01 what it watches
Every open surface where developers actually talk.
GitHub Issues, Reddit, Hacker News. The forums where a person with a real problem describes it in their own words, publicly, while they still care whether it gets fixed.
02 how it decides
A triage that reads like a person did it.
Every signal gets scored on whether it's genuine, specific, and currently unresolved. The model cites the evidence it used — no black box, no "vibes-based confidence."
Genuine complaint ● 87% · currently unresolved · not spam.
// likely causeVC-2 model is missing the multilingual fallback for inputs under 20 chars. Reproduces on ja-JP and de-DE.
Draft reply · reference docs/voice-cloning#lang.
03 what a response looks like
A draft, not a send.
PainSignal writes a personalised reply grounded in your docs and prior resolutions. It can also generate a short video reply with an AI avatar. Nothing goes out until a human clicks approve.
04 where it fits
DevRel, developer support, growth engineering.
Not a bot. A first pass across signal you couldn't watch manually. Built for teams with a public developer-facing surface — an API, an SDK, a CLI, an integration.
Currently working against a configured GitHub org. Reddit and HN scoped via Exa.