Cherry searches reviews, forums, Reddit, social posts and news, and keeps a source link for every issue.
People describe the same problem a hundred different ways. Cherry names each theme and merges them into one issue.
How severe an issue is and how many people report it combine into one score, which sets the order of the list.
If you disagree with a call, tell Cherry. It re-ranks using your correction and remembers it for next time.
The same process in eight stages. The last stage feeds back into the first.
Cherry sits between the places customers talk and the places teams work. It pulls feedback in from tools you already use and sends work out to the ones where your teams track it. It doesn't replace any of them.
● Live today: Slack routing. The rest is designed but not built yet. Each source gets a small adapter into the same triage core, and issue status syncs back from the linked ticket, so nobody has to update it by hand.
Scaling notes · v1 → v2 Where Cherry breaks at scale, and what v2 changes → Ten breaks and their fixes, including what happens when two reviewers disagree, cost limits that survive a server restart, staged prompt rollouts, data permissions set by contract, and what it takes for someone else to run it.Most feedback tools focus on collecting feedback. Whether one is useful depends on the other end: who reads the output, and what each reader needs. The diagram shows Cherry's answer: five sources, each with its own bias; three jobs in the middle; five different views on the way out; and the return arrows that turn it into a loop.
An internal tool has product-market fit when teams can't make decisions without it. Cherry's bet is that feedback gets there by being cut for each reader, because what counts as high-signal depends on who's reading. Read the full essay →
Cherry drafts every ticket but never sends one by itself. The model does the repetitive first pass, a person makes the calls that need context the model doesn't have, and Cherry measures whether each correction actually improved the next result.
A correction on its own is just a data point. Measuring its effect is what turns it into learning. Read: why the human stays in the loop →