AI clustering

One feature request. Asked 47 ways.

Supoid groups duplicate feedback the moment it lands — so you see the volume, not the noise. Tunable threshold, full transparency on every merge, no per-resolution metering.

$0.86
per 1,000 items

Real cost from our pipeline (text-embedding-3-small + Haiku 4.5 metadata). Featurebase charges $290 for the same volume — 100× more.

0.82
default threshold

Tunable per workspace. Slide it up to merge less aggressively, down to merge more. Everyone else's AI is a black box.

~10s
ingest to clustered

Embedding + metadata + cluster match end-to-end. Inngest fans out so high-volume workspaces don't queue.

Why this matters

Stop reading the same request 47 times

Three customers ask for dark mode. Twelve more next month. AI clustering shows you a single 'Dark mode (47 votes)' card with the volume signal you actually need.

Glassbox AI, not a black box

Every cluster has a 'Why merged?' panel — see the cosine similarity score, the reference embedding, and which items contributed. Disagree? Adjust the threshold or split the cluster.

Sentiment + category, automatic

Each piece of feedback lands tagged with sentiment (frustrated / neutral / excited) and category (feature / bug / integration / billing) before you even open the inbox.

PII-redacted before any AI call

Emails, phone numbers, common credentials regex-masked at the edge. Zero-data-retention agreements with both OpenAI and Anthropic via OpenRouter.

How it works

  1. 01

    Feedback lands

    Customer submits via the public board, embeddable widget, email, or REST API. The raw text + author metadata lands in Postgres.

  2. 02

    Embedding + metadata in parallel

    Inngest fans out two AI calls: text-embedding-3-small for the 1536-dim vector, and Haiku 4.5 for the sentiment/category/PII pass.

  3. 03

    Threshold-based cluster match

    We run an HNSW cosine search against existing cluster centroids. Above threshold (default 0.82) → merge in, update the centroid via running mean. Below → seed a new cluster.

  4. 04

    Summary on demand

    Once a cluster has 5+ members, Sonnet 4.5 drafts a one-line summary. Pin it in your inbox, drop it into a roadmap card, or close the loop with everyone who voted.

Side by side

FeatureSupoidFeaturebase / Canny
Tunable threshold
"Why merged?" reasoning panel
AI cost (per 1,000 items)$0.86$100–$290
Sentiment + category on intake
PII redaction before AI callvaries
Per-resolution metering surprise
Included in Free plan

Ready to hear what your customers actually want?

Six minutes from sign-up to your first clustered, tagged, actionable feedback. Free forever for solo founders.

No credit card · Cancel anytime · GDPR self-service