Back

Starglass

Observation runtime for agent systems watching changing external state.

Watching external systems sounds simple until runs restart, sources churn, payloads duplicate, and downstream agents need stable event identity instead of raw snapshots.

What I built

An observation layer for systems that need to monitor external state and turn meaningful change into normalized events.

Built-in source families cover HTTP, feeds, and filesystem inputs.

How it works

Define an observation target, choose or implement a SourceAdapter, let the runtime execute the poll, persist compact checkpoints, dedupe, and hand normalized events downstream.

What it proves

The output is a stable event envelope with enough identity and compact state to resume safely after cron jobs, restarts, or long-running watch loops.