Nekotag
User feedback, labelled
Nekotag gathers what people write about your apps, from app stores such as Myket and Cafe Bazaar, from a support desk, or from any source you set up in the panel. It has an AI label each comment with your team's own tag list, and shows what changed from one month to the next.
Example comments, written for this page. Each label is checked against the tag list before it is kept.
How it works
A job names the products, the sources and the Jalali months. Nekotag does the rest in order, and the Jobs page shows each step as it happens, with what the AI calls cost.
Each source is set up in the panel: its type (an app store, a support desk, or one you add), its platforms and its login. Myket and Cafe Bazaar reviews come from their developer panels, support tickets from the desk's database. Each comment is stored once, with its rating, app version and platform.
Comments go to the AI in batches, with your tag list and prompt. Each answer is checked: a tag pair that isn't on the list, or an invented comment id, never gets saved.
Reports and Explore turn the tags into shares, trends and comparisons, down to the comments behind each number. Everything exports to CSV that opens cleanly in Excel, Persian included.
What it guarantees
Tagging costs money and the numbers end up in decisions, so the pipeline is strict about both.
Taxonomy
Each product and source has its own tag list and prompt, edited in the panel. Every saved prompt version is kept, and each run records the exact one it used.
Validation
The AI's answer is matched exactly against the list. Anything else is dropped with a warning, so a typo in a tag name can't start a new category.
Cost
Every answered batch is saved before the next one goes out. A stopped or crashed job resumes where it left off, and an outage waits instead of throwing rows away.
Fit
Each call stays inside the model's context window, with room for the answers. A prompt too large for it stops before anything is spent.
Language
Right-to-left screens, Jalali months and dates, and every word in both Persian and English, including the server's messages.
Access
Sign-in with two-step codes, sessions that only the server can read, admin-only setup pages, and no button that deletes data without showing what it will remove.
Explore
| Sub tag | Share A | Share B | Change |
|---|---|---|---|
| پخش ویدیو - قطعی | 8.4% | 13.1% | +4.7 |
| دانلود - خطا | 3.9% | 6.2% | +2.3 |
| قیمت اشتراک | 11.0% | 9.6% | −1.4 |
| مصرف اینترنت | 20.3% | 13.8% | −6.5 |
Change is in percentage points of all tags in each period.
| What | Sub tag | Now | Usual |
|---|---|---|---|
| Up | پخش ویدیو - قطعی | 13.1% | 7.9% |
| New | پرداخت - خطا | 4.2% | 0% |
| More negative | قیمت اشتراک | 71% | 48% |
| Down | مصرف اینترنت | 13.8% | 19.6% |
Sentiment rows count the AI's negative answers for that tag; the score comes from comments your team checked.
Alerts and quality
Nekotag runs on your own server with your own database. Set the sources, write the tag list, and the first month is labelled in an afternoon.