Berth is a local-first desktop app for people who use AI coding agents seriously enough that their local setup has become hard to see. It reads the files, indexes and session traces that shape those agents, then shows them without changing them.
What Berth reads
The current scanner focuses on local agent assets and the evidence needed to explain where each one came from.
Instructions and memories
Skills, commands and plugins
MCP servers, hooks and permissions
Sessions, usage metadata and scan sources
How the scanner is organized
The scan engine lives in @berth/scan-engine instead of inside the UI. That package owns the shared asset model, adapter registry, CLI and adapter API.
Standalone package boundary
Worker-backed desktop runtime
SQLite stale-while-revalidate cache
Adapter v1 contracts for new agents
What Berth does not do
Berth is an inspector, not an editor. It avoids write operations, account systems and remote telemetry so local agent configuration can be reviewed safely.