I wanted to keep an eye on a handful of web pages — a supplier’s pricing page, a regulatory portal, a competitor’s changelog — without either checking them by hand or drowning in alerts every time a footer date or a cookie banner changed. Classic change-detection tools fire on every diff; what I actually wanted was a judgement: did anything I care about change?

Lauscher fetches each source on its own schedule, normalises the page to clean text, and only when a technical change is detected does it call an LLM to evaluate that change against a per-source watch brief you write in plain language. The LLM returns a structured verdict — relevant or not, a severity, a summary, and its reasoning — and you only get a notification (email or webhook) when the result clears your threshold. Every evaluation is fully traceable in the dashboard: the diff, the prompt, the model’s reasoning, and the token cost.

Two design choices keep it practical for self-hosting. The LLM is the expensive part, so it never runs unless a cheap hash-and-diff stage anticipates a real change first, and a monthly budget with a hard stop caps the spend. And the whole thing runs as a single Docker container — Go backend with a background scheduler, embedded React frontend, SQLite for storage — with the Anthropic API as the only external dependency.

“Lauscher” is Hessian dialect for “ears”.

Deploy your own instance with Docker