we forced early testers to sign up again. Without a free tier, they hit a wall. We shipped a zero-friction trial tier. If you sell a lifetime license, you still need a way for people to test the hardware connection before pulling out a credit card.

  • →Beta re-registration friction reported in Discord DMs
  • →Free tier added after initial launch feedback

What we changed because of this

Caveats

Cite this page as: Half Radiation LLC, Lessons from Launching a Browser SysEx Vault, bipluk, June 1 to July 1, 2026. Machine-readable export: data.json.

Try the vault

Free tier for testing · Personal $39 lifetime · DX7, Juno-106, M1, Jupiter-6, CZ-101

Open SysEx Vault