gbrain-operations

Delete the stale lock row from the gbrain_cycle_locks table in the PGLite database, or remove the leftover postmaster.pid file under ~/.gbrain/pglite. Then rerun gbrain doctor to confirm the database is healthy before syncing again.

Create a sync script that runs gbrain sync and gbrain embed --stale, then register it with hermes cron every 15 minutes using no-agent mode. Schedule the dream cycle nightly with a cron entry that runs gbrain dream on the default profile.

Kernel upgrades can change WASM memory mapping so file-backed PGLite databases crash while in-memory ones still work. The recommended fix is migrating to native PostgreSQL in WSL, since gbrain re-syncs from the brain markdown files.

PGLite is zero-config but brittle under kernel upgrades and blocks concurrent worker processes needed by dream cycle subagents. Local PostgreSQL offers crash recovery, faster queries, and working subagent workers, making it the recommended choice for production or WSL setups.

The embedding provider quota is exhausted or rate-limited, often when an expired OpenAI key is used instead of the Backup Provider key. Reduce batch size with --batch-size, switch providers with --provider, or verify quota via the provider's API.

Run the import as a background process with notify_on_complete enabled and a timeout of 3600 seconds, since importing thousands of files can take over two hours. Foreground execution hits the 300-second cap and kills the import mid-run.