Triage

Your tickets are lying: a bug tracker, not a mood ring.

The LLM is a sensor, not a judge. The model turns messy reality into typed claims that carry their evidence. Deterministic code verifies those claims, makes every decision and computes every number. Humans approve anything that acts.

What it measures

Triage takes raw customer feedback and pulls out what each person is actually complaining about, rather than reducing it to a sentiment score.

Complaints about the same underlying problem are grouped into one issue, which then moves through a real state machine that refuses illegal transitions.

Severity is a formula over how often a problem appears and how recently, so the ranking is arithmetic you can audit, not a judgement call.

Published limits

verbatim from EVAL.md, produced by a real eval run On a golden set of 30 synthetic feedback items with 4 planted clusters, clustering is pure (1.0), every item is assigned (coverage 1.0), every legal state transition is accepted and every planted illegal one rejected (1.0), and severity ordering is monotone in frequency and recency (1.0); the set is synthetic, so it does not measure clustering quality on real customer feedback.

This is an API

There is no web application here. Every endpoint below is JSON, and both reads and writes require a bearer token; unauthenticated requests are refused with 401. The interactive schema at /docs is deliberately public so the route map can be read without credentials.

methodpathwhat it does
POST/api/v1/feedback/importbulk import feedback items
POST/api/v1/feedback/{id}/aspects/extractLLM aspect extraction (key-gated)
POST/api/v1/triage/runcluster, register issues, score severity
GET/api/v1/issuesthe issue registry
GET/api/v1/issues/{id}issue detail with members and transitions
POST/api/v1/issues/{id}/transitionadvance state, illegal moves rejected

Links

version 0.3.2 · commit ae74973 · built 2026-08-03T02:39:22Z · environment production
Values are baked from build arguments. unknown means the image was built without them, which is reported rather than replaced with a plausible-looking string.