FeatureJet
A hosted feedback board for indie SaaS teams, with an API an agent can drive.
- Status
- Live
- Live and taking payment at featurejet.com — $15 a month or $119 a year, flat, with a seven-day trial. It was announced publicly on 2026-07-20; nobody has paid yet, and the distribution push is underway.
- Ship date
- Shipped July 18, 2026
- Went live and payment-capable on this date. The Product Hunt launch — the marketing push, not the switch-on — is set for 2026-08-11.
- Page updated
- This page is edited as the product changes, not archived at launch.
FeatureJet is a hosted feature-request board. Your users post ideas, other users vote on them, and when something ships the people who asked for it get told. That is the whole product. It exists because feedback arrives scattered across email, DMs, and support threads, and the founders receiving it end up building for whoever complained most recently instead of for whatever the most people wanted.
Why this one
It is a well-served category — the incumbents are good — but priced for companies with a head of product. A solo founder with four hundred users needs the same signal and cannot justify the same line item.
The part that is actually different
Every board operation is available over a documented public API, with first-party client libraries and a hosted MCP server. The board can be read and written by an agent: triaging inbound requests, updating statuses, notifying voters when something ships. My own roadmap is managed that way, which is the only reason I trust that the API is complete — if it were missing something, I would be the one blocked by it.
Where it actually stands
It is live. The site is up at featurejet.com, the pricing page works, and the signup, trial, and payment flow all run end to end — anyone who wants it can pay for it today. What it does not have yet is customers: it was announced on 20 July 2026 — the switch from built-but-quiet to out in the world, first directory submission included — and nobody has paid yet. The Product Hunt launch, the louder peak, is set for 11 August 2026.
So the numbers stay honest and mostly empty. Revenue is zero because no one has paid, not because it cannot be measured. The board metrics above will read "not measured" until the measurement pipeline is wired to this page, and they will read low for a while after that. Selling the thing and getting it used are the work of the next ninety days, in public, with the real figures either way.
Numbers
Nothing here has a source yet, so nothing here shows a number. These will fill in when the measurement pipeline is wired to this page — not before.
Learnings
Written at the time, dated, and kept even when they stop being flattering.
- What didn't
A hand-written llms.txt went live factually wrong
I wrote the machine-readable summary of the product by hand, from memory, and it described capabilities that did not match what had actually shipped. It was wrong the day it went up. Hand-maintained summaries of a moving product are a category of file that is always drifting toward wrong; the only fix that holds is generating them from the same source the pages render, and testing that they agree.
- What didn't
A post page that rendered four words per line on a phone
A two-column layout on a post page never collapsed at small widths, so the body column got squeezed to roughly four words per line. It looked fine in every desktop check. It had presumably looked like that for as long as the page existed, because nobody had opened it at 390 pixels wide.
- What I'd do differently
Make the internal id mismatch loud instead of silent
FeatureJet's identifier inside the measurement system is not the same string as its name. Anything that looks up its numbers with the obvious name gets a clean, confident, entirely wrong answer: zero. A missing mapping should fail loudly. A metric that cannot resolve its source should refuse to render a number at all, which is the rule this site now runs on.
- What worked
Shipping the SDKs and the MCP server with the API, not after it
The API went out with an official Python client, an official Node client, and a hosted MCP server on the same day. That was not a growth tactic; it was the only way I could dogfood the thing, because the board is maintained by agents rather than by me clicking around in a UI. The side effect is that "an agent can run your feedback board" is a true sentence I can demonstrate rather than a positioning line I made up.