BrunoThe Git-native API client. Local-first, open source.
LaunchPad
Your launch, on every trajectory.
12 followers
Your launch, on every trajectory.
12 followers
Write one product brief — LaunchPad drafts native posts for Hacker News, Product Hunt, X, Reddit, the app stores and 42+ channels, each in its real voice and character limits. Everything free for your first launch.
Does LaunchPad actually learn from past launches to refine tone over time, or is each draft generated fresh from the brief without any feedback loop?
Great question@glzar686203 , and the honest answer: no feedback loop yet — every draft is generated fresh from your brief plus the channel's instruction block.
Some of that is a real constraint I chose on purpose. The on-device engine means your drafts never leave your machine, and there's no server-side AI at all — so there's nowhere for a "learn from everyone's launches" model to live without breaking the privacy promise. I'd rather ship no learning than quiet telemetry.
But there's a version of this that fits the architecture, and your comment just moved it up the list: per-user learning that stays in your account. You already can edit any section's prompt and it drives every future generation — the missing piece is the tool noticing your edits. If you always strip the hashtags from your X drafts or flatten the adjectives in your HN titles, that's signal sitting right there in the diff between what was generated and what you marked ready. Feeding those observations back into your prompts — visibly, as suggested prompt edits you approve rather than silent drift — keeps the loop inspectable and keeps your data yours.
Solo-built, so the roadmap is literally comments like this one. Thanks for the sharp question.
How does it actually learn each channel's voice - is it trained on real posts from those communities or does it just follow format rules? Curious how authentic it feels vs generic AI-sounding copy.
Hi@seherserme69440 Neither trained nor format rules, honestly — it's closer to editorial judgment written down. There's no fine-tuning on scraped community posts (that'd be expensive, legally murky, and — the real reason — wouldn't help: models already know what HN sounds like, they just default to LinkedIn-speak unless told not to).
Instead, each of the 42 channels carries a hand-written instruction block (this instruction is customisable by the user as well and the AI picks this up for content generation) that's basically an opinion formed by reading what dies on that platform: the HN one pushes a flat, technically-honest register because excitement gets buried there; the Firefox Add-ons one says the audience is allergic to hype and "collects nothing" beats any feature bullet; the Reddit one is mostly warnings.
Format rules exist too (real per-store character limits, enforced in code with retries, red counter when the model blows past them anyway) — but structure without voice is how you get the same generic post reformatted 42 times. As for how authentic it feels: better on channels I personally read (HN, Reddit, the extension stores) than ones I don't, and I'd rather admit that than pretend otherwise — every prompt is user-editable per section, and when someone who knows a community tells me a draft feels off, that channel's opinion gets rewritten the same week.
There's a sample launch you can judge for yourself, no sign-up: launch.tapdot.org/demo
The per-channel voice matching is genuinely impressive, especially nailing the dry, technical tone that HN readers actually respond to. Most "write for 42 platforms" tools just spin up the same generic post 42 times, so this clearly takes some real craft.
Thanks@nereriskin22522, this one means a lot — the HN voice went through more rewrites than any other channel.
The honest answer to how it works: there's no clever prompt-engineering trick. Every one of the 42 channels has a hand-written instruction block that came from reading actual posts on that platform and paying attention to what dies there.
HN's prompt literally says things like "no marketing fluff, no generic launch-speak" and pushes toward the flat, technically-honest register — because the fastest way to get buried on HN is to sound excited. The Reddit one is mostly warnings about what not to do. The AMO (Firefox Add-ons) one notes that the audience is privacy-conscious and "allergic to hype," so mentioning that your extension collects nothing is worth more than any feature bullet.
And none of it is locked in — every section's prompt is editable right in the editor, so if you'd rather your HN post sound a little different, you rewrite the instruction once and it drives every generation after that. Saved to your account, per channel.
Does the "free for your first launch" mean each user gets one launch total, or one launch per product they ship?
Hi @semihkrcayf3f It's one launch total per account — but "launch" might be bigger than you're picturing.
A launch in LaunchPad is a whole product project: your brief plus drafts across all 42 channels, unlimited regenerations, every AI engine, custom channels, the works. It's not "one post" or "one platform" — it's everything for one product, forever free. You can keep coming back to it, re-draft for a v2 announcement, edit prompts, whatever — it stays yours and stays free.
The line is: a second product needs Pro ($9/mo). So if you ship one thing a year, you may honestly never pay me — that's fine, that user costs me nearly nothing since the AI runs on your device or your own key. If you're shipping multiple products, that's exactly who Pro is for.
And to be upfront about the obvious workaround: yes, you could delete your launch and start a fresh one for your next product. You'd lose all the old drafts and ready-states, but nothing stops you. I'd rather have a paywall people can reason about than one that plays games.
the voice matching across platforms is genuinely impressive, each channel actually sounds like itself instead of that generic "announcement" tone most tools spit out. the 42+ coverage for free on the first launch is a bold move, respect for betting on the craft over the upsell.
Thanks@kadirhakc2dln — though I'll be honest, the free-first-launch bet is less brave than it looks. The AI runs on your device (Chrome's built-in Gemini Nano) or your own OpenAI/Claude key, straight from the browser — my servers never touch a prompt.
So a free user costs me almost nothing, and crippling the free tier would've just been theater. The actual bet is simpler: If the full product drafts your launch well once, you'll come back with your next product — and that's the only thing Pro charges for. Betting on craft over upsell sounds noble, but really the architecture made generosity cheap and honesty was the only marketing budget I had.