cerno.

// cerno · Latin, verb · to sift, to distinguish

AI-first bug intake, dedup, and routing. Plain-English reports come in; Cerno classifies them, semantically dedupes against known defects, fills the fields, and presents them for a one-click confirm. Bug intake is a sifting problem, not a storage problem — the LLM is the router, not the human.

request early access how we work →
cerno · intake
in › "app crashes when I export a big report to PDF"
───
class · bug (0.94)
dupe · matches #1182 "PDF export OOM on large datasets" (0.89)
route · team:exports · severity:high
ready to confirm — no new ticket created
// the thesis

Every product-support team drowns in duplicate, half-described reports. The cost isn't storage — it's the human triage tax paid on every one. Cerno moves that work to the model: sift first, store second, and only put a human in the loop to confirm.

dedup before save

A report is matched against known defects semantically — by meaning, not keywords — before it ever becomes a new ticket. Duplicates collapse instead of pile up.

route before triage

Classification and routing happen at intake. The right team sees it first; nobody plays switchboard.

summarize before notify

People get a clean, deduped summary — not a raw firehose. Signal up, noise down.

// how it works
01

Accept

A report arrives in plain English — from a form, an inbox, a chat, or an API. No schema for the reporter to learn.

02

Classify

Cerno decides what it is — bug, feature request, or question — and extracts the details worth keeping.

03

Dedup

The report is embedded and compared against existing defects by meaning. Likely duplicates surface with a confidence score and the matching ticket.

04

Populate

Severity, area, and routing fields are auto-filled — a draft ticket, not a blank form.

05

Confirm

A human approves, merges into the duplicate, or edits. The model proposes; the team disposes.

// who it's for

Product-support and engineering teams at small-to-mid software companies taking 20+ reports a day — enough volume that duplicates and mis-routing quietly cost real hours every week.

// architecture

RAG-first on a single Postgres. No separate vector store, no orchestration framework tax — a clean seam where the AI lives.

Next.js PostgreSQL + pgvector Claude API Voyage embeddings shadcn/ui AWS

Want Cerno early?

Cerno is in active development and opening to a small group of design partners. If you run support intake at volume and want a say in how it works, reach out.

info@runthyme.com