Skip to content

Organize Customer Feedback into Topics You Define

One note, one primary topic. The default topics are bug, feature, usability, pricing, praise, and other. Rewrite the descriptions if your product uses different words.

Live sends the text to this site's server, then to the official Laya Hugging Face Space. No account is required for one message.

Rough budget: about 4 of 320 tokens for the text, and about 140 of 180 for the questions. The English checkpoint context is 512 tokens, with about 192 reserved for the question. This estimate is characters divided by 4, not the real tokenizer.

Categories

The key is the label that comes back. The description tells the model what belongs there.

No result yet

Load an example or paste a message, then choose Run classification. Loading an example does not send it.

Saving rules to an account is not available. There is no signup. Copy the rules if you want to keep them.

Why these topics are not ticket departments

A support ticket asks which team acts next. Feedback asks what the note is about. A bug is a broken behavior. A feature is something the product does not do. Usability means it works but is hard to use. Pricing is cost, plan, or value. Praise asks for no change. Other is for notes that are too thin, or that you refuse to force into a theme.

The instructions tell the model to pick one primary label even when the note mixes topics. That is a product limit, not a hidden multi-tagger. The mixed example below is the request we recorded:

{
  "state": {
    "message": "I love the new dashboard, but the filter control is buried and the Pro plan feels expensive for a solo user."
  },
  "questions": {
    "topic": {
      "type": "choice",
      "instructions": "Choose the main topic of this customer feedback. Pick one primary label even if more than one topic appears.",
      "criteria": {
        "bug": "Something is broken, crashes, or behaves incorrectly",
        "feature": "A request for a capability the product does not have",
        "usability": "The product works, but it is confusing or hard to use",
        "pricing": "Comments about cost, plans, discounts, or value",
        "praise": "Positive feedback that does not ask for a change",
        "other": "Does not fit the categories above, or there is not enough information"
      }
    }
  }
}

How to read a mixed note

If product needs every theme, do not use a single choice as the system of record. Split the sentence, or ask one yes/no question per theme in the playground. If you only need a queue, write which theme wins: for example “if the note mentions a broken action, choose bug even if it also praises the design.”

Recorded calls

One call each on 2026-09-29 to the official Space convaiinnovations/laya-demo, playground endpoint, state {"message": "..."} and the question shown on that tool. The model field came back as laya. latency_ms is the Space response field, not this website's end-to-end time and not the upstream GPU figure. A later call can differ. These six calls are not an accuracy score.

A broken export

Every time I export a CSV the download fails with error 500. I need the file for a meeting today.

Human reference. Bug. The export fails with an error.

This call returned. bug. Model field laya. Space-reported latency 131.8 ms. Input tokens 136. Confidence 0.9532. Date 2026-09-29. Space convaiinnovations/laya-demo.

  • bug: 0.9871
  • usability: 0.0069
  • feature: 0.0033
  • other: 0.0019
  • pricing: 0.0004
  • praise: 0.0004

A high probability on one clear message is still a prediction. This site does not treat 0.95 as safe to automate.

Praise, usability, and price in one note

I love the new dashboard, but the filter control is buried and the Pro plan feels expensive for a solo user.

Human reference. Three topics: praise, usability, and pricing. A single primary label drops two of them. A reviewer who needs the actionable part might pick usability or pricing.

This call returned. praise. Model field laya. Space-reported latency 141.6 ms. Input tokens 137. Confidence 0.4727. Date 2026-09-29. Space convaiinnovations/laya-demo.

  • praise: 0.7422
  • usability: 0.1075
  • feature: 0.0581
  • other: 0.04
  • bug: 0.03
  • pricing: 0.0222

This call emphasized praise and gave pricing 0.0222. The template is one label, not multi-label tagging. Split the note or add more questions if you need every topic.

An explicit 'not a feature request'

This is not a feature request. Please do not add more buttons. I am only saying the current search already works well.

Human reference. Praise or other. The writer rejects a feature request and says search already works.

This call returned. feature. Model field laya. Space-reported latency 141.9 ms. Input tokens 139. Confidence 0.2811. Date 2026-09-29. Space convaiinnovations/laya-demo.

  • feature: 0.5154
  • usability: 0.1612
  • bug: 0.158
  • praise: 0.1484
  • other: 0.0138
  • pricing: 0.0033

This call missed. The label was feature at 0.5154 even though the sentence says it is not a feature request. Confidence 0.2811 is not a reason to trust it. Do not automate a workflow on negations from this template.

Limits

The negation example returned feature. Use that as a reason to review “not a request” wording, not as a story that the template is finished. Ticket routing uses different categories on the support ticket classifier.

Questions

Can one note receive every topic it mentions?
Not with this template. It asks for one primary topic. A recorded mixed note about praise, a buried filter, and price came back as praise, with pricing near 0.02. Split the note or ask separate questions if you need every topic.
Will it catch “this is not a feature request”?
Not reliably. A recorded call on 29 Sep 2026 labeled that sentence as feature. Treat negations as a review case, not as a solved task.